release: 0.2.5

feat: scratchpad
additional improvements changelog fix: wording
2026-02-22 08:07:40 +01:00 · 2025-10-24 22:13:08 +02:00 · 2025-10-22 22:29:41 +02:00 · 2025-10-20 13:08:00 +02:00 · 2025-10-19 09:18:17 +02:00 · 2025-10-16 20:37:51 +02:00
184 changed files with 13034 additions and 2826 deletions
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -0,0 +1,56 @@
+# Changelog
+## Unreleased
+
+## Version 0.2.5 (2025-10-24)
+- Added manual knowledge entity creation flows using a modal, with the option for suggested relationships
+- Scratchpad feature, with the feature to convert scratchpads to content.
+- Added knowledge entity search results to the global search
+- Backend fixes for improved performance when ingesting and retrieval
+
+## Version 0.2.4 (2025-10-15)
+- Improved retrieval performance. Ingestion and chat now utilizes full text search, vector comparison and graph traversal.
+- Ingestion task archive
+
+## Version 0.2.3 (2025-10-12)
+- Fix changing vector dimensions on a fresh database (#3)
+
+## Version 0.2.2 (2025-10-07)
+- Support for ingestion of PDF files
+- Improved ingestion speed
+- Fix deletion of items work as expected
+- Fix enabling GPT-5 use via OpenAI API
+
+## Version 0.2.1 (2025-09-24)
+- Fixed API JSON responses so iOS Shortcuts integrations keep working.
+
+## Version 0.2.0 (2025-09-23)
+- Revamped the UI with a neobrutalist theme, better dark mode, and a D3-based knowledge graph.
+- Added pagination for entities and content plus new observability metrics on the dashboard.
+- Enabled audio ingestion and merged the new storage backend.
+- Improved performance, request filtering, and journalctl/systemd compatibility.
+
+## Version 0.1.4 (2025-07-01)
+- Added image ingestion with configurable system settings and updated Docker Compose docs.
+- Hardened admin flows by fixing concurrent API/database calls and normalizing task statuses.
+
+## Version 0.1.3 (2025-06-08)
+- Added support for AI providers beyond OpenAI.
+- Made the HTTP port configurable for deployments.
+- Smoothed graph mapper failures, long content tiles, and refreshed project documentation.
+
+## Version 0.1.2 (2025-05-26)
+- Introduced full-text search across indexed knowledge.
+- Polished the UI with consistent titles, icon fallbacks, and improved markdown scrolling.
+- Fixed search result links and SurrealDB vector formatting glitches.
+
+## Version 0.1.1 (2025-05-13)
+- Added streaming feedback to ingestion tasks for clearer progress updates.
+- Made the data storage path configurable.
+- Improved release tooling with Chromium-enabled Nix flakes, Docker builds, and migration/template fixes.
+
+## Version 0.1.0 (2025-05-06)
+- Initial release with a SurrealDB-backed ingestion pipeline, job queue, vector search, and knowledge graph storage.
+- Delivered a chat experience featuring streaming responses, conversation history, markdown rendering, and customizable system prompts.
+- Introduced an admin console with analytics, registration and timezone controls, and job monitoring.
+- Shipped a Tailwind/daisyUI web UI with responsive layouts, modals, content viewers, and editing flows.
+- Provided readability-based content ingestion, API/HTML ingress routes, and Docker/Docker Compose tooling.
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -36,6 +36,15 @@ version = "2.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "512761e0bb2578dd7380c6baaa0f4ce03e84f95e960231d1dec8bf4d7d6e2627"

+[[package]]
+name = "adobe-cmap-parser"
+version = "0.4.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ae8abfa9a4688de8fc9f42b3f013b6fffec18ed8a554f5f113577e0b9b3212a3"
+dependencies = [
+ "pom",
+]
+
 [[package]]
 name = "aead"
 version = "0.5.2"
@@ -116,6 +125,21 @@ dependencies = [
 "memchr",
 ]

+[[package]]
+name = "alloc-no-stdlib"
+version = "2.0.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "cc7bb162ec39d46ab1ca8c77bf72e890535becd1751bb45f64c597edb4c8c6b3"
+
+[[package]]
+name = "alloc-stdlib"
+version = "0.2.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "94fb8275041c72129eb51b7d0322c29b8387a0386127718b096429201a5d6ece"
+dependencies = [
+ "alloc-no-stdlib",
+]
+
 [[package]]
 name = "allocator-api2"
 version = "0.2.21"
@@ -299,12 +323,16 @@ dependencies = [
 ]

 [[package]]
-name = "async-convert"
-version = "1.0.0"
+name = "async-compression"
+version = "0.4.30"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "6d416feee97712e43152cd42874de162b8f9b77295b1c85e5d92725cc8310bae"
+checksum = "977eb15ea9efd848bb8a4a1a2500347ed7f0bf794edf0dc3ddcf439f43d36b23"
 dependencies = [
- "async-trait",
+ "compression-codecs",
+ "compression-core",
+ "futures-core",
+ "pin-project-lite",
+ "tokio",
 ]

 [[package]]
@@ -394,30 +422,41 @@ dependencies = [

 [[package]]
 name = "async-openai"
-version = "0.24.1"
+version = "0.29.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "c6db3286b4f52b6556ac5208fb575d035eca61a2bf40d7e75d1db2733ffc599f"
+checksum = "d4fc47ec9e669d562e0755f59e1976d157546910e403f3c2da856d0a4d3cdc07"
 dependencies = [
- "async-convert",
+ "async-openai-macros",
 "backoff",
 "base64 0.22.1",
 "bytes",
 "derive_builder",
 "eventsource-stream",
 "futures",
- "rand 0.8.5",
+ "rand 0.9.1",
 "reqwest",
 "reqwest-eventsource",
 "secrecy",
 "serde",
 "serde_json",
- "thiserror 1.0.69",
+ "thiserror 2.0.12",
 "tokio",
 "tokio-stream",
 "tokio-util",
 "tracing",
 ]

+[[package]]
+name = "async-openai-macros"
+version = "0.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0289cba6d5143bfe8251d57b4a8cac036adf158525a76533a7082ba65ec76398"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.101",
+]
+
 [[package]]
 name = "async-recursion"
 version = "1.1.1"
@@ -743,15 +782,6 @@ version = "1.7.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "89e25b6adfb930f02d1981565a6e5d9c547ac15a96606256d3b59040e5cd4ca3"

-[[package]]
-name = "basic-toml"
-version = "0.1.10"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ba62675e8242a4c4e806d12f11d136e626e6c8361d6b829310732241652a178a"
-dependencies = [
- "serde",
-]
-
 [[package]]
 name = "bcrypt"
 version = "0.15.1"
@@ -862,6 +892,15 @@ dependencies = [
 "generic-array",
 ]

+[[package]]
+name = "block-padding"
+version = "0.3.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a8894febbff9f758034a5b8e12d87918f56dfc64a8e1fe757d65e29041538d93"
+dependencies = [
+ "generic-array",
+]
+
 [[package]]
 name = "blowfish"
 version = "0.9.1"
@@ -895,6 +934,27 @@ dependencies = [
 "syn 2.0.101",
 ]

+[[package]]
+name = "brotli"
+version = "8.0.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4bd8b9603c7aa97359dbd97ecf258968c95f3adddd6db2f7e7a5bef101c84560"
+dependencies = [
+ "alloc-no-stdlib",
+ "alloc-stdlib",
+ "brotli-decompressor",
+]
+
+[[package]]
+name = "brotli-decompressor"
+version = "5.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "874bb8112abecc98cbd6d81ea4fa7e94fb9449648c93cc89aa40c81c24d7de03"
+dependencies = [
+ "alloc-no-stdlib",
+ "alloc-stdlib",
+]
+
 [[package]]
 name = "bumpalo"
 version = "3.17.0"
@@ -923,6 +983,12 @@ dependencies = [
 "syn 1.0.109",
 ]

+[[package]]
+name = "bytecount"
+version = "0.6.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "175812e0be2bccb6abe50bb8d566126198344f707e304f45c648fd8f2cc0365e"
+
 [[package]]
 name = "bytemuck"
 version = "1.23.0"
@@ -953,12 +1019,23 @@ dependencies = [
 "rustversion",
 ]

+[[package]]
+name = "cbc"
+version = "0.1.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "26b52a9543ae338f279b96b0b9fed9c8093744685043739079ce85cd58f289a6"
+dependencies = [
+ "cipher",
+]
+
 [[package]]
 name = "cc"
 version = "1.2.21"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "8691782945451c1c383942c4874dbe63814f61cb57ef773cda2972682b7bb3c0"
 dependencies = [
+ "jobserver",
+ "libc",
 "shlex",
 ]

@@ -1019,6 +1096,12 @@ dependencies = [
 "unicode-security",
 ]

+[[package]]
+name = "cff-parser"
+version = "0.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "31f5b6e9141c036f3ff4ce7b2f7e432b0f00dee416ddcd4f17741d189ddc2e9d"
+
 [[package]]
 name = "cfg-if"
 version = "1.0.0"
@@ -1221,6 +1304,7 @@ dependencies = [
 "axum_session_auth",
 "axum_session_surreal",
 "axum_typed_multipart",
+ "bytes",
 "chrono",
 "chrono-tz",
 "config",
@@ -1233,15 +1317,18 @@ dependencies = [
 "minijinja-autoreload",
 "minijinja-contrib",
 "minijinja-embed",
+ "object_store 0.11.2",
 "reqwest",
 "serde",
 "serde_json",
 "sha2",
+ "state-machines",
 "surrealdb",
 "surrealdb-migrations",
 "tempfile",
 "thiserror 1.0.69",
 "tokio",
+ "tokio-retry",
 "tracing",
 "url",
 "uuid",
@@ -1258,6 +1345,7 @@ dependencies = [
 "futures",
 "serde",
 "serde_json",
+ "state-machines",
 "surrealdb",
 "thiserror 1.0.69",
 "tokio",
@@ -1265,6 +1353,26 @@ dependencies = [
 "uuid",
 ]

+[[package]]
+name = "compression-codecs"
+version = "0.4.30"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "485abf41ac0c8047c07c87c72c8fb3eb5197f6e9d7ded615dfd1a00ae00a0f64"
+dependencies = [
+ "brotli",
+ "compression-core",
+ "flate2",
+ "memchr",
+ "zstd",
+ "zstd-safe",
+]
+
+[[package]]
+name = "compression-core"
+version = "0.4.29"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e47641d3deaf41fb1538ac1f54735925e275eaf3bf4d55c81b137fba797e5cbb"
+
 [[package]]
 name = "concurrent-queue"
 version = "2.5.0"
@@ -1765,12 +1873,6 @@ dependencies = [
 "dtoa",
 ]

-[[package]]
-name = "dyn-clone"
-version = "1.0.19"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "1c7a8fb8a9fbf66c1f703fe16184d10ca0ee9d23be5b4436400408ba54a95005"
-
 [[package]]
 name = "earcutr"
 version = "0.4.3"
@@ -1781,6 +1883,15 @@ dependencies = [
 "num-traits",
 ]

+[[package]]
+name = "ecb"
+version = "0.1.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1a8bfa975b1aec2145850fcaa1c6fe269a16578c44705a532ae3edc92b8881c7"
+dependencies = [
+ "cipher",
+]
+
 [[package]]
 name = "either"
 version = "1.15.0"
@@ -1823,16 +1934,6 @@ version = "1.0.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "877a4ace8713b0bcf2a4e7eec82529c029f1d0619886d18145fea96c3ffe5c0f"

-[[package]]
-name = "erased-serde"
-version = "0.4.6"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "e004d887f51fcb9fef17317a2f3525c887d8aa3f4f50fed920816a688284a5b7"
-dependencies = [
- "serde",
- "typeid",
-]
-
 [[package]]
 name = "errno"
 version = "0.3.11"
@@ -1843,6 +1944,15 @@ dependencies = [
 "windows-sys 0.59.0",
 ]

+[[package]]
+name = "euclid"
+version = "0.20.14"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2bb7ef65b3777a325d1eeefefab5b6d4959da54747e33bd6258e789640f307ad"
+dependencies = [
+ "num-traits",
+]
+
 [[package]]
 name = "event-listener"
 version = "5.4.0"
@@ -2405,6 +2515,7 @@ dependencies = [
 "axum_session_auth",
 "axum_session_surreal",
 "axum_typed_multipart",
+ "chrono",
 "chrono-tz",
 "common",
 "composite-retrieval",
@@ -2415,7 +2526,6 @@ dependencies = [
 "minijinja-autoreload",
 "minijinja-contrib",
 "minijinja-embed",
- "plotly",
 "serde",
 "serde_json",
 "surrealdb",
@@ -2426,6 +2536,8 @@ dependencies = [
 "tower-http",
 "tower-serve-static",
 "tracing",
+ "url",
+ "uuid",
 ]

 [[package]]
@@ -2504,15 +2616,6 @@ version = "1.0.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "df3b46402a9d5adb4c86a0cf463f42e19994e3ee891101b1841f30a545cb49a9"

-[[package]]
-name = "humansize"
-version = "2.1.3"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "6cb51c9a029ddc91b07a787f1d86b53ccfa49b0e86688c946ebe8d3555685dd7"
-dependencies = [
- "libm",
-]
-
 [[package]]
 name = "humantime"
 version = "2.2.0"
@@ -2816,17 +2919,22 @@ name = "ingestion-pipeline"
 version = "0.1.0"
 dependencies = [
 "async-openai",
+ "async-trait",
 "axum",
 "axum_typed_multipart",
+ "base64 0.22.1",
 "chrono",
 "common",
 "composite-retrieval",
 "dom_smoothie",
 "futures",
 "headless_chrome",
+ "lopdf 0.32.0",
+ "pdf-extract",
 "reqwest",
 "serde",
 "serde_json",
+ "state-machines",
 "surrealdb",
 "tempfile",
 "text-splitter",
@@ -2862,6 +2970,7 @@ version = "0.1.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "879f10e63c20629ecabbb64a8010319738c66a5cd0c29b02d63d272b03751d01"
 dependencies = [
+ "block-padding",
 "generic-array",
 ]

@@ -2928,6 +3037,16 @@ version = "1.0.15"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "4a5f13b858c8d314ee3e8f639011f7ccefe71f97f96e50151fb991f267928e2c"

+[[package]]
+name = "jobserver"
+version = "0.1.34"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9afb3de4395d6b3e67a780b6de64b51c978ecf11cb9a462c66be7d4ca9039d33"
+dependencies = [
+ "getrandom 0.3.2",
+ "libc",
+]
+
 [[package]]
 name = "js-sys"
 version = "0.3.77"
@@ -3081,6 +3200,12 @@ dependencies = [
 "thiserror 1.0.69",
 ]

+[[package]]
+name = "linked-hash-map"
+version = "0.5.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0717cef1bc8b636c6e1c1bbdefc09e6322da8a9321966e8928ef80d20f7f770f"
+
 [[package]]
 name = "linux-raw-sys"
 version = "0.9.4"
@@ -3109,6 +3234,51 @@ version = "0.4.27"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "13dc2df351e3202783a1fe0d44375f7295ffb4049267b0f3018346dc122a1d94"

+[[package]]
+name = "lopdf"
+version = "0.32.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e775e4ee264e8a87d50a9efef7b67b4aa988cf94e75630859875fc347e6c872b"
+dependencies = [
+ "chrono",
+ "encoding_rs",
+ "flate2",
+ "itoa",
+ "linked-hash-map",
+ "log",
+ "md5",
+ "nom 7.1.3",
+ "rayon",
+ "time",
+ "weezl",
+]
+
+[[package]]
+name = "lopdf"
+version = "0.36.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "59fa2559e99ba0f26a12458aabc754432c805bbb8cba516c427825a997af1fb7"
+dependencies = [
+ "aes",
+ "bitflags 2.9.0",
+ "cbc",
+ "ecb",
+ "encoding_rs",
+ "flate2",
+ "indexmap 2.9.0",
+ "itoa",
+ "log",
+ "md-5",
+ "nom 8.0.0",
+ "nom_locate",
+ "rand 0.9.1",
+ "rangemap",
+ "sha2",
+ "stringprep",
+ "thiserror 2.0.12",
+ "weezl",
+]
+
 [[package]]
 name = "lru"
 version = "0.12.5"
@@ -3126,7 +3296,7 @@ checksum = "c41e0c4fef86961ac6d6f8a82609f55f31b05e4fce149ac5710e439df7619ba4"

 [[package]]
 name = "main"
-version = "0.1.1"
+version = "0.2.5"
 dependencies = [
 "anyhow",
 "api-router",
@@ -3141,8 +3311,10 @@ dependencies = [
 "surrealdb",
 "thiserror 1.0.69",
 "tokio",
+ "tower",
 "tracing",
 "tracing-subscriber",
+ "uuid",
 ]

 [[package]]
@@ -3222,6 +3394,12 @@ dependencies = [
 "digest",
 ]

+[[package]]
+name = "md5"
+version = "0.7.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "490cc448043f947bae3cbee9c203358d62dbee0db12107a74be5c30ccfd09771"
+
 [[package]]
 name = "memchr"
 version = "2.7.4"
@@ -3454,6 +3632,17 @@ dependencies = [
 "memchr",
 ]

+[[package]]
+name = "nom_locate"
+version = "5.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0b577e2d69827c4740cba2b52efaad1c4cc7c73042860b199710b3575c68438d"
+dependencies = [
+ "bytecount",
+ "memchr",
+ "nom 8.0.0",
+]
+
 [[package]]
 name = "nonempty"
 version = "0.7.0"
@@ -3576,6 +3765,27 @@ dependencies = [
 "memchr",
 ]

+[[package]]
+name = "object_store"
+version = "0.11.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3cfccb68961a56facde1163f9319e0d15743352344e7808a11795fb99698dcaf"
+dependencies = [
+ "async-trait",
+ "bytes",
+ "chrono",
+ "futures",
+ "humantime",
+ "itertools 0.13.0",
+ "parking_lot",
+ "percent-encoding",
+ "snafu",
+ "tokio",
+ "tracing",
+ "url",
+ "walkdir",
+]
+
 [[package]]
 name = "object_store"
 version = "0.12.0"
@@ -3755,6 +3965,23 @@ dependencies = [
 "sha2",
 ]

+[[package]]
+name = "pdf-extract"
+version = "0.9.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6c2f44c6c642e359e2fe7f662bf5438db3811b6b4be60afc6de04b619ce51e1a"
+dependencies = [
+ "adobe-cmap-parser",
+ "cff-parser",
+ "encoding_rs",
+ "euclid",
+ "log",
+ "lopdf 0.36.0",
+ "postscript",
+ "type1-encoding-parser",
+ "unicode-normalization",
+]
+
 [[package]]
 name = "pem"
 version = "3.0.5"
@@ -3934,36 +4161,6 @@ version = "0.3.32"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "7edddbd0b52d732b21ad9a5fab5c704c14cd949e5e9a1ec5929a24fded1b904c"

-[[package]]
-name = "plotly"
-version = "0.12.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "0746e9faf2b051db76470fd428cbc0db792db05346dedaae4a75b16d7be503b5"
-dependencies = [
- "dyn-clone",
- "erased-serde",
- "once_cell",
- "plotly_derive",
- "rand 0.8.5",
- "rinja",
- "serde",
- "serde_json",
- "serde_repr",
- "serde_with",
-]
-
-[[package]]
-name = "plotly_derive"
-version = "0.12.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "2d683930282f098b9f524e2596e3e63483507ac499231c96127fcb166bc05d26"
-dependencies = [
- "darling",
- "proc-macro2",
- "quote",
- "syn 2.0.101",
-]
-
 [[package]]
 name = "polyval"
 version = "0.6.2"
@@ -3976,6 +4173,18 @@ dependencies = [
 "universal-hash",
 ]

+[[package]]
+name = "pom"
+version = "1.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "60f6ce597ecdcc9a098e7fddacb1065093a3d66446fa16c675e7e71d1b5c28e6"
+
+[[package]]
+name = "postscript"
+version = "0.14.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "78451badbdaebaf17f053fd9152b3ffb33b516104eacb45e7864aaa9c712f306"
+
 [[package]]
 name = "powerfmt"
 version = "0.2.0"
@@ -4241,6 +4450,12 @@ dependencies = [
 "getrandom 0.3.2",
 ]

+[[package]]
+name = "rangemap"
+version = "1.6.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f93e7e49bb0bf967717f7bd674458b3d6b0c5f48ec7e3038166026a69fc22223"
+
 [[package]]
 name = "rawpointer"
 version = "0.2.1"
@@ -4501,49 +4716,6 @@ dependencies = [
 "windows-sys 0.52.0",
 ]

-[[package]]
-name = "rinja"
-version = "0.3.5"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "3dc4940d00595430b3d7d5a01f6222b5e5b51395d1120bdb28d854bb8abb17a5"
-dependencies = [
- "humansize",
- "itoa",
- "percent-encoding",
- "rinja_derive",
- "serde",
- "serde_json",
-]
-
-[[package]]
-name = "rinja_derive"
-version = "0.3.5"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "08d9ed0146aef6e2825f1b1515f074510549efba38d71f4554eec32eb36ba18b"
-dependencies = [
- "basic-toml",
- "memchr",
- "mime",
- "mime_guess",
- "proc-macro2",
- "quote",
- "rinja_parser",
- "rustc-hash",
- "serde",
- "syn 2.0.101",
-]
-
-[[package]]
-name = "rinja_parser"
-version = "0.3.5"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "93f9a866e2e00a7a1fb27e46e9e324a6f7c0e7edc4543cae1d38f4e4a100c610"
-dependencies = [
- "memchr",
- "nom 7.1.3",
- "serde",
-]
-
 [[package]]
 name = "rkyv"
 version = "0.7.45"
@@ -4846,9 +5018,9 @@ checksum = "1c107b6f4780854c8b126e228ea8869f4d7b71260f962fefb57b996b8959ba6b"

 [[package]]
 name = "secrecy"
-version = "0.8.0"
+version = "0.10.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "9bd1c54ea06cfd2f6b63219704de0b9b4f72dcc2b8fdef820be6cd799780e91e"
+checksum = "e891af845473308773346dc847b2c23ee78fe442e0472ac50e22a18a93d3ae5a"
 dependencies = [
 "serde",
 "zeroize",
@@ -4994,17 +5166,6 @@ dependencies = [
 "serde",
 ]

-[[package]]
-name = "serde_repr"
-version = "0.1.20"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "175ee3e80ae9982737ca543e96133087cbd9a485eecc3bc4de9c1a37b47ea59c"
-dependencies = [
- "proc-macro2",
- "quote",
- "syn 2.0.101",
-]
-
 [[package]]
 name = "serde_spanned"
 version = "0.6.8"
@@ -5159,6 +5320,27 @@ dependencies = [
 "serde",
 ]

+[[package]]
+name = "snafu"
+version = "0.8.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6e84b3f4eacbf3a1ce05eac6763b4d629d60cbc94d632e4092c54ade71f1e1a2"
+dependencies = [
+ "snafu-derive",
+]
+
+[[package]]
+name = "snafu-derive"
+version = "0.8.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c1c97747dbf44bb1ca44a561ece23508e99cb592e862f22222dcf42f51d1e451"
+dependencies = [
+ "heck",
+ "proc-macro2",
+ "quote",
+ "syn 2.0.101",
+]
+
 [[package]]
 name = "snap"
 version = "1.1.1"
@@ -5223,6 +5405,34 @@ dependencies = [
 "windows-sys 0.59.0",
 ]

+[[package]]
+name = "state-machines"
+version = "0.2.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "806ba0bf43ae158b229036d8a84601649a58d9761e718b5e0e07c2953803f4c1"
+dependencies = [
+ "state-machines-core",
+ "state-machines-macro",
+]
+
+[[package]]
+name = "state-machines-core"
+version = "0.2.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "949cc50e84bed6234117f28a0ba2980dc35e9c17984ffe4e0a3364fba3e77540"
+
+[[package]]
+name = "state-machines-macro"
+version = "0.2.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8322f5aa92d31b3c05faa1ec3231b82da479a20706836867d67ae89ce74927bd"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "state-machines-core",
+ "syn 2.0.101",
+]
+
 [[package]]
 name = "static_assertions_next"
 version = "1.1.2"
@@ -5266,6 +5476,17 @@ dependencies = [
 "quote",
 ]

+[[package]]
+name = "stringprep"
+version = "0.1.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7b4df3d392d81bd458a8a621b8bffbd2302a12ffe288a9d931670948749463b1"
+dependencies = [
+ "unicode-bidi",
+ "unicode-normalization",
+ "unicode-properties",
+]
+
 [[package]]
 name = "strsim"
 version = "0.11.1"
@@ -5388,7 +5609,7 @@ dependencies = [
 "ndarray-stats",
 "num-traits",
 "num_cpus",
- "object_store",
+ "object_store 0.12.0",
 "parking_lot",
 "pbkdf2",
 "pharos",
@@ -5793,6 +6014,17 @@ dependencies = [
 "tokio",
 ]

+[[package]]
+name = "tokio-retry"
+version = "0.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7f57eb36ecbe0fc510036adff84824dd3c24bb781e21bfa67b69d556aa85214f"
+dependencies = [
+ "pin-project",
+ "rand 0.8.5",
+ "tokio",
+]
+
 [[package]]
 name = "tokio-rustls"
 version = "0.26.2"
@@ -5900,8 +6132,10 @@ version = "0.6.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "403fa3b783d4b626a8ad51d766ab03cb6d2dbfc46b1c5d4448395e6628dc9697"
 dependencies = [
+ "async-compression",
 "bitflags 2.9.0",
 "bytes",
+ "futures-core",
 "futures-util",
 "http",
 "http-body",
@@ -6086,10 +6320,13 @@ dependencies = [
 ]

 [[package]]
-name = "typeid"
-version = "1.0.3"
+name = "type1-encoding-parser"
+version = "0.1.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "bc7d623258602320d5c55d1bc22793b57daff0ec7efc270ea7d55ce1d5f5471c"
+checksum = "d3d6cc09e1a99c7e01f2afe4953789311a1c50baebbdac5b477ecf78e2e92a5b"
+dependencies = [
+ "pom",
+]

 [[package]]
 name = "typenum"
@@ -6126,6 +6363,12 @@ version = "2.8.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "75b844d17643ee918803943289730bec8aac480150456169e647ed0b576ba539"

+[[package]]
+name = "unicode-bidi"
+version = "0.3.18"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5c1cb5db39152898a79168971543b1cb5020dff7fe43c8dc468b0885f5e29df5"
+
 [[package]]
 name = "unicode-ident"
 version = "1.0.18"
@@ -6141,6 +6384,12 @@ dependencies = [
 "tinyvec",
 ]

+[[package]]
+name = "unicode-properties"
+version = "0.1.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e70f2a8b45122e719eb623c01822704c4e0907e7e426a05927e1a1cfff5b75d0"
+
 [[package]]
 name = "unicode-script"
 version = "0.5.7"
@@ -6470,6 +6719,12 @@ dependencies = [
 "rustls-pki-types",
 ]

+[[package]]
+name = "weezl"
+version = "0.1.10"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a751b3277700db47d3e574514de2eced5e54dc8a5436a3bf7a0b248b2cee16f3"
+
 [[package]]
 name = "which"
 version = "7.0.3"
@@ -6994,3 +7249,31 @@ dependencies = [
 "quote",
 "syn 2.0.101",
 ]
+
+[[package]]
+name = "zstd"
+version = "0.13.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e91ee311a569c327171651566e07972200e76fcfe2242a4fa446149a3881c08a"
+dependencies = [
+ "zstd-safe",
+]
+
+[[package]]
+name = "zstd-safe"
+version = "7.2.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8f49c4d5f0abb602a93fb8736af2a4f4dd9512e36f7f570d66e65ff867ed3b9d"
+dependencies = [
+ "zstd-sys",
+]
+
+[[package]]
+name = "zstd-sys"
+version = "2.0.16+zstd.1.5.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "91e19ebc2adc8f83e43039e79776e3fda8ca919132d68a1fed6a5faca2683748"
+dependencies = [
+ "cc",
+ "pkg-config",
+]
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -12,7 +12,7 @@ resolver = "2"

 [workspace.dependencies]
 anyhow = "1.0.94"
-async-openai = "0.24.1"
+async-openai = "0.29.3"
 async-stream = "0.3.6"
 async-trait = "0.1.88"
 axum-htmx = "0.7.0"
@@ -34,7 +34,6 @@ minijinja-autoreload = "2.5.0"
 minijinja-contrib = { version = "2.6.0", features = ["datetime", "timezone"] }
 minijinja-embed = { version = "2.8.0" }
 minijinja = { version = "2.5.0", features = ["loader", "multi_template"] }
-plotly = "0.12.1"
 reqwest = {version = "0.12.12", features = ["charset", "json"]}
 serde_json = "1.0.128"
 serde = { version = "1", features = ["derive"] }
@@ -46,13 +45,29 @@ text-splitter = "0.18.1"
 thiserror = "1.0.63"
 tokio-util = { version = "0.7.15", features = ["io"] }
 tokio = { version = "1", features = ["full"] }
-tower-http = { version = "0.6.2", features = ["fs"] }
+tower-http = { version = "0.6.2", features = ["fs", "compression-full"] }
 tower-serve-static = "0.1.1"
 tracing = "0.1.40"
 tracing-subscriber = { version = "0.3.18", features = ["env-filter"] }
 url = { version = "2.5.2", features = ["serde"] }
 uuid = { version = "1.10.0", features = ["v4", "serde"] }
+tokio-retry = "0.3.0"
+base64 = "0.22.1"
+object_store = { version = "0.11.2" }
+bytes = "1.7.1"
+state-machines = "0.2.0"

 [profile.dist]
 inherits = "release"
 lto = "thin"
+
+[workspace.lints.clippy]
+perf = { level = "warn",  priority = -1 }
+pedantic = { level = "warn",  priority = -1 }
+nursery = { level = "warn",  priority = -1 }
+cargo = { level = "warn",  priority = -1 }
+
+needless_question_mark = "allow"
+single_call_fn = "allow"
+must_use_candidate = "allow"
+missing_errors_doc = "allow"
--- a/README.md
+++ b/README.md
@@ -6,28 +6,38 @@
 [![License: AGPL v3](https://img.shields.io/badge/License-AGPL_v3-blue.svg)](https://www.gnu.org/licenses/agpl-3.0)
 [![Latest Release](https://img.shields.io/github/v/release/perstarkse/minne?sort=semver)](https://github.com/perstarkse/minne/releases/latest)

+![Screenshot](screenshot-graph.webp)
+
+## Demo deployment
+
+To test _Minne_ out, enter [this](https://minne-demo.stark.pub) read-only demo deployment to view and test functionality out.
+
 ## The "Why" Behind Minne

 For a while I've been fascinated by Zettelkasten-style PKM systems. While tools like Logseq and Obsidian are excellent, I found the manual linking process to be a hindrance for me. I also wanted a centralized storage and easy access across devices.

-While developing Minne, I discovered [KaraKeep](https://karakeep.com/) (formerly Hoarder), which is an excellent application in a similar space – you probably want to check it out! However, if you're specifically interested in a PKM that leverages a **SurrealDB as its backend, utilizing both vector and graph retrieval**, offers the **possibility to chat with your knowledge resource**, and provides a blend of manual and AI-driven organization, then Minne might be for you.
+While developing Minne, I discovered [KaraKeep](https://karakeep.com/) (formerly Hoarder), which is an excellent application in a similar space – you probably want to check it out! However, if you're interested in a PKM that builds an automatic network between related concepts using AI, offers search and the **possibility to chat with your knowledge resource**, and provides a blend of manual and AI-driven organization, then Minne might be worth testing.

 ## Core Philosophy & Features

-Minne is designed to make it incredibly easy to save snippets of text, URLs, and other content(coming if there is demand). Simply send content along with a category tag. Minne then ingests this, leveraging AI to create relevant nodes and relationships within its graph database, alongside your manual categorization. This graph backend, powered by SurrealDB, allows for discoverable connections between your pieces of knowledge.
+Minne is designed to make it incredibly easy to save snippets of text, URLs, and other content (limited, pending demand). Simply send content along with a category tag. Minne then ingests this, leveraging AI to create relevant nodes and relationships within its graph database, alongside your manual categorization. This graph backend allows for discoverable connections between your pieces of knowledge.

-You can converse with your knowledge base through an LLM-powered chat interface (via OpenAI API). For those who like to see the bigger picture, Minne also includes an **experimental feature to visually explore your knowledge graph.**
+You can converse with your knowledge base through an LLM-powered chat interface (via OpenAI compatible API, like Ollama or others). For those who like to see the bigger picture, Minne also includes an feature to visually explore your knowledge graph.

-The application is built for speed and efficiency using Rust with a Server-Side Rendered (SSR) frontend (HTMX and minimal JavaScript). It's fully responsive, offering a complete mobile interface for reading, editing, and managing your content, including the graph database itself. **PWA (Progressive Web App) support** means you can "install" Minne to your device for a native-like experience. For quick capture on the go, especially on iOS, a [**dedicated Shortcut**](https://www.icloud.com/shortcuts/9aa960600ec14329837ba4169f57a166) makes sending content to your Minne instance a breeze.
+You may switch and choose between models used, and have the possiblity to change the prompts to your liking. There is the option to change embeddings length, making it easy to test another embedding model.
+
+The application is built for speed and efficiency using Rust with a Server-Side Rendered (SSR) frontend (HTMX and minimal JavaScript). It's fully responsive, offering a complete mobile interface for reading, editing, and managing your content, including the graph database itself. **PWA (Progressive Web App) support** means you can "install" Minne to your device for a native-like experience. For quick capture on the go on iOS, a [**Shortcut**](https://www.icloud.com/shortcuts/e433fbd7602f4e2eaa70dca162323477) makes sending content to your Minne instance a breeze.
+
+A hybrid retrieval layer blends embeddings, full-text search, and graph signals to surface the best context when augmenting chat responses and when analyzing new content during ingestion.

 Minne is open source (AGPL), self-hostable, and can be deployed flexibly: via Nix, Docker Compose, pre-built binaries, or by building from source. It can run as a single `main` binary or as separate `server` and `worker` processes for optimized resource allocation.

 ## Tech Stack

- **Backend:** Rust
- **Frontend:** Server-Side Rendering (SSR) with HTMX, Axum, Minijinja, and plain JavaScript for interactivity.
- **Database:** SurrealDB (as a graph database)
- **AI Integration:** OpenAI API (for chat and content processing)
+- **Backend:** Rust. Server-Side Rendering (SSR). Axum. Minijinja for templating.
+- **Frontend:** HTML. HTMX and plain JavaScript for interactivity.
+- **Database:** SurrealDB
+- **AI Integration:** OpenAI API compatible endpoint (for chat and content processing), with support for structured outputs.
 - **Web Content Processing:** Relies on a Chromium instance for robust webpage fetching/rendering.

 ## Prerequisites
@@ -62,7 +72,7 @@ This is a great way to manage Minne and its SurrealDB dependency together.
 1. Create a `docker-compose.yml` file:

   ```yaml
-   version: '3.8'
+   version: "3.8"
   services:
     minne:
       image: ghcr.io/perstarkse/minne:latest # Pulls the latest pre-built image
@@ -80,6 +90,8 @@ This is a great way to manage Minne and its SurrealDB dependency together.
         SURREALDB_DATABASE: "minne_db"
         SURREALDB_NAMESPACE: "minne_ns"
         OPENAI_API_KEY: "your_openai_api_key_here" # IMPORTANT: Replace with your actual key
+         #OPENAI_BASE_URL: "your_ollama_address" # Uncomment this and change it to override the default openai base url
+         HTTP_PORT: 3000
         DATA_DIR: "/data" # Data directory inside the container
         RUST_LOG: "minne=info,tower_http=info" # Example logging level
       volumes:
@@ -128,18 +140,6 @@ This is a great way to manage Minne and its SurrealDB dependency together.
       driver: bridge
   ```

-1. Create a `.env` file in the same directory as your `docker-compose.yml` (recommended for sensitive data):
-
-   ```env
-   OPENAI_API_KEY="your_openai_api_key_here"
-   # You can override other environment variables here if needed
-   # e.g., if you want to expose SurrealDB differently or use different credentials.
-   # SURREALDB_USERNAME_MINNE="custom_user" # If changing Minne's access credentials
-   # SURREALDB_PASSWORD_MINNE="custom_pass"
-   ```
-
-   *(If using a `.env` file, ensure variables in `docker-compose.yml`'s `environment` section reference them like `${OPENAI_API_KEY}` or are directly set if not sensitive and common across setups)*
-
 1. Run:

   ```bash
@@ -179,7 +179,7 @@ Binaries for Windows, macOS, and Linux (combined `main` version) are available o
     ```bash
     cargo run --release --bin worker
     ```
-   The compiled binaries will be in `target/release/`.
+     The compiled binaries will be in `target/release/`.

 ## Configuration

@@ -192,14 +192,14 @@ Minne can be configured using environment variables or a `config.yaml` file plac
 - `SURREALDB_PASSWORD`: Password for SurrealDB (e.g., `root_password`).
 - `SURREALDB_DATABASE`: Database name in SurrealDB (e.g., `minne_db`).
 - `SURREALDB_NAMESPACE`: Namespace in SurrealDB (e.g., `minne_ns`).
- `OPENAI_API_KEY`: Your API key for OpenAI (e.g., `sk-YourActualOpenAIKeyGoesHere`).
- `DATA_DIR`: Directory to store local data like fetched webpage content (e.g., `./minne_app_data`).
+- `OPENAI_API_KEY`: Your API key for OpenAI compatible endpoint (e.g., `sk-YourActualOpenAIKeyGoesHere`).
+- `HTTP_PORT`: Port for the Minne server to listen on (Default: `3000`).

 **Optional Configuration:**

 - `RUST_LOG`: Controls logging level (e.g., `minne=info,tower_http=debug`).
- `HTTP_PORT`: Port for the Minne server to listen on (Default: `3000`).
- `CHROME_ADDRESS`: Address of a remote Chrome DevTools Protocol endpoint (e.g., `http://localhost:9222`, if not using local Chromium managed by Minne/Docker/Nix).
+- `DATA_DIR`: Directory to store local data like fetched webpage content (e.g., `./data`).
+- `OPENAI_BASE_URL`: Base URL to a OpenAI API provider, such as Ollama.

 **Example `config.yaml`:**

@@ -211,8 +211,8 @@ surrealdb_database: "minne_db"
 surrealdb_namespace: "minne_ns"
 openai_api_key: "sk-YourActualOpenAIKeyGoesHere"
 data_dir: "./minne_app_data"
+http_port: 3000
 # rust_log: "info"
-# http_port: 3000
 ```

 ## Application Architecture (Binaries)
@@ -231,23 +231,50 @@ Once Minne is running:

 1. Access the web interface at `http://localhost:3000` (or your configured port).
 1. On iOS, consider setting up the [Minne iOS Shortcut](https://www.icloud.com/shortcuts/9aa960600ec14329837ba4169f57a166) for effortless content sending. **Add the shortcut, replace the [insert_url] and the [insert_api_key] snippets**.
-1. Start adding notes, URLs and explore your growing knowledge graph.
+1. Add notes, URLs, **audio files**, and explore your growing knowledge graph.
 1. Engage with the chat interface to query your saved content.
 1. Try the experimental visual graph explorer to see connections.

+## AI Configuration & Model Selection
+
+Minne relies on an OpenAI-compatible API for processing content, generating graph relationships, and powering the chat feature.
+
+**Environment Variables / `config.yaml` keys:**
+
+- `OPENAI_API_KEY` (required): Your API key for the chosen AI provider.
+- `OPENAI_BASE_URL` (optional): Use this to override the default OpenAI API URL (`https://api.openai.com/v1`). This is essential for using local models via services like Ollama, or other API providers.
+  - **Example for Ollama:** `http://<your-ollama-ip>:11434/v1`
+
+### Changing Models
+
+Once you have configured the `OPENAI_BASE_URL` to point to your desired provider, you can select the specific models Minne should use.
+
+1. Navigate to the `/admin` page in your Minne instance.
+1. The page will list the models available from your configured endpoint. You can select different models for processing content and for chat.
+1. **Important:** For content processing, Minne relies on structured outputs (function calling). The model and provider you select for this task **must** support this feature.
+1. **Embedding Dimensions:** If you change the embedding model, you **must** update the "Embedding Dimensions" setting in the admin panel to match the output dimensions of your new model (e.g., `text-embedding-3-small` uses 1536, `nomic-embed-text` uses 768). Mismatched dimensions will cause errors. Some newer models will accept a dimension argument, and for these setting the dimensions to whatever should work.
+
 ## Roadmap

 I've developed Minne primarily for my own use, but having been in the selfhosted space for a long time, and using the efforts by others, I thought I'd share with the community. Feature requests are welcome.
 The roadmap as of now is:

- Handle uploaded images wisely.
- An updated explorer of the graph database, and potentially the vector space.
+~~- Handle uploaded images wisely.~~
+~~- An updated explorer of the graph database.~~
 - A TUI frontend which opens your system default editor for improved writing and document management.

 ## Contributing

 Contributions are welcome! Whether it's bug reports, feature suggestions, documentation improvements, or code contributions, please feel free to open an issue or submit a pull request.

+## Development
+
+Run test with
+```rust
+cargo test
+```
+There is currently a variety of unit tests for commonly used functions. Additional tests, especially integration tests would be very welcome.
+
 ## License

 Minne is licensed under the **GNU Affero General Public License v3.0 (AGPL-3.0)**. See the [LICENSE](LICENSE) file for details. This means if you run a modified version of Minne as a network service, you must also offer the source code of that modified version to its users.
--- a/api-router/Cargo.toml
+++ b/api-router/Cargo.toml
@@ -4,6 +4,9 @@ version = "0.1.0"
 edition = "2021"
 license = "AGPL-3.0-or-later"

+[lints]
+workspace = true
+
 [dependencies]
 tokio = { workspace = true }
 serde = { workspace = true }
--- a/api-router/src/api_state.rs
+++ b/api-router/src/api_state.rs
@@ -23,7 +23,7 @@ impl ApiState {

        surreal_db_client.apply_migrations().await?;

-        let app_state = ApiState {
+        let app_state = Self {
            db: surreal_db_client.clone(),
            config: config.clone(),
        };
--- a/api-router/src/error.rs
+++ b/api-router/src/error.rs
@@ -27,40 +27,40 @@ impl From<AppError> for ApiError {
        match err {
            AppError::Database(_) | AppError::OpenAI(_) => {
                tracing::error!("Internal error: {:?}", err);
-                ApiError::InternalError("Internal server error".to_string())
+                Self::InternalError("Internal server error".to_string())
            }
-            AppError::NotFound(msg) => ApiError::NotFound(msg),
-            AppError::Validation(msg) => ApiError::ValidationError(msg),
-            AppError::Auth(msg) => ApiError::Unauthorized(msg),
-            _ => ApiError::InternalError("Internal server error".to_string()),
+            AppError::NotFound(msg) => Self::NotFound(msg),
+            AppError::Validation(msg) => Self::ValidationError(msg),
+            AppError::Auth(msg) => Self::Unauthorized(msg),
+            _ => Self::InternalError("Internal server error".to_string()),
        }
    }
 }
 impl IntoResponse for ApiError {
    fn into_response(self) -> Response {
        let (status, error_response) = match self {
-            ApiError::InternalError(message) => (
+            Self::InternalError(message) => (
                StatusCode::INTERNAL_SERVER_ERROR,
                ErrorResponse {
                    error: message,
                    status: "error".to_string(),
                },
            ),
-            ApiError::ValidationError(message) => (
+            Self::ValidationError(message) => (
                StatusCode::BAD_REQUEST,
                ErrorResponse {
                    error: message,
                    status: "error".to_string(),
                },
            ),
-            ApiError::NotFound(message) => (
+            Self::NotFound(message) => (
                StatusCode::NOT_FOUND,
                ErrorResponse {
                    error: message,
                    status: "error".to_string(),
                },
            ),
-            ApiError::Unauthorized(message) => (
+            Self::Unauthorized(message) => (
                StatusCode::UNAUTHORIZED,
                ErrorResponse {
                    error: message,
--- a/api-router/src/lib.rs
+++ b/api-router/src/lib.rs
@@ -6,7 +6,7 @@ use axum::{
    Router,
 };
 use middleware_api_auth::api_auth;
-use routes::{categories::get_categories, ingress::ingest_data};
+use routes::{categories::get_categories, ingress::ingest_data, liveness::live, readiness::ready};

 pub mod api_state;
 pub mod error;
@@ -19,9 +19,17 @@ where
    S: Clone + Send + Sync + 'static,
    ApiState: FromRef<S>,
 {
-    Router::new()
+    // Public, unauthenticated endpoints (for k8s/systemd probes)
+    let public = Router::new()
+        .route("/ready", get(ready))
+        .route("/live", get(live));
+
+    // Protected API endpoints (require auth)
+    let protected = Router::new()
        .route("/ingress", post(ingest_data))
        .route("/categories", get(get_categories))
        .layer(DefaultBodyLimit::max(1024 * 1024 * 1024))
-        .route_layer(from_fn_with_state(app_state.clone(), api_auth))
+        .route_layer(from_fn_with_state(app_state.clone(), api_auth));
+
+    public.merge(protected)
 }
--- a/api-router/src/middleware_api_auth.rs
+++ b/api-router/src/middleware_api_auth.rs
@@ -13,14 +13,12 @@ pub async fn api_auth(
    mut request: Request,
    next: Next,
 ) -> Result<Response, ApiError> {
-    let api_key = extract_api_key(&request).ok_or(ApiError::Unauthorized(
-        "You have to be authenticated".to_string(),
-    ))?;
+    let api_key = extract_api_key(&request)
+        .ok_or_else(|| ApiError::Unauthorized("You have to be authenticated".to_string()))?;

    let user = User::find_by_api_key(&api_key, &state.db).await?;
-    let user = user.ok_or(ApiError::Unauthorized(
-        "You have to be authenticated".to_string(),
-    ))?;
+    let user =
+        user.ok_or_else(|| ApiError::Unauthorized("You have to be authenticated".to_string()))?;

    request.extensions_mut().insert(user);

@@ -37,7 +35,7 @@ fn extract_api_key(request: &Request) -> Option<String> {
                .headers()
                .get("Authorization")
                .and_then(|v| v.to_str().ok())
-                .and_then(|auth| auth.strip_prefix("Bearer ").map(|s| s.trim()))
+                .and_then(|auth| auth.strip_prefix("Bearer ").map(str::trim))
        })
        .map(String::from)
 }
--- a/api-router/src/routes/ingress.rs
+++ b/api-router/src/routes/ingress.rs
@@ -1,4 +1,4 @@
-use axum::{extract::State, http::StatusCode, response::IntoResponse, Extension};
+use axum::{extract::State, http::StatusCode, response::IntoResponse, Extension, Json};
 use axum_typed_multipart::{FieldData, TryFromMultipart, TypedMultipart};
 use common::{
    error::AppError,
@@ -8,6 +8,7 @@ use common::{
    },
 };
 use futures::{future::try_join_all, TryFutureExt};
+use serde_json::json;
 use tempfile::NamedTempFile;
 use tracing::info;

@@ -45,12 +46,10 @@ pub async fn ingest_data(

    let futures: Vec<_> = payloads
        .into_iter()
-        .map(|object| {
-            IngestionTask::create_and_add_to_db(object.clone(), user.id.clone(), &state.db)
-        })
+        .map(|object| IngestionTask::create_and_add_to_db(object, user.id.clone(), &state.db))
        .collect();

-    try_join_all(futures).await.map_err(AppError::from)?;
+    try_join_all(futures).await?;

-    Ok(StatusCode::OK)
+    Ok((StatusCode::OK, Json(json!({ "status": "success" }))))
 }
--- a/api-router/src/routes/liveness.rs
+++ b/api-router/src/routes/liveness.rs
@@ -0,0 +1,7 @@
+use axum::{http::StatusCode, response::IntoResponse, Json};
+use serde_json::json;
+
+/// Liveness probe: always returns 200 to indicate the process is running.
+pub async fn live() -> impl IntoResponse {
+    (StatusCode::OK, Json(json!({"status": "ok"})))
+}
--- a/api-router/src/routes/mod.rs
+++ b/api-router/src/routes/mod.rs
@@ -1,2 +1,4 @@
 pub mod categories;
 pub mod ingress;
+pub mod liveness;
+pub mod readiness;
--- a/api-router/src/routes/readiness.rs
+++ b/api-router/src/routes/readiness.rs
@@ -0,0 +1,25 @@
+use axum::{extract::State, http::StatusCode, response::IntoResponse, Json};
+use serde_json::json;
+
+use crate::api_state::ApiState;
+
+/// Readiness probe: returns 200 if core dependencies are ready, else 503.
+pub async fn ready(State(state): State<ApiState>) -> impl IntoResponse {
+    match state.db.client.query("RETURN true").await {
+        Ok(_) => (
+            StatusCode::OK,
+            Json(json!({
+                "status": "ok",
+                "checks": { "db": "ok" }
+            })),
+        ),
+        Err(e) => (
+            StatusCode::SERVICE_UNAVAILABLE,
+            Json(json!({
+                "status": "error",
+                "checks": { "db": "fail" },
+                "reason": e.to_string()
+            })),
+        ),
+    }
+}
--- a/common/Cargo.toml
+++ b/common/Cargo.toml
@@ -4,6 +4,9 @@ version = "0.1.0"
 edition = "2021"
 license = "AGPL-3.0-or-later"

+[lints]
+workspace = true
+
 [dependencies]
 # Workspace dependencies
 tokio = { workspace = true }
@@ -38,6 +41,10 @@ sha2 = { workspace = true }
 url = { workspace = true }
 uuid = { workspace = true } 
 surrealdb-migrations = { workspace = true } 
+tokio-retry = { workspace = true }
+object_store = { workspace = true }
+bytes = { workspace = true }
+state-machines = { workspace = true }


 [features]
--- a/common/migrations/20250503_215025_initial_setup.surql
+++ b/common/migrations/20250503_215025_initial_setup.surql
@@ -13,6 +13,8 @@ CREATE system_settings:current CONTENT {
    require_email_verification: false,
    query_model: "gpt-4o-mini",
    processing_model: "gpt-4o-mini",
+    embedding_model: "text-embedding-3-small",
+    embedding_dimensions: 1536,
    query_system_prompt: "You are a knowledgeable assistant with access to a specialized knowledge base. You will be provided with relevant knowledge entities from the database as context. Each knowledge entity contains a name, description, and type, representing different concepts, ideas, and information.\nYour task is to:\n1. Carefully analyze the provided knowledge entities in the context\n2. Answer user questions based on this information\n3. Provide clear, concise, and accurate responses\n4. When referencing information, briefly mention which knowledge entity it came from\n5. If the provided context doesn't contain enough information to answer the question confidently, clearly state this\n6. If only partial information is available, explain what you can answer and what information is missing\n7. Avoid making assumptions or providing information not supported by the context\n8. Output the references to the documents. Use the UUIDs and make sure they are correct!\nRemember:\n- Be direct and honest about the limitations of your knowledge\n- Cite the relevant knowledge entities when providing information, but only provide the UUIDs in the reference array\n- If you need to combine information from multiple entities, explain how they connect\n- Don't speculate beyond what's provided in the context\nExample response formats:\n\"Based on [Entity Name], [answer...]\"\n\"I found relevant information in multiple entries: [explanation...]\"\n\"I apologize, but the provided context doesn't contain information about [topic]\"",
    ingestion_system_prompt: "You are an AI assistant. You will receive a text content, along with user context and a category. Your task is to provide a structured JSON object representing the content in a graph format suitable for a graph database. You will also be presented with some existing knowledge_entities from the database, do not replicate these! Your task is to create meaningful knowledge entities from the submitted content. Try and infer as much as possible from the users context and category when creating these. If the user submits a large content, create more general entities. If the user submits a narrow and precise content, try and create precise knowledge entities.\nThe JSON should have the following structure:\n{\n\"knowledge_entities\": [\n{\n\"key\": \"unique-key-1\",\n\"name\": \"Entity Name\",\n\"description\": \"A detailed description of the entity.\",\n\"entity_type\": \"TypeOfEntity\"\n},\n// More entities...\n],\n\"relationships\": [\n{\n\"type\": \"RelationshipType\",\n\"source\": \"unique-key-1 or UUID from existing database\",\n\"target\": \"unique-key-1 or UUID from existing database\"\n},\n// More relationships...\n]\n}\nGuidelines:\n1. Do NOT generate any IDs or UUIDs. Use a unique `key` for each knowledge entity.\n2. Each KnowledgeEntity should have a unique `key`, a meaningful `name`, and a descriptive `description`.\n3. Define the type of each KnowledgeEntity using the following categories: Idea, Project, Document, Page, TextSnippet.\n4. Establish relationships between entities using types like RelatedTo, RelevantTo, SimilarTo.\n5. Use the `source` key to indicate the originating entity and the `target` key to indicate the related entity\"\n6. You will be presented with a few existing KnowledgeEntities that are similar to the current ones. They will have an existing UUID. When creating relationships to these entities, use their UUID.\n7. Only create relationships between existing KnowledgeEntities.\n8. Entities that exist already in the database should NOT be created again. If there is only a minor overlap, skip creating a new entity.\n9. A new relationship MUST include a newly created KnowledgeEntity."
 };
--- a/common/migrations/20250606_234535_add_embedding_model_and_dimensions_to_system_settings.surql
+++ b/common/migrations/20250606_234535_add_embedding_model_and_dimensions_to_system_settings.surql
@@ -0,0 +1,7 @@
+DEFINE FIELD IF NOT EXISTS embedding_model ON system_settings TYPE string;
+DEFINE FIELD IF NOT EXISTS embedding_dimensions ON system_settings TYPE int;
+
+UPDATE system_settings:current SET
+	embedding_model =  "text-embedding-3-small",
+    embedding_dimensions = 1536
+WHERE embedding_model == NONE && embedding_dimensions == NONE;
--- a/common/migrations/20250627_120000_add_image_processing_settings.surql
+++ b/common/migrations/20250627_120000_add_image_processing_settings.surql
@@ -0,0 +1,7 @@
+DEFINE FIELD IF NOT EXISTS image_processing_model ON system_settings TYPE string;
+DEFINE FIELD IF NOT EXISTS image_processing_prompt ON system_settings TYPE string;
+
+UPDATE system_settings:current SET
+	image_processing_model =  "gpt-4o-mini",
+    image_processing_prompt = "Analyze this image and respond based on its primary content:\n - If the image is mainly text (document, screenshot, sign), transcribe the text verbatim.\n - If the image is mainly visual (photograph, art, landscape), provide a concise description of the scene.\n - For hybrid images (diagrams, ads), briefly describe the visual, then transcribe the text under a Text: heading.\n\n Respond directly with the analysis."
+WHERE image_processing_model == NONE && image_processing_prompt == NONE;
--- a/common/migrations/20250627_231035_remove_job_table.surql
+++ b/common/migrations/20250627_231035_remove_job_table.surql
@@ -0,0 +1 @@
+REMOVE TABLE job;
--- a/common/migrations/20250701_000000_add_voice_processing_model_to_system_settings.surql
+++ b/common/migrations/20250701_000000_add_voice_processing_model_to_system_settings.surql
@@ -0,0 +1,5 @@
+DEFINE FIELD IF NOT EXISTS voice_processing_model ON system_settings TYPE string;
+
+UPDATE system_settings:current SET
+    voice_processing_model = "whisper-1"
+WHERE voice_processing_model == NONE; 
--- a/common/migrations/20250921_120004_fix_datetime_fields.surql
+++ b/common/migrations/20250921_120004_fix_datetime_fields.surql
@@ -0,0 +1,115 @@
+-- Align timestamp fields with SurrealDB native datetime type.
+
+-- User timestamps
+DEFINE FIELD OVERWRITE created_at ON user FLEXIBLE;
+DEFINE FIELD OVERWRITE updated_at ON user FLEXIBLE;
+
+UPDATE user SET created_at = type::datetime(created_at)
+WHERE type::is::string(created_at) AND created_at != "";
+
+UPDATE user SET updated_at = type::datetime(updated_at)
+WHERE type::is::string(updated_at) AND updated_at != "";
+
+DEFINE FIELD OVERWRITE created_at ON user TYPE datetime;
+DEFINE FIELD OVERWRITE updated_at ON user TYPE datetime;
+
+-- Text content timestamps
+DEFINE FIELD OVERWRITE created_at ON text_content FLEXIBLE;
+DEFINE FIELD OVERWRITE updated_at ON text_content FLEXIBLE;
+
+UPDATE text_content SET created_at = type::datetime(created_at)
+WHERE type::is::string(created_at) AND created_at != "";
+
+UPDATE text_content SET updated_at = type::datetime(updated_at)
+WHERE type::is::string(updated_at) AND updated_at != "";
+
+DEFINE FIELD OVERWRITE created_at ON text_content TYPE datetime;
+DEFINE FIELD OVERWRITE updated_at ON text_content TYPE datetime;
+
+REBUILD INDEX text_content_created_at_idx ON text_content;
+
+-- Text chunk timestamps
+DEFINE FIELD OVERWRITE created_at ON text_chunk FLEXIBLE;
+DEFINE FIELD OVERWRITE updated_at ON text_chunk FLEXIBLE;
+
+UPDATE text_chunk SET created_at = type::datetime(created_at)
+WHERE type::is::string(created_at) AND created_at != "";
+
+UPDATE text_chunk SET updated_at = type::datetime(updated_at)
+WHERE type::is::string(updated_at) AND updated_at != "";
+
+DEFINE FIELD OVERWRITE created_at ON text_chunk TYPE datetime;
+DEFINE FIELD OVERWRITE updated_at ON text_chunk TYPE datetime;
+
+-- Knowledge entity timestamps
+DEFINE FIELD OVERWRITE created_at ON knowledge_entity FLEXIBLE;
+DEFINE FIELD OVERWRITE updated_at ON knowledge_entity FLEXIBLE;
+
+UPDATE knowledge_entity SET created_at = type::datetime(created_at)
+WHERE type::is::string(created_at) AND created_at != "";
+
+UPDATE knowledge_entity SET updated_at = type::datetime(updated_at)
+WHERE type::is::string(updated_at) AND updated_at != "";
+
+DEFINE FIELD OVERWRITE created_at ON knowledge_entity TYPE datetime;
+DEFINE FIELD OVERWRITE updated_at ON knowledge_entity TYPE datetime;
+
+REBUILD INDEX knowledge_entity_created_at_idx ON knowledge_entity;
+
+-- Conversation timestamps
+DEFINE FIELD OVERWRITE created_at ON conversation FLEXIBLE;
+DEFINE FIELD OVERWRITE updated_at ON conversation FLEXIBLE;
+
+UPDATE conversation SET created_at = type::datetime(created_at)
+WHERE type::is::string(created_at) AND created_at != "";
+
+UPDATE conversation SET updated_at = type::datetime(updated_at)
+WHERE type::is::string(updated_at) AND updated_at != "";
+
+DEFINE FIELD OVERWRITE created_at ON conversation TYPE datetime;
+DEFINE FIELD OVERWRITE updated_at ON conversation TYPE datetime;
+
+REBUILD INDEX conversation_created_at_idx ON conversation;
+
+-- Message timestamps
+DEFINE FIELD OVERWRITE created_at ON message FLEXIBLE;
+DEFINE FIELD OVERWRITE updated_at ON message FLEXIBLE;
+
+UPDATE message SET created_at = type::datetime(created_at)
+WHERE type::is::string(created_at) AND created_at != "";
+
+UPDATE message SET updated_at = type::datetime(updated_at)
+WHERE type::is::string(updated_at) AND updated_at != "";
+
+DEFINE FIELD OVERWRITE created_at ON message TYPE datetime;
+DEFINE FIELD OVERWRITE updated_at ON message TYPE datetime;
+
+REBUILD INDEX message_updated_at_idx ON message;
+
+-- Ingestion task timestamps
+DEFINE FIELD OVERWRITE created_at ON ingestion_task FLEXIBLE;
+DEFINE FIELD OVERWRITE updated_at ON ingestion_task FLEXIBLE;
+
+UPDATE ingestion_task SET created_at = type::datetime(created_at)
+WHERE type::is::string(created_at) AND created_at != "";
+
+UPDATE ingestion_task SET updated_at = type::datetime(updated_at)
+WHERE type::is::string(updated_at) AND updated_at != "";
+
+DEFINE FIELD OVERWRITE created_at ON ingestion_task TYPE datetime;
+DEFINE FIELD OVERWRITE updated_at ON ingestion_task TYPE datetime;
+
+REBUILD INDEX idx_ingestion_task_created ON ingestion_task;
+
+-- File timestamps
+DEFINE FIELD OVERWRITE created_at ON file FLEXIBLE;
+DEFINE FIELD OVERWRITE updated_at ON file FLEXIBLE;
+
+UPDATE file SET created_at = type::datetime(created_at)
+WHERE type::is::string(created_at) AND created_at != "";
+
+UPDATE file SET updated_at = type::datetime(updated_at)
+WHERE type::is::string(updated_at) AND updated_at != "";
+
+DEFINE FIELD OVERWRITE created_at ON file TYPE datetime;
+DEFINE FIELD OVERWRITE updated_at ON file TYPE datetime;
--- a/common/migrations/20251003_000001_add_fts_for_entities_and_chunks.surql
+++ b/common/migrations/20251003_000001_add_fts_for_entities_and_chunks.surql
@@ -0,0 +1,17 @@
+-- Add FTS indexes for searching name and description on entities
+
+DEFINE ANALYZER IF NOT EXISTS app_en_fts_analyzer
+    TOKENIZERS class
+    FILTERS lowercase, ascii, snowball(english);
+
+DEFINE INDEX IF NOT EXISTS knowledge_entity_fts_name_idx ON TABLE knowledge_entity
+    FIELDS name
+    SEARCH ANALYZER app_en_fts_analyzer BM25;
+
+DEFINE INDEX IF NOT EXISTS knowledge_entity_fts_description_idx ON TABLE knowledge_entity
+    FIELDS description
+    SEARCH ANALYZER app_en_fts_analyzer BM25;
+
+DEFINE INDEX IF NOT EXISTS text_chunk_fts_chunk_idx ON TABLE text_chunk
+    FIELDS chunk
+    SEARCH ANALYZER app_en_fts_analyzer BM25;
--- a/common/migrations/20251012_205900_state_machine_migration.surql
+++ b/common/migrations/20251012_205900_state_machine_migration.surql
@@ -0,0 +1,173 @@
+-- State machine migration for ingestion_task records
+
+DEFINE FIELD IF NOT EXISTS state ON TABLE ingestion_task TYPE option<string>;
+DEFINE FIELD IF NOT EXISTS attempts ON TABLE ingestion_task TYPE option<number>;
+DEFINE FIELD IF NOT EXISTS max_attempts ON TABLE ingestion_task TYPE option<number>;
+DEFINE FIELD IF NOT EXISTS scheduled_at ON TABLE ingestion_task TYPE option<datetime>;
+DEFINE FIELD IF NOT EXISTS locked_at ON TABLE ingestion_task TYPE option<datetime>;
+DEFINE FIELD IF NOT EXISTS lease_duration_secs ON TABLE ingestion_task TYPE option<number>;
+DEFINE FIELD IF NOT EXISTS worker_id ON TABLE ingestion_task TYPE option<string>;
+DEFINE FIELD IF NOT EXISTS error_code ON TABLE ingestion_task TYPE option<string>;
+DEFINE FIELD IF NOT EXISTS error_message ON TABLE ingestion_task TYPE option<string>;
+DEFINE FIELD IF NOT EXISTS last_error_at ON TABLE ingestion_task TYPE option<datetime>;
+DEFINE FIELD IF NOT EXISTS priority ON TABLE ingestion_task TYPE option<number>;
+
+REMOVE FIELD status ON TABLE ingestion_task;
+DEFINE FIELD status ON TABLE ingestion_task TYPE option<object>;
+
+DEFINE INDEX IF NOT EXISTS idx_ingestion_task_state_sched ON TABLE ingestion_task FIELDS state, scheduled_at;
+
+LET $needs_migration = (SELECT count() AS count FROM type::table('ingestion_task') WHERE state = NONE)[0].count;
+
+IF $needs_migration > 0 THEN {
+    -- Created -> Pending
+    UPDATE type::table('ingestion_task')
+    SET
+        state = "Pending",
+        attempts = 0,
+        max_attempts = 3,
+        scheduled_at = IF created_at != NONE THEN created_at ELSE time::now() END,
+        locked_at = NONE,
+        lease_duration_secs = 300,
+        worker_id = NONE,
+        error_code = NONE,
+        error_message = NONE,
+        last_error_at = NONE,
+        priority = 0
+    WHERE state = NONE
+      AND status != NONE
+      AND status.name = "Created";
+
+    -- InProgress -> Processing
+    UPDATE type::table('ingestion_task')
+    SET
+        state = "Processing",
+        attempts = IF status.attempts != NONE THEN status.attempts ELSE 1 END,
+        max_attempts = 3,
+        scheduled_at = IF status.last_attempt != NONE THEN status.last_attempt ELSE time::now() END,
+        locked_at = IF status.last_attempt != NONE THEN status.last_attempt ELSE time::now() END,
+        lease_duration_secs = 300,
+        worker_id = NONE,
+        error_code = NONE,
+        error_message = NONE,
+        last_error_at = NONE,
+        priority = 0
+    WHERE state = NONE
+      AND status != NONE
+      AND status.name = "InProgress";
+
+    -- Completed -> Succeeded
+    UPDATE type::table('ingestion_task')
+    SET
+        state = "Succeeded",
+        attempts = 1,
+        max_attempts = 3,
+        scheduled_at = IF updated_at != NONE THEN updated_at ELSE time::now() END,
+        locked_at = NONE,
+        lease_duration_secs = 300,
+        worker_id = NONE,
+        error_code = NONE,
+        error_message = NONE,
+        last_error_at = NONE,
+        priority = 0
+    WHERE state = NONE
+      AND status != NONE
+      AND status.name = "Completed";
+
+    -- Error -> DeadLetter (terminal failure)
+    UPDATE type::table('ingestion_task')
+    SET
+        state = "DeadLetter",
+        attempts = 3,
+        max_attempts = 3,
+        scheduled_at = IF updated_at != NONE THEN updated_at ELSE time::now() END,
+        locked_at = NONE,
+        lease_duration_secs = 300,
+        worker_id = NONE,
+        error_code = NONE,
+        error_message = status.message,
+        last_error_at = IF updated_at != NONE THEN updated_at ELSE time::now() END,
+        priority = 0
+    WHERE state = NONE
+      AND status != NONE
+      AND status.name = "Error";
+
+    -- Cancelled -> Cancelled
+    UPDATE type::table('ingestion_task')
+    SET
+        state = "Cancelled",
+        attempts = 0,
+        max_attempts = 3,
+        scheduled_at = IF updated_at != NONE THEN updated_at ELSE time::now() END,
+        locked_at = NONE,
+        lease_duration_secs = 300,
+        worker_id = NONE,
+        error_code = NONE,
+        error_message = NONE,
+        last_error_at = NONE,
+        priority = 0
+    WHERE state = NONE
+      AND status != NONE
+      AND status.name = "Cancelled";
+
+    -- Fallback for any remaining records missing state
+    UPDATE type::table('ingestion_task')
+    SET
+        state = "Pending",
+        attempts = 0,
+        max_attempts = 3,
+        scheduled_at = IF updated_at != NONE THEN updated_at ELSE time::now() END,
+        locked_at = NONE,
+        lease_duration_secs = 300,
+        worker_id = NONE,
+        error_code = NONE,
+        error_message = NONE,
+        last_error_at = NONE,
+        priority = 0
+    WHERE state = NONE;
+} END;
+
+-- Ensure defaults for newly added fields
+UPDATE type::table('ingestion_task')
+SET max_attempts = 3
+WHERE max_attempts = NONE;
+
+UPDATE type::table('ingestion_task')
+SET lease_duration_secs = 300
+WHERE lease_duration_secs = NONE;
+
+UPDATE type::table('ingestion_task')
+SET attempts = 0
+WHERE attempts = NONE;
+
+UPDATE type::table('ingestion_task')
+SET priority = 0
+WHERE priority = NONE;
+
+UPDATE type::table('ingestion_task')
+SET scheduled_at = IF updated_at != NONE THEN updated_at ELSE time::now() END
+WHERE scheduled_at = NONE;
+
+UPDATE type::table('ingestion_task')
+SET locked_at = NONE
+WHERE locked_at = NONE;
+
+UPDATE type::table('ingestion_task')
+SET worker_id = NONE
+WHERE worker_id != NONE AND worker_id = "";
+
+UPDATE type::table('ingestion_task')
+SET error_code = NONE
+WHERE error_code = NONE;
+
+UPDATE type::table('ingestion_task')
+SET error_message = NONE
+WHERE error_message = NONE;
+
+UPDATE type::table('ingestion_task')
+SET last_error_at = NONE
+WHERE last_error_at = NONE;
+
+UPDATE type::table('ingestion_task')
+SET status = NONE
+WHERE status != NONE;
--- a/common/migrations/20251022_120302_add_scratchpad_table.surql
+++ b/common/migrations/20251022_120302_add_scratchpad_table.surql
@@ -0,0 +1,24 @@
+-- Add scratchpad table and schema
+
+-- Define scratchpad table and schema
+DEFINE TABLE IF NOT EXISTS scratchpad SCHEMALESS;
+
+-- Standard fields from stored_object! macro
+DEFINE FIELD IF NOT EXISTS created_at ON scratchpad TYPE datetime;
+DEFINE FIELD IF NOT EXISTS updated_at ON scratchpad TYPE datetime;
+
+-- Custom fields from the Scratchpad struct
+DEFINE FIELD IF NOT EXISTS user_id ON scratchpad TYPE string;
+DEFINE FIELD IF NOT EXISTS title ON scratchpad TYPE string;
+DEFINE FIELD IF NOT EXISTS content ON scratchpad TYPE string;
+DEFINE FIELD IF NOT EXISTS last_saved_at ON scratchpad TYPE datetime;
+DEFINE FIELD IF NOT EXISTS is_dirty ON scratchpad TYPE bool DEFAULT false;
+DEFINE FIELD IF NOT EXISTS is_archived ON scratchpad TYPE bool DEFAULT false;
+DEFINE FIELD IF NOT EXISTS archived_at ON scratchpad TYPE option<datetime>;
+DEFINE FIELD IF NOT EXISTS ingested_at ON scratchpad TYPE option<datetime>;
+
+-- Indexes based on query patterns
+DEFINE INDEX IF NOT EXISTS scratchpad_user_idx ON scratchpad FIELDS user_id;
+DEFINE INDEX IF NOT EXISTS scratchpad_user_archived_idx ON scratchpad FIELDS user_id, is_archived;
+DEFINE INDEX IF NOT EXISTS scratchpad_updated_idx ON scratchpad FIELDS updated_at;
+DEFINE INDEX IF NOT EXISTS scratchpad_archived_idx ON scratchpad FIELDS archived_at;
--- a/common/migrations/definitions/20250514_142342_add_full_text_search_text_content.json
+++ b/common/migrations/definitions/20250514_142342_add_full_text_search_text_content.json
@@ -0,0 +1 @@
+{"schemas":"--- original\n+++ modified\n@@ -98,7 +98,7 @@\n DEFINE INDEX IF NOT EXISTS knowledge_entity_user_id_idx ON knowledge_entity FIELDS user_id;\n DEFINE INDEX IF NOT EXISTS knowledge_entity_source_id_idx ON knowledge_entity FIELDS source_id;\n DEFINE INDEX IF NOT EXISTS knowledge_entity_entity_type_idx ON knowledge_entity FIELDS entity_type;\n-DEFINE INDEX IF NOT EXISTS knowledge_entity_created_at_idx ON knowledge_entity FIELDS created_at; # For get_latest_knowledge_entities\n+DEFINE INDEX IF NOT EXISTS knowledge_entity_created_at_idx ON knowledge_entity FIELDS created_at;\n\n # Defines the schema for the 'message' table.\n\n@@ -157,6 +157,8 @@\n DEFINE FIELD IF NOT EXISTS require_email_verification ON system_settings TYPE bool;\n DEFINE FIELD IF NOT EXISTS query_model ON system_settings TYPE string;\n DEFINE FIELD IF NOT EXISTS processing_model ON system_settings TYPE string;\n+DEFINE FIELD IF NOT EXISTS embedding_model ON system_settings TYPE string;\n+DEFINE FIELD IF NOT EXISTS embedding_dimensions ON system_settings TYPE int;\n DEFINE FIELD IF NOT EXISTS query_system_prompt ON system_settings TYPE string;\n DEFINE FIELD IF NOT EXISTS ingestion_system_prompt ON system_settings TYPE string;\n\n","events":null}
--- a/common/migrations/definitions/20250606_234535_add_embedding_model_and_dimensions_to_system_settings.json
+++ b/common/migrations/definitions/20250606_234535_add_embedding_model_and_dimensions_to_system_settings.json
@@ -0,0 +1 @@
+{"schemas":"--- original\n+++ modified\n@@ -51,23 +51,23 @@\n\n # Defines the schema for the 'ingestion_task' table (used by IngestionTask).\n\n-DEFINE TABLE IF NOT EXISTS job SCHEMALESS;\n+DEFINE TABLE IF NOT EXISTS ingestion_task SCHEMALESS;\n\n # Standard fields\n-DEFINE FIELD IF NOT EXISTS created_at ON job TYPE string;\n-DEFINE FIELD IF NOT EXISTS updated_at ON job TYPE string;\n+DEFINE FIELD IF NOT EXISTS created_at ON ingestion_task TYPE string;\n+DEFINE FIELD IF NOT EXISTS updated_at ON ingestion_task TYPE string;\n\n # Custom fields from the IngestionTask struct\n # IngestionPayload is complex, store as object\n-DEFINE FIELD IF NOT EXISTS content ON job TYPE object;\n+DEFINE FIELD IF NOT EXISTS content ON ingestion_task TYPE object;\n # IngestionTaskStatus can hold data (InProgress), store as object\n-DEFINE FIELD IF NOT EXISTS status ON job TYPE object;\n-DEFINE FIELD IF NOT EXISTS user_id ON job TYPE string;\n+DEFINE FIELD IF NOT EXISTS status ON ingestion_task TYPE object;\n+DEFINE FIELD IF NOT EXISTS user_id ON ingestion_task TYPE string;\n\n # Indexes explicitly defined in build_indexes and useful for get_unfinished_tasks\n-DEFINE INDEX IF NOT EXISTS idx_job_status ON job FIELDS status;\n-DEFINE INDEX IF NOT EXISTS idx_job_user ON job FIELDS user_id;\n-DEFINE INDEX IF NOT EXISTS idx_job_created ON job FIELDS created_at;\n+DEFINE INDEX IF NOT EXISTS idx_ingestion_task_status ON ingestion_task FIELDS status;\n+DEFINE INDEX IF NOT EXISTS idx_ingestion_task_user ON ingestion_task FIELDS user_id;\n+DEFINE INDEX IF NOT EXISTS idx_ingestion_task_created ON ingestion_task FIELDS created_at;\n\n # Defines the schema for the 'knowledge_entity' table.\n\n","events":null}
--- a/common/migrations/definitions/20250627_120000_add_image_processing_settings.json
+++ b/common/migrations/definitions/20250627_120000_add_image_processing_settings.json
@@ -0,0 +1 @@
+{"schemas":"--- original\n+++ modified\n@@ -57,10 +57,7 @@\n DEFINE FIELD IF NOT EXISTS created_at ON ingestion_task TYPE string;\n DEFINE FIELD IF NOT EXISTS updated_at ON ingestion_task TYPE string;\n\n-# Custom fields from the IngestionTask struct\n-# IngestionPayload is complex, store as object\n DEFINE FIELD IF NOT EXISTS content ON ingestion_task TYPE object;\n-# IngestionTaskStatus can hold data (InProgress), store as object\n DEFINE FIELD IF NOT EXISTS status ON ingestion_task TYPE object;\n DEFINE FIELD IF NOT EXISTS user_id ON ingestion_task TYPE string;\n\n@@ -157,10 +154,12 @@\n DEFINE FIELD IF NOT EXISTS require_email_verification ON system_settings TYPE bool;\n DEFINE FIELD IF NOT EXISTS query_model ON system_settings TYPE string;\n DEFINE FIELD IF NOT EXISTS processing_model ON system_settings TYPE string;\n+DEFINE FIELD IF NOT EXISTS image_processing_model ON system_settings TYPE string;\n DEFINE FIELD IF NOT EXISTS embedding_model ON system_settings TYPE string;\n DEFINE FIELD IF NOT EXISTS embedding_dimensions ON system_settings TYPE int;\n DEFINE FIELD IF NOT EXISTS query_system_prompt ON system_settings TYPE string;\n DEFINE FIELD IF NOT EXISTS ingestion_system_prompt ON system_settings TYPE string;\n+DEFINE FIELD IF NOT EXISTS image_processing_prompt ON system_settings TYPE string;\n\n # Defines the schema for the 'text_chunk' table.\n\n","events":null}
--- a/common/migrations/definitions/20250701_000000_add_voice_processing_model_to_system_settings.json
+++ b/common/migrations/definitions/20250701_000000_add_voice_processing_model_to_system_settings.json
@@ -0,0 +1 @@
+{"schemas":"--- original\n+++ modified\n@@ -160,6 +160,7 @@\n DEFINE FIELD IF NOT EXISTS query_system_prompt ON system_settings TYPE string;\n DEFINE FIELD IF NOT EXISTS ingestion_system_prompt ON system_settings TYPE string;\n DEFINE FIELD IF NOT EXISTS image_processing_prompt ON system_settings TYPE string;\n+DEFINE FIELD IF NOT EXISTS voice_processing_model ON system_settings TYPE string;\n\n # Defines the schema for the 'text_chunk' table.\n\n","events":null}
--- a/common/migrations/definitions/20250921_120004_fix_datetime_fields.json
+++ b/common/migrations/definitions/20250921_120004_fix_datetime_fields.json
@@ -0,0 +1 @@
+{"schemas":"--- original\n+++ modified\n@@ -18,8 +18,8 @@\n DEFINE TABLE IF NOT EXISTS conversation SCHEMALESS;\n\n # Standard fields\n-DEFINE FIELD IF NOT EXISTS created_at ON conversation TYPE string;\n-DEFINE FIELD IF NOT EXISTS updated_at ON conversation TYPE string;\n+DEFINE FIELD IF NOT EXISTS created_at ON conversation TYPE datetime;\n+DEFINE FIELD IF NOT EXISTS updated_at ON conversation TYPE datetime;\n\n # Custom fields from the Conversation struct\n DEFINE FIELD IF NOT EXISTS user_id ON conversation TYPE string;\n@@ -34,8 +34,8 @@\n DEFINE TABLE IF NOT EXISTS file SCHEMALESS;\n\n # Standard fields\n-DEFINE FIELD IF NOT EXISTS created_at ON file TYPE string;\n-DEFINE FIELD IF NOT EXISTS updated_at ON file TYPE string;\n+DEFINE FIELD IF NOT EXISTS created_at ON file TYPE datetime;\n+DEFINE FIELD IF NOT EXISTS updated_at ON file TYPE datetime;\n\n # Custom fields from the FileInfo struct\n DEFINE FIELD IF NOT EXISTS sha256 ON file TYPE string;\n@@ -54,8 +54,8 @@\n DEFINE TABLE IF NOT EXISTS ingestion_task SCHEMALESS;\n\n # Standard fields\n-DEFINE FIELD IF NOT EXISTS created_at ON ingestion_task TYPE string;\n-DEFINE FIELD IF NOT EXISTS updated_at ON ingestion_task TYPE string;\n+DEFINE FIELD IF NOT EXISTS created_at ON ingestion_task TYPE datetime;\n+DEFINE FIELD IF NOT EXISTS updated_at ON ingestion_task TYPE datetime;\n\n DEFINE FIELD IF NOT EXISTS content ON ingestion_task TYPE object;\n DEFINE FIELD IF NOT EXISTS status ON ingestion_task TYPE object;\n@@ -71,8 +71,8 @@\n DEFINE TABLE IF NOT EXISTS knowledge_entity SCHEMALESS;\n\n # Standard fields\n-DEFINE FIELD IF NOT EXISTS created_at ON knowledge_entity TYPE string;\n-DEFINE FIELD IF NOT EXISTS updated_at ON knowledge_entity TYPE string;\n+DEFINE FIELD IF NOT EXISTS created_at ON knowledge_entity TYPE datetime;\n+DEFINE FIELD IF NOT EXISTS updated_at ON knowledge_entity TYPE datetime;\n\n # Custom fields from the KnowledgeEntity struct\n DEFINE FIELD IF NOT EXISTS source_id ON knowledge_entity TYPE string;\n@@ -102,8 +102,8 @@\n DEFINE TABLE IF NOT EXISTS message SCHEMALESS;\n\n # Standard fields\n-DEFINE FIELD IF NOT EXISTS created_at ON message TYPE string;\n-DEFINE FIELD IF NOT EXISTS updated_at ON message TYPE string;\n+DEFINE FIELD IF NOT EXISTS created_at ON message TYPE datetime;\n+DEFINE FIELD IF NOT EXISTS updated_at ON message TYPE datetime;\n\n # Custom fields from the Message struct\n DEFINE FIELD IF NOT EXISTS conversation_id ON message TYPE string;\n@@ -167,8 +167,8 @@\n DEFINE TABLE IF NOT EXISTS text_chunk SCHEMALESS;\n\n # Standard fields\n-DEFINE FIELD IF NOT EXISTS created_at ON text_chunk TYPE string;\n-DEFINE FIELD IF NOT EXISTS updated_at ON text_chunk TYPE string;\n+DEFINE FIELD IF NOT EXISTS created_at ON text_chunk TYPE datetime;\n+DEFINE FIELD IF NOT EXISTS updated_at ON text_chunk TYPE datetime;\n\n # Custom fields from the TextChunk struct\n DEFINE FIELD IF NOT EXISTS source_id ON text_chunk TYPE string;\n@@ -191,8 +191,8 @@\n DEFINE TABLE IF NOT EXISTS text_content SCHEMALESS;\n\n # Standard fields\n-DEFINE FIELD IF NOT EXISTS created_at ON text_content TYPE string;\n-DEFINE FIELD IF NOT EXISTS updated_at ON text_content TYPE string;\n+DEFINE FIELD IF NOT EXISTS created_at ON text_content TYPE datetime;\n+DEFINE FIELD IF NOT EXISTS updated_at ON text_content TYPE datetime;\n\n # Custom fields from the TextContent struct\n DEFINE FIELD IF NOT EXISTS text ON text_content TYPE string;\n@@ -215,8 +215,8 @@\n DEFINE TABLE IF NOT EXISTS user SCHEMALESS;\n\n # Standard fields\n-DEFINE FIELD IF NOT EXISTS created_at ON user TYPE string;\n-DEFINE FIELD IF NOT EXISTS updated_at ON user TYPE string;\n+DEFINE FIELD IF NOT EXISTS created_at ON user TYPE datetime;\n+DEFINE FIELD IF NOT EXISTS updated_at ON user TYPE datetime;\n\n # Custom fields from the User struct\n DEFINE FIELD IF NOT EXISTS email ON user TYPE string;\n","events":null}
--- a/common/migrations/definitions/20251022_120302_add_scratchpad_table.json
+++ b/common/migrations/definitions/20251022_120302_add_scratchpad_table.json
@@ -0,0 +1 @@
+{"schemas":"--- original\n+++ modified\n@@ -137,6 +137,30 @@\n DEFINE INDEX IF NOT EXISTS relates_to_metadata_source_id_idx ON relates_to FIELDS metadata.source_id;\n DEFINE INDEX IF NOT EXISTS relates_to_metadata_user_id_idx ON relates_to FIELDS metadata.user_id;\n\n+# Defines the schema for the 'scratchpad' table.\n+\n+DEFINE TABLE IF NOT EXISTS scratchpad SCHEMALESS;\n+\n+# Standard fields from stored_object! macro\n+DEFINE FIELD IF NOT EXISTS created_at ON scratchpad TYPE datetime;\n+DEFINE FIELD IF NOT EXISTS updated_at ON scratchpad TYPE datetime;\n+\n+# Custom fields from the Scratchpad struct\n+DEFINE FIELD IF NOT EXISTS user_id ON scratchpad TYPE string;\n+DEFINE FIELD IF NOT EXISTS title ON scratchpad TYPE string;\n+DEFINE FIELD IF NOT EXISTS content ON scratchpad TYPE string;\n+DEFINE FIELD IF NOT EXISTS last_saved_at ON scratchpad TYPE datetime;\n+DEFINE FIELD IF NOT EXISTS is_dirty ON scratchpad TYPE bool DEFAULT false;\n+DEFINE FIELD IF NOT EXISTS is_archived ON scratchpad TYPE bool DEFAULT false;\n+DEFINE FIELD IF NOT EXISTS archived_at ON scratchpad TYPE option<datetime>;\n+DEFINE FIELD IF NOT EXISTS ingested_at ON scratchpad TYPE option<datetime>;\n+\n+# Indexes based on query patterns\n+DEFINE INDEX IF NOT EXISTS scratchpad_user_idx ON scratchpad FIELDS user_id;\n+DEFINE INDEX IF NOT EXISTS scratchpad_user_archived_idx ON scratchpad FIELDS user_id, is_archived;\n+DEFINE INDEX IF NOT EXISTS scratchpad_updated_idx ON scratchpad FIELDS updated_at;\n+DEFINE INDEX IF NOT EXISTS scratchpad_archived_idx ON scratchpad FIELDS archived_at;\n+\n DEFINE TABLE OVERWRITE script_migration SCHEMAFULL\n     PERMISSIONS\n         FOR select FULL\n","events":null}
--- a/common/schemas/conversation.surql
+++ b/common/schemas/conversation.surql
@@ -3,8 +3,8 @@
 DEFINE TABLE IF NOT EXISTS conversation SCHEMALESS;

 # Standard fields
-DEFINE FIELD IF NOT EXISTS created_at ON conversation TYPE string;
-DEFINE FIELD IF NOT EXISTS updated_at ON conversation TYPE string;
+DEFINE FIELD IF NOT EXISTS created_at ON conversation TYPE datetime;
+DEFINE FIELD IF NOT EXISTS updated_at ON conversation TYPE datetime;

 # Custom fields from the Conversation struct
 DEFINE FIELD IF NOT EXISTS user_id ON conversation TYPE string;
--- a/common/schemas/file.surql
+++ b/common/schemas/file.surql
@@ -3,8 +3,8 @@
 DEFINE TABLE IF NOT EXISTS file SCHEMALESS;

 # Standard fields
-DEFINE FIELD IF NOT EXISTS created_at ON file TYPE string;
-DEFINE FIELD IF NOT EXISTS updated_at ON file TYPE string;
+DEFINE FIELD IF NOT EXISTS created_at ON file TYPE datetime;
+DEFINE FIELD IF NOT EXISTS updated_at ON file TYPE datetime;

 # Custom fields from the FileInfo struct
 DEFINE FIELD IF NOT EXISTS sha256 ON file TYPE string;
--- a/common/schemas/ingestion_task.surql
+++ b/common/schemas/ingestion_task.surql
@@ -1,19 +1,16 @@
 # Defines the schema for the 'ingestion_task' table (used by IngestionTask).

-DEFINE TABLE IF NOT EXISTS job SCHEMALESS;
+DEFINE TABLE IF NOT EXISTS ingestion_task SCHEMALESS;

 # Standard fields
-DEFINE FIELD IF NOT EXISTS created_at ON job TYPE string;
-DEFINE FIELD IF NOT EXISTS updated_at ON job TYPE string;
+DEFINE FIELD IF NOT EXISTS created_at ON ingestion_task TYPE datetime;
+DEFINE FIELD IF NOT EXISTS updated_at ON ingestion_task TYPE datetime;

-# Custom fields from the IngestionTask struct
-# IngestionPayload is complex, store as object
-DEFINE FIELD IF NOT EXISTS content ON job TYPE object;
-# IngestionTaskStatus can hold data (InProgress), store as object
-DEFINE FIELD IF NOT EXISTS status ON job TYPE object;
-DEFINE FIELD IF NOT EXISTS user_id ON job TYPE string;
+DEFINE FIELD IF NOT EXISTS content ON ingestion_task TYPE object;
+DEFINE FIELD IF NOT EXISTS status ON ingestion_task TYPE object;
+DEFINE FIELD IF NOT EXISTS user_id ON ingestion_task TYPE string;

 # Indexes explicitly defined in build_indexes and useful for get_unfinished_tasks
-DEFINE INDEX IF NOT EXISTS idx_job_status ON job FIELDS status;
-DEFINE INDEX IF NOT EXISTS idx_job_user ON job FIELDS user_id;
-DEFINE INDEX IF NOT EXISTS idx_job_created ON job FIELDS created_at;
+DEFINE INDEX IF NOT EXISTS idx_ingestion_task_status ON ingestion_task FIELDS status;
+DEFINE INDEX IF NOT EXISTS idx_ingestion_task_user ON ingestion_task FIELDS user_id;
+DEFINE INDEX IF NOT EXISTS idx_ingestion_task_created ON ingestion_task FIELDS created_at;
--- a/common/schemas/knowledge_entity.surql
+++ b/common/schemas/knowledge_entity.surql
@@ -3,8 +3,8 @@
 DEFINE TABLE IF NOT EXISTS knowledge_entity SCHEMALESS;

 # Standard fields
-DEFINE FIELD IF NOT EXISTS created_at ON knowledge_entity TYPE string;
-DEFINE FIELD IF NOT EXISTS updated_at ON knowledge_entity TYPE string;
+DEFINE FIELD IF NOT EXISTS created_at ON knowledge_entity TYPE datetime;
+DEFINE FIELD IF NOT EXISTS updated_at ON knowledge_entity TYPE datetime;

 # Custom fields from the KnowledgeEntity struct
 DEFINE FIELD IF NOT EXISTS source_id ON knowledge_entity TYPE string;
@@ -27,4 +27,4 @@ DEFINE INDEX IF NOT EXISTS idx_embedding_entities ON knowledge_entity FIELDS emb
 DEFINE INDEX IF NOT EXISTS knowledge_entity_user_id_idx ON knowledge_entity FIELDS user_id;
 DEFINE INDEX IF NOT EXISTS knowledge_entity_source_id_idx ON knowledge_entity FIELDS source_id;
 DEFINE INDEX IF NOT EXISTS knowledge_entity_entity_type_idx ON knowledge_entity FIELDS entity_type;
-DEFINE INDEX IF NOT EXISTS knowledge_entity_created_at_idx ON knowledge_entity FIELDS created_at; # For get_latest_knowledge_entities
+DEFINE INDEX IF NOT EXISTS knowledge_entity_created_at_idx ON knowledge_entity FIELDS created_at;
--- a/common/schemas/message.surql
+++ b/common/schemas/message.surql
@@ -3,8 +3,8 @@
 DEFINE TABLE IF NOT EXISTS message SCHEMALESS;

 # Standard fields
-DEFINE FIELD IF NOT EXISTS created_at ON message TYPE string;
-DEFINE FIELD IF NOT EXISTS updated_at ON message TYPE string;
+DEFINE FIELD IF NOT EXISTS created_at ON message TYPE datetime;
+DEFINE FIELD IF NOT EXISTS updated_at ON message TYPE datetime;

 # Custom fields from the Message struct
 DEFINE FIELD IF NOT EXISTS conversation_id ON message TYPE string;
--- a/common/schemas/scratchpad.surql
+++ b/common/schemas/scratchpad.surql
@@ -0,0 +1,23 @@
+# Defines the schema for the 'scratchpad' table.
+
+DEFINE TABLE IF NOT EXISTS scratchpad SCHEMALESS;
+
+# Standard fields from stored_object! macro
+DEFINE FIELD IF NOT EXISTS created_at ON scratchpad TYPE datetime;
+DEFINE FIELD IF NOT EXISTS updated_at ON scratchpad TYPE datetime;
+
+# Custom fields from the Scratchpad struct
+DEFINE FIELD IF NOT EXISTS user_id ON scratchpad TYPE string;
+DEFINE FIELD IF NOT EXISTS title ON scratchpad TYPE string;
+DEFINE FIELD IF NOT EXISTS content ON scratchpad TYPE string;
+DEFINE FIELD IF NOT EXISTS last_saved_at ON scratchpad TYPE datetime;
+DEFINE FIELD IF NOT EXISTS is_dirty ON scratchpad TYPE bool DEFAULT false;
+DEFINE FIELD IF NOT EXISTS is_archived ON scratchpad TYPE bool DEFAULT false;
+DEFINE FIELD IF NOT EXISTS archived_at ON scratchpad TYPE option<datetime>;
+DEFINE FIELD IF NOT EXISTS ingested_at ON scratchpad TYPE option<datetime>;
+
+# Indexes based on query patterns
+DEFINE INDEX IF NOT EXISTS scratchpad_user_idx ON scratchpad FIELDS user_id;
+DEFINE INDEX IF NOT EXISTS scratchpad_user_archived_idx ON scratchpad FIELDS user_id, is_archived;
+DEFINE INDEX IF NOT EXISTS scratchpad_updated_idx ON scratchpad FIELDS updated_at;
+DEFINE INDEX IF NOT EXISTS scratchpad_archived_idx ON scratchpad FIELDS archived_at;
--- a/common/schemas/system_settings.surql
+++ b/common/schemas/system_settings.surql
@@ -7,5 +7,10 @@ DEFINE FIELD IF NOT EXISTS registrations_enabled ON system_settings TYPE bool;
 DEFINE FIELD IF NOT EXISTS require_email_verification ON system_settings TYPE bool;
 DEFINE FIELD IF NOT EXISTS query_model ON system_settings TYPE string;
 DEFINE FIELD IF NOT EXISTS processing_model ON system_settings TYPE string;
+DEFINE FIELD IF NOT EXISTS image_processing_model ON system_settings TYPE string;
+DEFINE FIELD IF NOT EXISTS embedding_model ON system_settings TYPE string;
+DEFINE FIELD IF NOT EXISTS embedding_dimensions ON system_settings TYPE int;
 DEFINE FIELD IF NOT EXISTS query_system_prompt ON system_settings TYPE string;
 DEFINE FIELD IF NOT EXISTS ingestion_system_prompt ON system_settings TYPE string;
+DEFINE FIELD IF NOT EXISTS image_processing_prompt ON system_settings TYPE string;
+DEFINE FIELD IF NOT EXISTS voice_processing_model ON system_settings TYPE string;
--- a/common/schemas/text_chunk.surql
+++ b/common/schemas/text_chunk.surql
@@ -3,8 +3,8 @@
 DEFINE TABLE IF NOT EXISTS text_chunk SCHEMALESS;

 # Standard fields
-DEFINE FIELD IF NOT EXISTS created_at ON text_chunk TYPE string;
-DEFINE FIELD IF NOT EXISTS updated_at ON text_chunk TYPE string;
+DEFINE FIELD IF NOT EXISTS created_at ON text_chunk TYPE datetime;
+DEFINE FIELD IF NOT EXISTS updated_at ON text_chunk TYPE datetime;

 # Custom fields from the TextChunk struct
 DEFINE FIELD IF NOT EXISTS source_id ON text_chunk TYPE string;
--- a/common/schemas/text_content.surql
+++ b/common/schemas/text_content.surql
@@ -3,8 +3,8 @@
 DEFINE TABLE IF NOT EXISTS text_content SCHEMALESS;

 # Standard fields
-DEFINE FIELD IF NOT EXISTS created_at ON text_content TYPE string;
-DEFINE FIELD IF NOT EXISTS updated_at ON text_content TYPE string;
+DEFINE FIELD IF NOT EXISTS created_at ON text_content TYPE datetime;
+DEFINE FIELD IF NOT EXISTS updated_at ON text_content TYPE datetime;

 # Custom fields from the TextContent struct
 DEFINE FIELD IF NOT EXISTS text ON text_content TYPE string;
--- a/common/schemas/user.surql
+++ b/common/schemas/user.surql
@@ -4,8 +4,8 @@
 DEFINE TABLE IF NOT EXISTS user SCHEMALESS;

 # Standard fields
-DEFINE FIELD IF NOT EXISTS created_at ON user TYPE string;
-DEFINE FIELD IF NOT EXISTS updated_at ON user TYPE string;
+DEFINE FIELD IF NOT EXISTS created_at ON user TYPE datetime;
+DEFINE FIELD IF NOT EXISTS updated_at ON user TYPE datetime;

 # Custom fields from the User struct
 DEFINE FIELD IF NOT EXISTS email ON user TYPE string;
--- a/common/src/error.rs
+++ b/common/src/error.rs
@@ -29,8 +29,8 @@ pub enum AppError {
    Io(#[from] std::io::Error),
    #[error("Reqwest error: {0}")]
    Reqwest(#[from] reqwest::Error),
-    #[error("Tiktoken error: {0}")]
-    Tiktoken(#[from] anyhow::Error),
+    #[error("Anyhow error: {0}")]
+    Anyhow(#[from] anyhow::Error),
    #[error("Ingestion Processing error: {0}")]
    Processing(String),
    #[error("DOM smoothie error: {0}")]
--- a/common/src/storage/db.rs
+++ b/common/src/storage/db.rs
@@ -11,6 +11,7 @@ use surrealdb::{
    Error, Notification, Surreal,
 };
 use surrealdb_migrations::MigrationRunner;
+use tracing::debug;

 static MIGRATIONS_DIR: Dir<'_> = include_dir!("$CARGO_MANIFEST_DIR/");

@@ -50,6 +51,7 @@ impl SurrealDbClient {
    pub async fn create_session_store(
        &self,
    ) -> Result<SessionStore<SessionSurrealPool<Any>>, SessionError> {
+        debug!("Creating session store");
        SessionStore::new(
            Some(self.client.clone().into()),
            SessionConfig::default()
@@ -65,6 +67,7 @@ impl SurrealDbClient {
    /// the database and selecting the appropriate namespace and database, but before
    /// the application starts performing operations that rely on the schema.
    pub async fn apply_migrations(&self) -> Result<(), AppError> {
+        debug!("Applying migrations");
        MigrationRunner::new(&self.client)
            .load_files(&MIGRATIONS_DIR)
            .up()
@@ -76,15 +79,19 @@ impl SurrealDbClient {

    /// Operation to rebuild indexes
    pub async fn rebuild_indexes(&self) -> Result<(), Error> {
-        self.client
-            .query("REBUILD INDEX IF EXISTS idx_embedding_chunks ON text_chunk")
-            .await?;
-        self.client
-            .query("REBUILD INDEX IF EXISTS idx_embedding_entities ON knowledge_entity")
-            .await?;
-        self.client
-            .query("REBUILD INDEX IF EXISTS text_content_fts_idx ON text_content")
-            .await?;
+        debug!("Rebuilding indexes");
+        let rebuild_sql = r#"
+            BEGIN TRANSACTION;
+            REBUILD INDEX IF EXISTS idx_embedding_chunks ON text_chunk;
+            REBUILD INDEX IF EXISTS idx_embedding_entities ON knowledge_entity;
+            REBUILD INDEX IF EXISTS text_content_fts_idx ON text_content;
+            REBUILD INDEX IF EXISTS knowledge_entity_fts_name_idx ON knowledge_entity;
+            REBUILD INDEX IF EXISTS knowledge_entity_fts_description_idx ON knowledge_entity;
+            REBUILD INDEX IF EXISTS text_chunk_fts_chunk_idx ON text_chunk;
+            COMMIT TRANSACTION;
+        "#;
+
+        self.client.query(rebuild_sql).await?;
        Ok(())
    }

--- a/common/src/storage/mod.rs
+++ b/common/src/storage/mod.rs
@@ -1,2 +1,3 @@
 pub mod db;
+pub mod store;
 pub mod types;
--- a/common/src/storage/store.rs
+++ b/common/src/storage/store.rs
@@ -0,0 +1,284 @@
+use std::path::{Path, PathBuf};
+use std::sync::Arc;
+
+use anyhow::{anyhow, Result as AnyResult};
+use bytes::Bytes;
+use futures::stream::BoxStream;
+use futures::{StreamExt, TryStreamExt};
+use object_store::local::LocalFileSystem;
+use object_store::{path::Path as ObjPath, ObjectStore};
+
+use crate::utils::config::{AppConfig, StorageKind};
+
+pub type DynStore = Arc<dyn ObjectStore>;
+
+/// Build an object store instance anchored at the given filesystem `prefix`.
+///
+/// - For the `Local` backend, `prefix` is the absolute directory on disk that
+///   serves as the root for all object paths passed to the store.
+/// - `prefix` must already exist; this function will create it if missing.
+///
+/// Example (Local):
+/// - prefix: `/var/data`
+/// - object location: `user/uuid/file.txt`
+/// - absolute path: `/var/data/user/uuid/file.txt`
+pub async fn build_store(prefix: &Path, cfg: &AppConfig) -> object_store::Result<DynStore> {
+    match cfg.storage {
+        StorageKind::Local => {
+            if !prefix.exists() {
+                tokio::fs::create_dir_all(prefix).await.map_err(|e| {
+                    object_store::Error::Generic {
+                        store: "LocalFileSystem",
+                        source: e.into(),
+                    }
+                })?;
+            }
+            let store = LocalFileSystem::new_with_prefix(prefix)?;
+            Ok(Arc::new(store))
+        }
+    }
+}
+
+/// Resolve the absolute base directory used for local storage from config.
+///
+/// If `data_dir` is relative, it is resolved against the current working directory.
+pub fn resolve_base_dir(cfg: &AppConfig) -> PathBuf {
+    if cfg.data_dir.starts_with('/') {
+        PathBuf::from(&cfg.data_dir)
+    } else {
+        std::env::current_dir()
+            .unwrap_or_else(|_| PathBuf::from("."))
+            .join(&cfg.data_dir)
+    }
+}
+
+/// Build an object store rooted at the configured data directory.
+///
+/// This is the recommended way to obtain a store for logical object operations
+/// such as `put_bytes_at`, `get_bytes_at`, and `delete_prefix_at`.
+pub async fn build_store_root(cfg: &AppConfig) -> object_store::Result<DynStore> {
+    let base = resolve_base_dir(cfg);
+    build_store(&base, cfg).await
+}
+
+/// Write bytes to `file_name` within a filesystem `prefix` using the configured store.
+///
+/// Prefer [`put_bytes_at`] for location-based writes that do not need to compute
+/// a separate filesystem prefix.
+pub async fn put_bytes(
+    prefix: &Path,
+    file_name: &str,
+    data: Bytes,
+    cfg: &AppConfig,
+) -> object_store::Result<()> {
+    let store = build_store(prefix, cfg).await?;
+    let payload = object_store::PutPayload::from_bytes(data);
+    store.put(&ObjPath::from(file_name), payload).await?;
+    Ok(())
+}
+
+/// Write bytes to the provided logical object `location`, e.g. `"user/uuid/file"`.
+///
+/// The store root is taken from `AppConfig::data_dir` for the local backend.
+/// This performs an atomic write as guaranteed by `object_store`.
+pub async fn put_bytes_at(
+    location: &str,
+    data: Bytes,
+    cfg: &AppConfig,
+) -> object_store::Result<()> {
+    let store = build_store_root(cfg).await?;
+    let payload = object_store::PutPayload::from_bytes(data);
+    store.put(&ObjPath::from(location), payload).await?;
+    Ok(())
+}
+
+/// Read bytes from `file_name` within a filesystem `prefix` using the configured store.
+///
+/// Prefer [`get_bytes_at`] for location-based reads.
+pub async fn get_bytes(
+    prefix: &Path,
+    file_name: &str,
+    cfg: &AppConfig,
+) -> object_store::Result<Bytes> {
+    let store = build_store(prefix, cfg).await?;
+    let r = store.get(&ObjPath::from(file_name)).await?;
+    let b = r.bytes().await?;
+    Ok(b)
+}
+
+/// Read bytes from the provided logical object `location`.
+///
+/// Returns the full contents buffered in memory.
+pub async fn get_bytes_at(location: &str, cfg: &AppConfig) -> object_store::Result<Bytes> {
+    let store = build_store_root(cfg).await?;
+    let r = store.get(&ObjPath::from(location)).await?;
+    r.bytes().await
+}
+
+/// Get a streaming body for the provided logical object `location`.
+///
+/// Returns a fallible `BoxStream` of `Bytes`, suitable for use with
+/// `axum::body::Body::from_stream` to stream responses without buffering.
+pub async fn get_stream_at(
+    location: &str,
+    cfg: &AppConfig,
+) -> object_store::Result<BoxStream<'static, object_store::Result<Bytes>>> {
+    let store = build_store_root(cfg).await?;
+    let r = store.get(&ObjPath::from(location)).await?;
+    Ok(r.into_stream())
+}
+
+/// Delete all objects below the provided filesystem `prefix`.
+///
+/// This is a low-level variant for when a dedicated on-disk prefix is used for a
+/// particular object grouping. Prefer [`delete_prefix_at`] for location-based stores.
+pub async fn delete_prefix(prefix: &Path, cfg: &AppConfig) -> object_store::Result<()> {
+    let store = build_store(prefix, cfg).await?;
+    // list everything and delete
+    let locations = store.list(None).map_ok(|m| m.location).boxed();
+    store
+        .delete_stream(locations)
+        .try_collect::<Vec<_>>()
+        .await?;
+    // Best effort remove the directory itself
+    if tokio::fs::try_exists(prefix).await.unwrap_or(false) {
+        let _ = tokio::fs::remove_dir_all(prefix).await;
+    }
+    Ok(())
+}
+
+/// Delete all objects below the provided logical object `prefix`, e.g. `"user/uuid/"`.
+///
+/// After deleting, attempts a best-effort cleanup of the now-empty directory on disk
+/// when using the local backend.
+pub async fn delete_prefix_at(prefix: &str, cfg: &AppConfig) -> object_store::Result<()> {
+    let store = build_store_root(cfg).await?;
+    let prefix_path = ObjPath::from(prefix);
+    let locations = store
+        .list(Some(&prefix_path))
+        .map_ok(|m| m.location)
+        .boxed();
+    store
+        .delete_stream(locations)
+        .try_collect::<Vec<_>>()
+        .await?;
+    // Best effort remove empty directory on disk for local storage
+    let base_dir = resolve_base_dir(cfg).join(prefix);
+    if tokio::fs::try_exists(&base_dir).await.unwrap_or(false) {
+        let _ = tokio::fs::remove_dir_all(&base_dir).await;
+    }
+    Ok(())
+}
+
+/// Split an absolute filesystem path into `(parent_dir, file_name)`.
+pub fn split_abs_path(path: &str) -> AnyResult<(PathBuf, String)> {
+    let pb = PathBuf::from(path);
+    let parent = pb
+        .parent()
+        .ok_or_else(|| anyhow!("Path has no parent: {path}"))?
+        .to_path_buf();
+    let file = pb
+        .file_name()
+        .ok_or_else(|| anyhow!("Path has no file name: {path}"))?
+        .to_string_lossy()
+        .to_string();
+    Ok((parent, file))
+}
+
+/// Split a logical object location `"a/b/c"` into `("a/b", "c")`.
+pub fn split_object_path(path: &str) -> AnyResult<(String, String)> {
+    if let Some((p, f)) = path.rsplit_once('/') {
+        return Ok((p.to_string(), f.to_string()));
+    }
+    Err(anyhow!("Object path has no separator: {path}"))
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::utils::config::{PdfIngestMode::LlmFirst, StorageKind};
+    use bytes::Bytes;
+    use futures::TryStreamExt;
+    use uuid::Uuid;
+
+    fn test_config(root: &str) -> AppConfig {
+        AppConfig {
+            openai_api_key: "test".into(),
+            surrealdb_address: "test".into(),
+            surrealdb_username: "test".into(),
+            surrealdb_password: "test".into(),
+            surrealdb_namespace: "test".into(),
+            surrealdb_database: "test".into(),
+            data_dir: root.into(),
+            http_port: 0,
+            openai_base_url: "..".into(),
+            storage: StorageKind::Local,
+            pdf_ingest_mode: LlmFirst,
+        }
+    }
+
+    #[tokio::test]
+    async fn test_build_store_root_creates_base() {
+        let base = format!("/tmp/minne_store_test_{}", Uuid::new_v4());
+        let cfg = test_config(&base);
+        let _ = build_store_root(&cfg).await.expect("build store root");
+        assert!(tokio::fs::try_exists(&base).await.unwrap_or(false));
+        let _ = tokio::fs::remove_dir_all(&base).await;
+    }
+
+    #[tokio::test]
+    async fn test_put_get_bytes_at_and_delete_prefix_at() {
+        let base = format!("/tmp/minne_store_test_{}", Uuid::new_v4());
+        let cfg = test_config(&base);
+
+        let location_prefix = format!("{}/{}", "user1", Uuid::new_v4());
+        let file_name = "file.txt";
+        let location = format!("{}/{}", &location_prefix, file_name);
+        let payload = Bytes::from_static(b"hello world");
+
+        put_bytes_at(&location, payload.clone(), &cfg)
+            .await
+            .expect("put");
+        let got = get_bytes_at(&location, &cfg).await.expect("get");
+        assert_eq!(got.as_ref(), payload.as_ref());
+
+        // Delete the whole prefix and ensure retrieval fails
+        delete_prefix_at(&location_prefix, &cfg)
+            .await
+            .expect("delete prefix");
+        assert!(get_bytes_at(&location, &cfg).await.is_err());
+
+        let _ = tokio::fs::remove_dir_all(&base).await;
+    }
+
+    #[tokio::test]
+    async fn test_get_stream_at() {
+        let base = format!("/tmp/minne_store_test_{}", Uuid::new_v4());
+        let cfg = test_config(&base);
+
+        let location = format!("{}/{}/stream.bin", "user2", Uuid::new_v4());
+        let content = vec![7u8; 32 * 1024]; // 32KB payload
+
+        put_bytes_at(&location, Bytes::from(content.clone()), &cfg)
+            .await
+            .expect("put");
+
+        let stream = get_stream_at(&location, &cfg).await.expect("stream");
+        let combined: Vec<u8> = stream
+            .map_ok(|chunk| chunk.to_vec())
+            .try_fold(Vec::new(), |mut acc, mut chunk| async move {
+                acc.append(&mut chunk);
+                Ok(acc)
+            })
+            .await
+            .expect("collect");
+
+        assert_eq!(combined, content);
+
+        delete_prefix_at(&split_object_path(&location).unwrap().0, &cfg)
+            .await
+            .ok();
+
+        let _ = tokio::fs::remove_dir_all(&base).await;
+    }
+}
--- a/common/src/storage/types/conversation.rs
+++ b/common/src/storage/types/conversation.rs
@@ -67,7 +67,10 @@ impl Conversation {
        let _updated: Option<Self> = db
            .update((Self::table_name(), id))
            .patch(PatchOp::replace("/title", new_title.to_string()))
-            .patch(PatchOp::replace("/updated_at", Utc::now()))
+            .patch(PatchOp::replace(
+                "/updated_at",
+                surrealdb::Datetime::from(Utc::now()),
+            ))
            .await?;

        Ok(())
--- a/common/src/storage/types/file_info.rs
+++ b/common/src/storage/types/file_info.rs
@@ -1,18 +1,21 @@
 use axum_typed_multipart::FieldData;
 use mime_guess::from_path;
+use object_store::Error as ObjectStoreError;
 use sha2::{Digest, Sha256};
 use std::{
    io::{BufReader, Read},
-    path::{Path, PathBuf},
+    path::Path,
 };
 use tempfile::NamedTempFile;
 use thiserror::Error;
-use tokio::fs::remove_dir_all;
 use tracing::info;
 use uuid::Uuid;

 use crate::{
-    error::AppError, storage::db::SurrealDbClient, stored_object, utils::config::AppConfig,
+    error::AppError,
+    storage::{db::SurrealDbClient, store},
+    stored_object,
+    utils::config::AppConfig,
 };

 #[derive(Error, Debug)]
@@ -34,6 +37,9 @@ pub enum FileError {

    #[error("File name missing in metadata")]
    MissingFileName,
+
+    #[error("Object store error: {0}")]
+    ObjectStore(#[from] ObjectStoreError),
 }

 stored_object!(FileInfo, "file", {
@@ -51,7 +57,6 @@ impl FileInfo {
        user_id: &str,
        config: &AppConfig,
    ) -> Result<Self, FileError> {
-        info!("Data_dir: {:?}", config);
        let file = field_data.contents;
        let file_name = field_data
            .metadata
@@ -83,10 +88,7 @@ impl FileInfo {
            updated_at: now,
            file_name,
            sha256,
-            path: Self::persist_file(&uuid, file, &sanitized_file_name, user_id, config)
-                .await?
-                .to_string_lossy()
-                .into(),
+            path: Self::persist_file(&uuid, file, &sanitized_file_name, user_id, config).await?,
            mime_type: Self::guess_mime_type(Path::new(&sanitized_file_name)),
            user_id: user_id.to_string(),
        };
@@ -165,7 +167,7 @@ impl FileInfo {
        }
    }

-    /// Persists the file to the filesystem under `{data_dir}/{user_id}/{uuid}/{file_name}`.
+    /// Persists the file under the logical location `{user_id}/{uuid}/{file_name}`.
    ///
    /// # Arguments
    /// * `uuid` - The UUID of the file.
@@ -175,43 +177,24 @@ impl FileInfo {
    /// * `config` - Application configuration containing data directory path
    ///
    /// # Returns
-    /// * `Result<PathBuf, FileError>` - The persisted file path or an error.
+    /// * `Result<String, FileError>` - The logical object location or an error.
    async fn persist_file(
        uuid: &Uuid,
        file: NamedTempFile,
        file_name: &str,
        user_id: &str,
        config: &AppConfig,
-    ) -> Result<PathBuf, FileError> {
-        info!("Data dir: {:?}", config.data_dir);
-        // Convert relative path to absolute path
-        let base_dir = if config.data_dir.starts_with('/') {
-            Path::new(&config.data_dir).to_path_buf()
-        } else {
-            std::env::current_dir()
-                .map_err(FileError::Io)?
-                .join(&config.data_dir)
-        };
+    ) -> Result<String, FileError> {
+        // Logical object location relative to the store root
+        let location = format!("{}/{}/{}", user_id, uuid, file_name);
+        info!("Persisting to object location: {}", location);

-        let user_dir = base_dir.join(user_id); // Create the user directory
-        let uuid_dir = user_dir.join(uuid.to_string()); // Create the UUID directory under the user directory
-
-        // Create the user and UUID directories if they don't exist
-        tokio::fs::create_dir_all(&uuid_dir)
+        let bytes = tokio::fs::read(file.path()).await?;
+        store::put_bytes_at(&location, bytes.into(), config)
            .await
-            .map_err(FileError::Io)?;
+            .map_err(FileError::from)?;

-        // Define the final file path
-        let final_path = uuid_dir.join(file_name);
-        info!("Final path: {:?}", final_path);
-
-        // Copy the temporary file to the final path
-        tokio::fs::copy(file.path(), &final_path)
-            .await
-            .map_err(FileError::Io)?;
-        info!("Copied file to {:?}", final_path);
-
-        Ok(final_path)
+        Ok(location)
    }

    /// Retrieves a `FileInfo` by SHA256.
@@ -240,37 +223,26 @@ impl FileInfo {
    ///
    /// # Returns
    ///  `Result<(), FileError>`
-    pub async fn delete_by_id(id: &str, db_client: &SurrealDbClient) -> Result<(), AppError> {
+    pub async fn delete_by_id(
+        id: &str,
+        db_client: &SurrealDbClient,
+        config: &AppConfig,
+    ) -> Result<(), AppError> {
        // Get the FileInfo from the database
-        let file_info = match db_client.get_item::<FileInfo>(id).await? {
-            Some(info) => info,
-            None => {
-                return Err(AppError::from(FileError::FileNotFound(format!(
-                    "File with id {} was not found",
-                    id
-                ))))
-            }
+        let Some(file_info) = db_client.get_item::<FileInfo>(id).await? else {
+            return Ok(());
        };

-        // Remove the file and its parent directory
-        let file_path = Path::new(&file_info.path);
-        if file_path.exists() {
-            // Get the parent directory of the file
-            if let Some(parent_dir) = file_path.parent() {
-                // Remove the entire directory containing the file
-                remove_dir_all(parent_dir).await?;
-                info!("Removed directory {:?} and its contents", parent_dir);
-            } else {
-                return Err(AppError::from(FileError::FileNotFound(
-                    "File has no parent directory".to_string(),
-                )));
-            }
-        } else {
-            return Err(AppError::from(FileError::FileNotFound(format!(
-                "File at path {:?} was not found",
-                file_path
-            ))));
-        }
+        // Remove the object's parent prefix in the object store
+        let (parent_prefix, _file_name) = store::split_object_path(&file_info.path)
+            .map_err(|e| AppError::from(anyhow::anyhow!(e)))?;
+        store::delete_prefix_at(&parent_prefix, config)
+            .await
+            .map_err(|e| AppError::from(anyhow::anyhow!(e)))?;
+        info!(
+            "Removed object prefix {} and its contents via object_store",
+            parent_prefix
+        );

        // Delete the FileInfo from the database
        db_client.delete_item::<FileInfo>(id).await?;
@@ -298,6 +270,7 @@ impl FileInfo {
 #[cfg(test)]
 mod tests {
    use super::*;
+    use crate::utils::config::{PdfIngestMode::LlmFirst, StorageKind};
    use axum::http::HeaderMap;
    use axum_typed_multipart::FieldMetadata;
    use std::io::Write;
@@ -326,7 +299,7 @@ mod tests {
    }

    #[tokio::test]
-    async fn test_cross_filesystem_file_operations() {
+    async fn test_fileinfo_create_read_delete() {
        // Setup in-memory database for testing
        let namespace = "test_ns";
        let database = &Uuid::new_v4().to_string();
@@ -349,6 +322,10 @@ mod tests {
            surrealdb_password: "test_pass".to_string(),
            surrealdb_namespace: "test_ns".to_string(),
            surrealdb_database: "test_db".to_string(),
+            http_port: 3000,
+            openai_base_url: "..".to_string(),
+            storage: StorageKind::Local,
+            pdf_ingest_mode: LlmFirst,
        };

        // Test file creation
@@ -356,14 +333,11 @@ mod tests {
            .await
            .expect("Failed to create file across filesystems");

-        // Verify the file exists and has correct content
-        let file_path = Path::new(&file_info.path);
-        assert!(file_path.exists(), "File should exist at {:?}", file_path);
-
-        let file_content = tokio::fs::read_to_string(file_path)
+        // Verify the file exists via object_store and has correct content
+        let bytes = store::get_bytes_at(&file_info.path, &config)
            .await
-            .expect("Failed to read file content");
-        assert_eq!(file_content, String::from_utf8_lossy(content));
+            .expect("Failed to read file content via object_store");
+        assert_eq!(bytes, content.as_slice());

        // Test file reading
        let retrieved = FileInfo::get_by_id(&file_info.id, &db)
@@ -373,17 +347,20 @@ mod tests {
        assert_eq!(retrieved.sha256, file_info.sha256);

        // Test file deletion
-        FileInfo::delete_by_id(&file_info.id, &db)
+        FileInfo::delete_by_id(&file_info.id, &db, &config)
            .await
            .expect("Failed to delete file");
-        assert!(!file_path.exists(), "File should be deleted");
+        assert!(
+            store::get_bytes_at(&file_info.path, &config).await.is_err(),
+            "File should be deleted"
+        );

        // Clean up the test directory
        let _ = tokio::fs::remove_dir_all(&config.data_dir).await;
    }

    #[tokio::test]
-    async fn test_cross_filesystem_duplicate_detection() {
+    async fn test_fileinfo_duplicate_detection() {
        // Setup in-memory database for testing
        let namespace = "test_ns";
        let database = &Uuid::new_v4().to_string();
@@ -406,6 +383,10 @@ mod tests {
            surrealdb_password: "test_pass".to_string(),
            surrealdb_namespace: "test_ns".to_string(),
            surrealdb_database: "test_db".to_string(),
+            http_port: 3000,
+            openai_base_url: "..".to_string(),
+            storage: StorageKind::Local,
+            pdf_ingest_mode: LlmFirst,
        };

        // Store the original file
@@ -429,7 +410,7 @@ mod tests {
        assert_ne!(duplicate_file_info.file_name, duplicate_name);

        // Clean up
-        FileInfo::delete_by_id(&original_file_info.id, &db)
+        FileInfo::delete_by_id(&original_file_info.id, &db, &config)
            .await
            .expect("Failed to delete file");
        let _ = tokio::fs::remove_dir_all(&config.data_dir).await;
@@ -459,6 +440,10 @@ mod tests {
            surrealdb_password: "test_pass".to_string(),
            surrealdb_namespace: "test_ns".to_string(),
            surrealdb_database: "test_db".to_string(),
+            http_port: 3000,
+            openai_base_url: "..".to_string(),
+            storage: StorageKind::Local,
+            pdf_ingest_mode: LlmFirst,
        };
        let file_info = FileInfo::new(field_data, &db, user_id, &config).await;

@@ -472,6 +457,11 @@ mod tests {
        assert_eq!(file_info.file_name, file_name);
        assert!(!file_info.sha256.is_empty());
        assert!(!file_info.path.is_empty());
+        // path should be logical: "user_id/uuid/file_name"
+        let parts: Vec<&str> = file_info.path.split('/').collect();
+        assert_eq!(parts.len(), 3);
+        assert_eq!(parts[0], user_id);
+        assert_eq!(parts[2], file_name);
        assert!(file_info.mime_type.contains("text/plain"));

        // Verify it's in the database
@@ -508,6 +498,10 @@ mod tests {
            surrealdb_password: "test_pass".to_string(),
            surrealdb_namespace: "test_ns".to_string(),
            surrealdb_database: "test_db".to_string(),
+            http_port: 3000,
+            openai_base_url: "..".to_string(),
+            storage: StorageKind::Local,
+            pdf_ingest_mode: LlmFirst,
        };

        let field_data1 = create_test_file(content, file_name);
@@ -659,50 +653,28 @@ mod tests {
            .await
            .expect("Failed to start in-memory surrealdb");

-        // Create a FileInfo instance directly (without persistence to disk)
-        let now = Utc::now();
-        let file_id = Uuid::new_v4().to_string();
-
-        // Create a temporary directory that mimics the structure we would have on disk
-        let base_dir = Path::new("./data");
-        let user_id = "test_user";
-        let user_dir = base_dir.join(user_id);
-        let uuid_dir = user_dir.join(&file_id);
-
-        tokio::fs::create_dir_all(&uuid_dir)
-            .await
-            .expect("Failed to create test directories");
-
-        // Create a test file in the directory
-        let test_file_path = uuid_dir.join("test_file.txt");
-        tokio::fs::write(&test_file_path, b"test content")
-            .await
-            .expect("Failed to write test file");
-
-        // The file path should point to our test file
-        let file_info = FileInfo {
-            id: file_id.clone(),
-            user_id: "user123".to_string(),
-            created_at: now,
-            updated_at: now,
-            sha256: "test_sha256_hash".to_string(),
-            path: test_file_path.to_string_lossy().to_string(),
-            file_name: "test_file.txt".to_string(),
-            mime_type: "text/plain".to_string(),
+        // Create and persist a test file via FileInfo::new
+        let user_id = "user123";
+        let cfg = AppConfig {
+            data_dir: "./data".to_string(),
+            openai_api_key: "".to_string(),
+            surrealdb_address: "".to_string(),
+            surrealdb_username: "".to_string(),
+            surrealdb_password: "".to_string(),
+            surrealdb_namespace: "".to_string(),
+            surrealdb_database: "".to_string(),
+            http_port: 0,
+            openai_base_url: "".to_string(),
+            storage: crate::utils::config::StorageKind::Local,
+            pdf_ingest_mode: LlmFirst,
        };
-
-        // Store it in the database
-        db.store_item(file_info.clone())
+        let temp = create_test_file(b"test content", "test_file.txt");
+        let file_info = FileInfo::new(temp, &db, user_id, &cfg)
            .await
-            .expect("Failed to store file info");
-
-        // Verify file exists on disk
-        assert!(tokio::fs::try_exists(&test_file_path)
-            .await
-            .unwrap_or(false));
+            .expect("create file");

        // Delete the file
-        let delete_result = FileInfo::delete_by_id(&file_id, &db).await;
+        let delete_result = FileInfo::delete_by_id(&file_info.id, &db, &cfg).await;

        // Delete should be successful
        assert!(
@@ -713,7 +685,7 @@ mod tests {

        // Verify the file is removed from the database
        let retrieved: Option<FileInfo> = db
-            .get_item(&file_id)
+            .get_item(&file_info.id)
            .await
            .expect("Failed to query database");
        assert!(
@@ -721,14 +693,8 @@ mod tests {
            "FileInfo should be deleted from the database"
        );

-        // Verify directory is gone
-        assert!(
-            !tokio::fs::try_exists(&uuid_dir).await.unwrap_or(true),
-            "UUID directory should be deleted"
-        );
-
-        // Clean up test directory if it exists
-        let _ = tokio::fs::remove_dir_all(base_dir).await;
+        // Verify content no longer retrievable
+        assert!(store::get_bytes_at(&file_info.path, &cfg).await.is_err());
    }

    #[tokio::test]
@@ -741,16 +707,27 @@ mod tests {
            .expect("Failed to start in-memory surrealdb");

        // Try to delete a file that doesn't exist
-        let result = FileInfo::delete_by_id("nonexistent_id", &db).await;
+        let result = FileInfo::delete_by_id(
+            "nonexistent_id",
+            &db,
+            &AppConfig {
+                data_dir: "./data".to_string(),
+                openai_api_key: "".to_string(),
+                surrealdb_address: "".to_string(),
+                surrealdb_username: "".to_string(),
+                surrealdb_password: "".to_string(),
+                surrealdb_namespace: "".to_string(),
+                surrealdb_database: "".to_string(),
+                http_port: 0,
+                openai_base_url: "".to_string(),
+                storage: crate::utils::config::StorageKind::Local,
+                pdf_ingest_mode: LlmFirst,
+            },
+        )
+        .await;

-        // Should fail with FileNotFound error
-        assert!(result.is_err());
-        match result {
-            Err(AppError::File(_)) => {
-                // Expected error
-            }
-            _ => panic!("Expected FileNotFound error"),
-        }
+        // Should succeed even if the file record does not exist
+        assert!(result.is_ok());
    }
    #[tokio::test]
    async fn test_get_by_id() {
@@ -820,7 +797,7 @@ mod tests {
    }

    #[tokio::test]
-    async fn test_data_directory_configuration() {
+    async fn test_fileinfo_persist_with_custom_root() {
        // Setup in-memory database for testing
        let namespace = "test_ns";
        let database = &Uuid::new_v4().to_string();
@@ -844,6 +821,10 @@ mod tests {
            surrealdb_password: "test_pass".to_string(),
            surrealdb_namespace: "test_ns".to_string(),
            surrealdb_database: "test_db".to_string(),
+            http_port: 3000,
+            openai_base_url: "..".to_string(),
+            storage: StorageKind::Local,
+            pdf_ingest_mode: LlmFirst,
        };

        // Test file creation
@@ -851,27 +832,17 @@ mod tests {
            .await
            .expect("Failed to create file in custom data directory");

-        // Verify the file exists in the correct location
-        let file_path = Path::new(&file_info.path);
-        assert!(file_path.exists(), "File should exist at {:?}", file_path);
-
-        // Verify the file is in the correct data directory
-        assert!(
-            file_path.starts_with(custom_data_dir),
-            "File should be stored in the custom data directory"
-        );
-
-        // Verify the file has the correct content
-        let file_content = tokio::fs::read_to_string(file_path)
+        // Verify the file has the correct content via object_store
+        let file_content = store::get_bytes_at(&file_info.path, &config)
            .await
            .expect("Failed to read file content");
-        assert_eq!(file_content, String::from_utf8_lossy(content));
+        assert_eq!(file_content.as_ref(), content);

        // Test file deletion
-        FileInfo::delete_by_id(&file_info.id, &db)
+        FileInfo::delete_by_id(&file_info.id, &db, &config)
            .await
            .expect("Failed to delete file");
-        assert!(!file_path.exists(), "File should be deleted");
+        assert!(store::get_bytes_at(&file_info.path, &config).await.is_err());

        // Clean up the test directory
        let _ = tokio::fs::remove_dir_all(custom_data_dir).await;
--- a/common/src/storage/types/ingestion_task.rs
+++ b/common/src/storage/types/ingestion_task.rs
@@ -1,113 +1,529 @@
-use futures::Stream;
-use surrealdb::{opt::PatchOp, Notification};
+use std::time::Duration;
+
+use chrono::Duration as ChronoDuration;
+use state_machines::state_machine;
+use surrealdb::sql::Datetime as SurrealDatetime;
 use uuid::Uuid;

 use crate::{error::AppError, storage::db::SurrealDbClient, stored_object};

 use super::ingestion_payload::IngestionPayload;

-#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
-pub enum IngestionTaskStatus {
-    Created,
-    InProgress {
-        attempts: u32,
-        last_attempt: DateTime<Utc>,
-    },
-    Completed,
-    Error(String),
+pub const MAX_ATTEMPTS: u32 = 3;
+pub const DEFAULT_LEASE_SECS: i64 = 300;
+pub const DEFAULT_PRIORITY: i32 = 0;
+
+#[derive(Debug, Default, Clone, serde::Serialize, serde::Deserialize, PartialEq, Eq)]
+pub enum TaskState {
+    #[serde(rename = "Pending")]
+    #[default]
+    Pending,
+    #[serde(rename = "Reserved")]
+    Reserved,
+    #[serde(rename = "Processing")]
+    Processing,
+    #[serde(rename = "Succeeded")]
+    Succeeded,
+    #[serde(rename = "Failed")]
+    Failed,
+    #[serde(rename = "Cancelled")]
    Cancelled,
+    #[serde(rename = "DeadLetter")]
+    DeadLetter,
+}
+
+impl TaskState {
+    pub fn as_str(&self) -> &'static str {
+        match self {
+            TaskState::Pending => "Pending",
+            TaskState::Reserved => "Reserved",
+            TaskState::Processing => "Processing",
+            TaskState::Succeeded => "Succeeded",
+            TaskState::Failed => "Failed",
+            TaskState::Cancelled => "Cancelled",
+            TaskState::DeadLetter => "DeadLetter",
+        }
+    }
+
+    pub fn is_terminal(&self) -> bool {
+        matches!(
+            self,
+            TaskState::Succeeded | TaskState::Cancelled | TaskState::DeadLetter
+        )
+    }
+
+    pub fn display_label(&self) -> &'static str {
+        match self {
+            TaskState::Pending => "Pending",
+            TaskState::Reserved => "Reserved",
+            TaskState::Processing => "Processing",
+            TaskState::Succeeded => "Completed",
+            TaskState::Failed => "Retrying",
+            TaskState::Cancelled => "Cancelled",
+            TaskState::DeadLetter => "Dead Letter",
+        }
+    }
+}
+
+#[derive(Debug, Clone, serde::Serialize, serde::Deserialize, PartialEq, Eq, Default)]
+pub struct TaskErrorInfo {
+    pub code: Option<String>,
+    pub message: String,
+}
+
+#[derive(Debug, Clone, Copy)]
+enum TaskTransition {
+    StartProcessing,
+    Succeed,
+    Fail,
+    Cancel,
+    DeadLetter,
+    Release,
+}
+
+impl TaskTransition {
+    fn as_str(&self) -> &'static str {
+        match self {
+            TaskTransition::StartProcessing => "start_processing",
+            TaskTransition::Succeed => "succeed",
+            TaskTransition::Fail => "fail",
+            TaskTransition::Cancel => "cancel",
+            TaskTransition::DeadLetter => "deadletter",
+            TaskTransition::Release => "release",
+        }
+    }
+}
+
+mod lifecycle {
+    use super::state_machine;
+
+    state_machine! {
+        name: TaskLifecycleMachine,
+        initial: Pending,
+        states: [Pending, Reserved, Processing, Succeeded, Failed, Cancelled, DeadLetter],
+        events {
+            reserve {
+                transition: { from: Pending, to: Reserved }
+                transition: { from: Failed, to: Reserved }
+            }
+            start_processing {
+                transition: { from: Reserved, to: Processing }
+            }
+            succeed {
+                transition: { from: Processing, to: Succeeded }
+            }
+            fail {
+                transition: { from: Processing, to: Failed }
+            }
+            cancel {
+                transition: { from: Pending, to: Cancelled }
+                transition: { from: Reserved, to: Cancelled }
+                transition: { from: Processing, to: Cancelled }
+            }
+            deadletter {
+                transition: { from: Failed, to: DeadLetter }
+            }
+            release {
+                transition: { from: Reserved, to: Pending }
+            }
+        }
+    }
+
+    pub(super) fn pending() -> TaskLifecycleMachine<(), Pending> {
+        TaskLifecycleMachine::new(())
+    }
+
+    pub(super) fn reserved() -> TaskLifecycleMachine<(), Reserved> {
+        pending()
+            .reserve()
+            .expect("reserve transition from Pending should exist")
+    }
+
+    pub(super) fn processing() -> TaskLifecycleMachine<(), Processing> {
+        reserved()
+            .start_processing()
+            .expect("start_processing transition from Reserved should exist")
+    }
+
+    pub(super) fn failed() -> TaskLifecycleMachine<(), Failed> {
+        processing()
+            .fail()
+            .expect("fail transition from Processing should exist")
+    }
+}
+
+fn invalid_transition(state: &TaskState, event: TaskTransition) -> AppError {
+    AppError::Validation(format!(
+        "Invalid task transition: {} -> {}",
+        state.as_str(),
+        event.as_str()
+    ))
 }

 stored_object!(IngestionTask, "ingestion_task", {
    content: IngestionPayload,
-    status: IngestionTaskStatus,
-    user_id: String
+    state: TaskState,
+    user_id: String,
+    attempts: u32,
+    max_attempts: u32,
+    #[serde(serialize_with = "serialize_datetime", deserialize_with = "deserialize_datetime")]
+    scheduled_at: chrono::DateTime<chrono::Utc>,
+    #[serde(
+        serialize_with = "serialize_option_datetime",
+        deserialize_with = "deserialize_option_datetime",
+        default
+    )]
+    locked_at: Option<chrono::DateTime<chrono::Utc>>,
+    lease_duration_secs: i64,
+    worker_id: Option<String>,
+    error_code: Option<String>,
+    error_message: Option<String>,
+    #[serde(
+        serialize_with = "serialize_option_datetime",
+        deserialize_with = "deserialize_option_datetime",
+        default
+    )]
+    last_error_at: Option<chrono::DateTime<chrono::Utc>>,
+    priority: i32
 });

-pub const MAX_ATTEMPTS: u32 = 3;
-
 impl IngestionTask {
-    pub async fn new(content: IngestionPayload, user_id: String) -> Self {
-        let now = Utc::now();
+    pub fn new(content: IngestionPayload, user_id: String) -> Self {
+        let now = chrono::Utc::now();

        Self {
            id: Uuid::new_v4().to_string(),
            content,
-            status: IngestionTaskStatus::Created,
+            state: TaskState::Pending,
+            user_id,
+            attempts: 0,
+            max_attempts: MAX_ATTEMPTS,
+            scheduled_at: now,
+            locked_at: None,
+            lease_duration_secs: DEFAULT_LEASE_SECS,
+            worker_id: None,
+            error_code: None,
+            error_message: None,
+            last_error_at: None,
+            priority: DEFAULT_PRIORITY,
            created_at: now,
            updated_at: now,
-            user_id,
        }
    }

-    /// Creates a new job and stores it in the database
+    pub fn can_retry(&self) -> bool {
+        self.attempts < self.max_attempts
+    }
+
+    pub fn lease_duration(&self) -> Duration {
+        Duration::from_secs(self.lease_duration_secs.max(0) as u64)
+    }
+
    pub async fn create_and_add_to_db(
        content: IngestionPayload,
        user_id: String,
        db: &SurrealDbClient,
    ) -> Result<IngestionTask, AppError> {
-        let task = Self::new(content, user_id).await;
-
+        let task = Self::new(content, user_id);
        db.store_item(task.clone()).await?;
-
        Ok(task)
    }

-    // Update job status
-    pub async fn update_status(
-        id: &str,
-        status: IngestionTaskStatus,
+    pub async fn claim_next_ready(
        db: &SurrealDbClient,
-    ) -> Result<(), AppError> {
-        let _job: Option<Self> = db
-            .update((Self::table_name(), id))
-            .patch(PatchOp::replace("/status", status))
-            .patch(PatchOp::replace(
-                "/updated_at",
-                surrealdb::sql::Datetime::default(),
+        worker_id: &str,
+        now: chrono::DateTime<chrono::Utc>,
+        lease_duration: Duration,
+    ) -> Result<Option<IngestionTask>, AppError> {
+        debug_assert!(lifecycle::pending().reserve().is_ok());
+        debug_assert!(lifecycle::failed().reserve().is_ok());
+
+        const CLAIM_QUERY: &str = r#"
+            UPDATE (
+                SELECT * FROM type::table($table)
+                WHERE state IN $candidate_states
+                  AND scheduled_at <= $now
+                  AND (
+                        attempts < max_attempts
+                        OR state IN $sticky_states
+                  )
+                  AND (
+                        locked_at = NONE
+                        OR time::unix($now) - time::unix(locked_at) >= lease_duration_secs
+                  )
+                ORDER BY priority DESC, scheduled_at ASC, created_at ASC
+                LIMIT 1
+            )
+            SET state = $reserved_state,
+                attempts = if state IN $increment_states THEN
+                    if attempts + 1 > max_attempts THEN max_attempts ELSE attempts + 1 END
+                ELSE
+                    attempts
+                END,
+                locked_at = $now,
+                worker_id = $worker_id,
+                lease_duration_secs = $lease_secs,
+                updated_at = $now
+            RETURN *;
+        "#;
+
+        let mut result = db
+            .client
+            .query(CLAIM_QUERY)
+            .bind(("table", Self::table_name()))
+            .bind((
+                "candidate_states",
+                vec![
+                    TaskState::Pending.as_str(),
+                    TaskState::Failed.as_str(),
+                    TaskState::Reserved.as_str(),
+                    TaskState::Processing.as_str(),
+                ],
            ))
+            .bind((
+                "sticky_states",
+                vec![TaskState::Reserved.as_str(), TaskState::Processing.as_str()],
+            ))
+            .bind((
+                "increment_states",
+                vec![TaskState::Pending.as_str(), TaskState::Failed.as_str()],
+            ))
+            .bind(("reserved_state", TaskState::Reserved.as_str()))
+            .bind(("now", SurrealDatetime::from(now)))
+            .bind(("worker_id", worker_id.to_string()))
+            .bind(("lease_secs", lease_duration.as_secs() as i64))
            .await?;

-        Ok(())
+        let task: Option<IngestionTask> = result.take(0)?;
+        Ok(task)
    }

-    /// Listen for new jobs
-    pub async fn listen_for_tasks(
+    pub async fn mark_processing(&self, db: &SurrealDbClient) -> Result<IngestionTask, AppError> {
+        const START_PROCESSING_QUERY: &str = r#"
+            UPDATE type::thing($table, $id)
+            SET state = $processing,
+                updated_at = $now,
+                locked_at = $now
+            WHERE state = $reserved AND worker_id = $worker_id
+            RETURN *;
+        "#;
+
+        let now = chrono::Utc::now();
+        let mut result = db
+            .client
+            .query(START_PROCESSING_QUERY)
+            .bind(("table", Self::table_name()))
+            .bind(("id", self.id.clone()))
+            .bind(("processing", TaskState::Processing.as_str()))
+            .bind(("reserved", TaskState::Reserved.as_str()))
+            .bind(("now", SurrealDatetime::from(now)))
+            .bind(("worker_id", self.worker_id.clone().unwrap_or_default()))
+            .await?;
+
+        let updated: Option<IngestionTask> = result.take(0)?;
+        updated.ok_or_else(|| invalid_transition(&self.state, TaskTransition::StartProcessing))
+    }
+
+    pub async fn mark_succeeded(&self, db: &SurrealDbClient) -> Result<IngestionTask, AppError> {
+        const COMPLETE_QUERY: &str = r#"
+            UPDATE type::thing($table, $id)
+            SET state = $succeeded,
+                updated_at = $now,
+                locked_at = NONE,
+                worker_id = NONE,
+                scheduled_at = $now,
+                error_code = NONE,
+                error_message = NONE,
+                last_error_at = NONE
+            WHERE state = $processing AND worker_id = $worker_id
+            RETURN *;
+        "#;
+
+        let now = chrono::Utc::now();
+        let mut result = db
+            .client
+            .query(COMPLETE_QUERY)
+            .bind(("table", Self::table_name()))
+            .bind(("id", self.id.clone()))
+            .bind(("succeeded", TaskState::Succeeded.as_str()))
+            .bind(("processing", TaskState::Processing.as_str()))
+            .bind(("now", SurrealDatetime::from(now)))
+            .bind(("worker_id", self.worker_id.clone().unwrap_or_default()))
+            .await?;
+
+        let updated: Option<IngestionTask> = result.take(0)?;
+        updated.ok_or_else(|| invalid_transition(&self.state, TaskTransition::Succeed))
+    }
+
+    pub async fn mark_failed(
+        &self,
+        error: TaskErrorInfo,
+        retry_delay: Duration,
        db: &SurrealDbClient,
-    ) -> Result<impl Stream<Item = Result<Notification<Self>, surrealdb::Error>>, surrealdb::Error>
-    {
-        db.listen::<Self>().await
+    ) -> Result<IngestionTask, AppError> {
+        let now = chrono::Utc::now();
+        let retry_at = now
+            + ChronoDuration::from_std(retry_delay).unwrap_or_else(|_| ChronoDuration::seconds(30));
+
+        const FAIL_QUERY: &str = r#"
+            UPDATE type::thing($table, $id)
+            SET state = $failed,
+                updated_at = $now,
+                locked_at = NONE,
+                worker_id = NONE,
+                scheduled_at = $retry_at,
+                error_code = $error_code,
+                error_message = $error_message,
+                last_error_at = $now
+            WHERE state = $processing AND worker_id = $worker_id
+            RETURN *;
+        "#;
+
+        let mut result = db
+            .client
+            .query(FAIL_QUERY)
+            .bind(("table", Self::table_name()))
+            .bind(("id", self.id.clone()))
+            .bind(("failed", TaskState::Failed.as_str()))
+            .bind(("processing", TaskState::Processing.as_str()))
+            .bind(("now", SurrealDatetime::from(now)))
+            .bind(("retry_at", SurrealDatetime::from(retry_at)))
+            .bind(("error_code", error.code.clone()))
+            .bind(("error_message", error.message.clone()))
+            .bind(("worker_id", self.worker_id.clone().unwrap_or_default()))
+            .await?;
+
+        let updated: Option<IngestionTask> = result.take(0)?;
+        updated.ok_or_else(|| invalid_transition(&self.state, TaskTransition::Fail))
    }

-    /// Get all unfinished tasks, ie newly created and in progress up two times
-    pub async fn get_unfinished_tasks(db: &SurrealDbClient) -> Result<Vec<Self>, AppError> {
-        let jobs: Vec<Self> = db
+    pub async fn mark_dead_letter(
+        &self,
+        error: TaskErrorInfo,
+        db: &SurrealDbClient,
+    ) -> Result<IngestionTask, AppError> {
+        const DEAD_LETTER_QUERY: &str = r#"
+            UPDATE type::thing($table, $id)
+            SET state = $dead,
+                updated_at = $now,
+                locked_at = NONE,
+                worker_id = NONE,
+                scheduled_at = $now,
+                error_code = $error_code,
+                error_message = $error_message,
+                last_error_at = $now
+            WHERE state = $failed
+            RETURN *;
+        "#;
+
+        let now = chrono::Utc::now();
+        let mut result = db
+            .client
+            .query(DEAD_LETTER_QUERY)
+            .bind(("table", Self::table_name()))
+            .bind(("id", self.id.clone()))
+            .bind(("dead", TaskState::DeadLetter.as_str()))
+            .bind(("failed", TaskState::Failed.as_str()))
+            .bind(("now", SurrealDatetime::from(now)))
+            .bind(("error_code", error.code.clone()))
+            .bind(("error_message", error.message.clone()))
+            .await?;
+
+        let updated: Option<IngestionTask> = result.take(0)?;
+        updated.ok_or_else(|| invalid_transition(&self.state, TaskTransition::DeadLetter))
+    }
+
+    pub async fn mark_cancelled(&self, db: &SurrealDbClient) -> Result<IngestionTask, AppError> {
+        const CANCEL_QUERY: &str = r#"
+            UPDATE type::thing($table, $id)
+            SET state = $cancelled,
+                updated_at = $now,
+                locked_at = NONE,
+                worker_id = NONE
+            WHERE state IN $allow_states
+            RETURN *;
+        "#;
+
+        let now = chrono::Utc::now();
+        let mut result = db
+            .client
+            .query(CANCEL_QUERY)
+            .bind(("table", Self::table_name()))
+            .bind(("id", self.id.clone()))
+            .bind(("cancelled", TaskState::Cancelled.as_str()))
+            .bind((
+                "allow_states",
+                vec![
+                    TaskState::Pending.as_str(),
+                    TaskState::Reserved.as_str(),
+                    TaskState::Processing.as_str(),
+                ],
+            ))
+            .bind(("now", SurrealDatetime::from(now)))
+            .await?;
+
+        let updated: Option<IngestionTask> = result.take(0)?;
+        updated.ok_or_else(|| invalid_transition(&self.state, TaskTransition::Cancel))
+    }
+
+    pub async fn release(&self, db: &SurrealDbClient) -> Result<IngestionTask, AppError> {
+        const RELEASE_QUERY: &str = r#"
+            UPDATE type::thing($table, $id)
+            SET state = $pending,
+                updated_at = $now,
+                locked_at = NONE,
+                worker_id = NONE
+            WHERE state = $reserved
+            RETURN *;
+        "#;
+
+        let now = chrono::Utc::now();
+        let mut result = db
+            .client
+            .query(RELEASE_QUERY)
+            .bind(("table", Self::table_name()))
+            .bind(("id", self.id.clone()))
+            .bind(("pending", TaskState::Pending.as_str()))
+            .bind(("reserved", TaskState::Reserved.as_str()))
+            .bind(("now", SurrealDatetime::from(now)))
+            .await?;
+
+        let updated: Option<IngestionTask> = result.take(0)?;
+        updated.ok_or_else(|| invalid_transition(&self.state, TaskTransition::Release))
+    }
+
+    pub async fn get_unfinished_tasks(
+        db: &SurrealDbClient,
+    ) -> Result<Vec<IngestionTask>, AppError> {
+        let tasks: Vec<IngestionTask> = db
            .query(
-                "SELECT * FROM type::table($table) 
-             WHERE 
-                status = 'Created' 
-                OR (
-                    status.InProgress != NONE 
-                    AND status.InProgress.attempts < $max_attempts
-                )
-             ORDER BY created_at ASC",
+                "SELECT * FROM type::table($table)
+                 WHERE state IN $active_states
+                 ORDER BY scheduled_at ASC, created_at ASC",
            )
            .bind(("table", Self::table_name()))
-            .bind(("max_attempts", MAX_ATTEMPTS))
+            .bind((
+                "active_states",
+                vec![
+                    TaskState::Pending.as_str(),
+                    TaskState::Reserved.as_str(),
+                    TaskState::Processing.as_str(),
+                    TaskState::Failed.as_str(),
+                ],
+            ))
            .await?
            .take(0)?;

-        Ok(jobs)
+        Ok(tasks)
    }
 }

 #[cfg(test)]
 mod tests {
    use super::*;
-    use chrono::Utc;
+    use crate::storage::types::ingestion_payload::IngestionPayload;

-    // Helper function to create a test ingestion payload
-    fn create_test_payload(user_id: &str) -> IngestionPayload {
+    fn create_payload(user_id: &str) -> IngestionPayload {
        IngestionPayload::Text {
            text: "Test content".to_string(),
            context: "Test context".to_string(),
@@ -116,180 +532,197 @@ mod tests {
        }
    }

+    async fn memory_db() -> SurrealDbClient {
+        let namespace = "test_ns";
+        let database = Uuid::new_v4().to_string();
+        SurrealDbClient::memory(namespace, &database)
+            .await
+            .expect("in-memory surrealdb")
+    }
+
    #[tokio::test]
-    async fn test_new_ingestion_task() {
+    async fn test_new_task_defaults() {
        let user_id = "user123";
-        let payload = create_test_payload(user_id);
+        let payload = create_payload(user_id);
+        let task = IngestionTask::new(payload.clone(), user_id.to_string());

-        let task = IngestionTask::new(payload.clone(), user_id.to_string()).await;
-
-        // Verify task properties
        assert_eq!(task.user_id, user_id);
        assert_eq!(task.content, payload);
-        assert!(matches!(task.status, IngestionTaskStatus::Created));
-        assert!(!task.id.is_empty());
+        assert_eq!(task.state, TaskState::Pending);
+        assert_eq!(task.attempts, 0);
+        assert_eq!(task.max_attempts, MAX_ATTEMPTS);
+        assert!(task.locked_at.is_none());
+        assert!(task.worker_id.is_none());
    }

    #[tokio::test]
-    async fn test_create_and_add_to_db() {
-        // Setup in-memory database
-        let namespace = "test_ns";
-        let database = &Uuid::new_v4().to_string();
-        let db = SurrealDbClient::memory(namespace, database)
-            .await
-            .expect("Failed to start in-memory surrealdb");
-
+    async fn test_create_and_store_task() {
+        let db = memory_db().await;
        let user_id = "user123";
-        let payload = create_test_payload(user_id);
+        let payload = create_payload(user_id);

-        // Create and store task
-        IngestionTask::create_and_add_to_db(payload.clone(), user_id.to_string(), &db)
+        let created =
+            IngestionTask::create_and_add_to_db(payload.clone(), user_id.to_string(), &db)
+                .await
+                .expect("store");
+
+        let stored: Option<IngestionTask> = db
+            .get_item::<IngestionTask>(&created.id)
            .await
-            .expect("Failed to create and add task to db");
+            .expect("fetch");

-        // Query to verify task was stored
-        let query = format!(
-            "SELECT * FROM {} WHERE user_id = '{}'",
-            IngestionTask::table_name(),
-            user_id
-        );
-        let mut result = db.query(query).await.expect("Query failed");
-        let tasks: Vec<IngestionTask> = result.take(0).unwrap_or_default();
-
-        // Verify task is in the database
-        assert!(!tasks.is_empty(), "Task should exist in the database");
-        let stored_task = &tasks[0];
-        assert_eq!(stored_task.user_id, user_id);
-        assert!(matches!(stored_task.status, IngestionTaskStatus::Created));
+        let stored = stored.expect("task exists");
+        assert_eq!(stored.id, created.id);
+        assert_eq!(stored.state, TaskState::Pending);
+        assert_eq!(stored.attempts, 0);
    }

    #[tokio::test]
-    async fn test_update_status() {
-        // Setup in-memory database
-        let namespace = "test_ns";
-        let database = &Uuid::new_v4().to_string();
-        let db = SurrealDbClient::memory(namespace, database)
-            .await
-            .expect("Failed to start in-memory surrealdb");
-
+    async fn test_claim_and_transition() {
+        let db = memory_db().await;
        let user_id = "user123";
-        let payload = create_test_payload(user_id);
+        let payload = create_payload(user_id);
+        let task = IngestionTask::new(payload, user_id.to_string());
+        db.store_item(task.clone()).await.expect("store");

-        // Create task manually
-        let task = IngestionTask::new(payload.clone(), user_id.to_string()).await;
-        let task_id = task.id.clone();
+        let worker_id = "worker-1";
+        let now = chrono::Utc::now();
+        let claimed = IngestionTask::claim_next_ready(&db, worker_id, now, Duration::from_secs(60))
+            .await
+            .expect("claim");

-        // Store task
-        db.store_item(task).await.expect("Failed to store task");
+        let claimed = claimed.expect("task claimed");
+        assert_eq!(claimed.state, TaskState::Reserved);
+        assert_eq!(claimed.worker_id.as_deref(), Some(worker_id));

-        // Update status to InProgress
-        let now = Utc::now();
-        let new_status = IngestionTaskStatus::InProgress {
-            attempts: 1,
-            last_attempt: now,
+        let processing = claimed.mark_processing(&db).await.expect("processing");
+        assert_eq!(processing.state, TaskState::Processing);
+
+        let succeeded = processing.mark_succeeded(&db).await.expect("succeeded");
+        assert_eq!(succeeded.state, TaskState::Succeeded);
+        assert!(succeeded.worker_id.is_none());
+        assert!(succeeded.locked_at.is_none());
+    }
+
+    #[tokio::test]
+    async fn test_fail_and_dead_letter() {
+        let db = memory_db().await;
+        let user_id = "user123";
+        let payload = create_payload(user_id);
+        let task = IngestionTask::new(payload, user_id.to_string());
+        db.store_item(task.clone()).await.expect("store");
+
+        let worker_id = "worker-dead";
+        let now = chrono::Utc::now();
+        let claimed = IngestionTask::claim_next_ready(&db, worker_id, now, Duration::from_secs(60))
+            .await
+            .expect("claim")
+            .expect("claimed");
+
+        let processing = claimed.mark_processing(&db).await.expect("processing");
+
+        let error_info = TaskErrorInfo {
+            code: Some("pipeline_error".into()),
+            message: "failed".into(),
        };

-        IngestionTask::update_status(&task_id, new_status.clone(), &db)
+        let failed = processing
+            .mark_failed(error_info.clone(), Duration::from_secs(30), &db)
            .await
-            .expect("Failed to update status");
+            .expect("failed update");
+        assert_eq!(failed.state, TaskState::Failed);
+        assert_eq!(failed.error_message.as_deref(), Some("failed"));
+        assert!(failed.worker_id.is_none());
+        assert!(failed.locked_at.is_none());
+        assert!(failed.scheduled_at > now);

-        // Verify status updated
-        let updated_task: Option<IngestionTask> = db
-            .get_item::<IngestionTask>(&task_id)
+        let dead = failed
+            .mark_dead_letter(error_info.clone(), &db)
            .await
-            .expect("Failed to get updated task");
+            .expect("dead letter");
+        assert_eq!(dead.state, TaskState::DeadLetter);
+        assert_eq!(dead.error_message.as_deref(), Some("failed"));
+    }

-        assert!(updated_task.is_some());
-        let updated_task = updated_task.unwrap();
+    #[tokio::test]
+    async fn test_mark_processing_requires_reservation() {
+        let db = memory_db().await;
+        let user_id = "user123";
+        let payload = create_payload(user_id);

-        match updated_task.status {
-            IngestionTaskStatus::InProgress { attempts, .. } => {
-                assert_eq!(attempts, 1);
+        let task = IngestionTask::new(payload.clone(), user_id.to_string());
+        db.store_item(task.clone()).await.expect("store");
+
+        let err = task
+            .mark_processing(&db)
+            .await
+            .expect_err("processing should fail without reservation");
+
+        match err {
+            AppError::Validation(message) => {
+                assert!(
+                    message.contains("Pending -> start_processing"),
+                    "unexpected message: {message}"
+                );
            }
-            _ => panic!("Expected InProgress status"),
+            other => panic!("expected validation error, got {other:?}"),
        }
    }

    #[tokio::test]
-    async fn test_get_unfinished_tasks() {
-        // Setup in-memory database
-        let namespace = "test_ns";
-        let database = &Uuid::new_v4().to_string();
-        let db = SurrealDbClient::memory(namespace, database)
-            .await
-            .expect("Failed to start in-memory surrealdb");
-
+    async fn test_mark_failed_requires_processing() {
+        let db = memory_db().await;
        let user_id = "user123";
-        let payload = create_test_payload(user_id);
+        let payload = create_payload(user_id);

-        // Create tasks with different statuses
-        let created_task = IngestionTask::new(payload.clone(), user_id.to_string()).await;
+        let task = IngestionTask::new(payload.clone(), user_id.to_string());
+        db.store_item(task.clone()).await.expect("store");

-        let mut in_progress_task = IngestionTask::new(payload.clone(), user_id.to_string()).await;
-        in_progress_task.status = IngestionTaskStatus::InProgress {
-            attempts: 1,
-            last_attempt: Utc::now(),
-        };
-
-        let mut max_attempts_task = IngestionTask::new(payload.clone(), user_id.to_string()).await;
-        max_attempts_task.status = IngestionTaskStatus::InProgress {
-            attempts: MAX_ATTEMPTS,
-            last_attempt: Utc::now(),
-        };
-
-        let mut completed_task = IngestionTask::new(payload.clone(), user_id.to_string()).await;
-        completed_task.status = IngestionTaskStatus::Completed;
-
-        let mut error_task = IngestionTask::new(payload.clone(), user_id.to_string()).await;
-        error_task.status = IngestionTaskStatus::Error("Test error".to_string());
-
-        // Store all tasks
-        db.store_item(created_task)
+        let err = task
+            .mark_failed(
+                TaskErrorInfo {
+                    code: None,
+                    message: "boom".into(),
+                },
+                Duration::from_secs(30),
+                &db,
+            )
            .await
-            .expect("Failed to store created task");
-        db.store_item(in_progress_task)
-            .await
-            .expect("Failed to store in-progress task");
-        db.store_item(max_attempts_task)
-            .await
-            .expect("Failed to store max-attempts task");
-        db.store_item(completed_task)
-            .await
-            .expect("Failed to store completed task");
-        db.store_item(error_task)
-            .await
-            .expect("Failed to store error task");
+            .expect_err("failing should require processing state");

-        // Get unfinished tasks
-        let unfinished_tasks = IngestionTask::get_unfinished_tasks(&db)
+        match err {
+            AppError::Validation(message) => {
+                assert!(
+                    message.contains("Pending -> fail"),
+                    "unexpected message: {message}"
+                );
+            }
+            other => panic!("expected validation error, got {other:?}"),
+        }
+    }
+
+    #[tokio::test]
+    async fn test_release_requires_reservation() {
+        let db = memory_db().await;
+        let user_id = "user123";
+        let payload = create_payload(user_id);
+
+        let task = IngestionTask::new(payload.clone(), user_id.to_string());
+        db.store_item(task.clone()).await.expect("store");
+
+        let err = task
+            .release(&db)
            .await
-            .expect("Failed to get unfinished tasks");
+            .expect_err("release should require reserved state");

-        // Verify only Created and InProgress with attempts < MAX_ATTEMPTS are returned
-        assert_eq!(unfinished_tasks.len(), 2);
-
-        let statuses: Vec<_> = unfinished_tasks
-            .iter()
-            .map(|task| match &task.status {
-                IngestionTaskStatus::Created => "Created",
-                IngestionTaskStatus::InProgress { attempts, .. } => {
-                    if *attempts < MAX_ATTEMPTS {
-                        "InProgress<MAX"
-                    } else {
-                        "InProgress>=MAX"
-                    }
-                }
-                IngestionTaskStatus::Completed => "Completed",
-                IngestionTaskStatus::Error(_) => "Error",
-                IngestionTaskStatus::Cancelled => "Cancelled",
-            })
-            .collect();
-
-        assert!(statuses.contains(&"Created"));
-        assert!(statuses.contains(&"InProgress<MAX"));
-        assert!(!statuses.contains(&"InProgress>=MAX"));
-        assert!(!statuses.contains(&"Completed"));
-        assert!(!statuses.contains(&"Error"));
-        assert!(!statuses.contains(&"Cancelled"));
+        match err {
+            AppError::Validation(message) => {
+                assert!(
+                    message.contains("Pending -> release"),
+                    "unexpected message: {message}"
+                );
+            }
+            other => panic!("expected validation error, got {other:?}"),
+        }
    }
 }
--- a/common/src/storage/types/knowledge_entity.rs
+++ b/common/src/storage/types/knowledge_entity.rs
@@ -1,8 +1,15 @@
+use std::collections::HashMap;
+
 use crate::{
    error::AppError, storage::db::SurrealDbClient, stored_object,
    utils::embedding::generate_embedding,
 };
 use async_openai::{config::OpenAIConfig, Client};
+use tokio_retry::{
+    strategy::{jitter, ExponentialBackoff},
+    Retry,
+};
+use tracing::{error, info};
 use uuid::Uuid;

 #[derive(Debug, Serialize, Deserialize, Clone, PartialEq)]
@@ -33,6 +40,38 @@ impl From<String> for KnowledgeEntityType {
    }
 }

+#[derive(Debug, Deserialize, Serialize)]
+pub struct KnowledgeEntitySearchResult {
+    #[serde(deserialize_with = "deserialize_flexible_id")]
+    pub id: String,
+    #[serde(
+        serialize_with = "serialize_datetime",
+        deserialize_with = "deserialize_datetime",
+        default
+    )]
+    pub created_at: DateTime<Utc>,
+    #[serde(
+        serialize_with = "serialize_datetime",
+        deserialize_with = "deserialize_datetime",
+        default
+    )]
+    pub updated_at: DateTime<Utc>,
+
+    pub source_id: String,
+    pub name: String,
+    pub description: String,
+    pub entity_type: KnowledgeEntityType,
+    #[serde(default)]
+    pub metadata: Option<serde_json::Value>,
+    pub user_id: String,
+
+    pub score: f32,
+    #[serde(default)]
+    pub highlighted_name: Option<String>,
+    #[serde(default)]
+    pub highlighted_description: Option<String>,
+}
+
 stored_object!(KnowledgeEntity, "knowledge_entity", {
    source_id: String,
    name: String,
@@ -68,6 +107,50 @@ impl KnowledgeEntity {
        }
    }

+    pub async fn search(
+        db: &SurrealDbClient,
+        search_terms: &str,
+        user_id: &str,
+        limit: usize,
+    ) -> Result<Vec<KnowledgeEntitySearchResult>, AppError> {
+        let sql = r#"
+            SELECT
+                id,
+                created_at,
+                updated_at,
+                source_id,
+                name,
+                description,
+                entity_type,
+                metadata,
+                user_id,
+                search::highlight('<b>', '</b>', 0) AS highlighted_name,
+                search::highlight('<b>', '</b>', 1) AS highlighted_description,
+                (
+                    IF search::score(0) != NONE THEN search::score(0) ELSE 0 END +
+                    IF search::score(1) != NONE THEN search::score(1) ELSE 0 END
+                ) AS score
+            FROM knowledge_entity
+            WHERE
+                (
+                    name @0@ $terms OR
+                    description @1@ $terms
+                )
+                AND user_id = $user_id
+            ORDER BY score DESC
+            LIMIT $limit;
+        "#;
+
+        Ok(db
+            .client
+            .query(sql)
+            .bind(("terms", search_terms.to_owned()))
+            .bind(("user_id", user_id.to_owned()))
+            .bind(("limit", limit))
+            .await?
+            .take(0)?)
+    }
+
    pub async fn delete_by_source_id(
        source_id: &str,
        db_client: &SurrealDbClient,
@@ -94,7 +177,9 @@ impl KnowledgeEntity {
            "name: {}, description: {}, type: {:?}",
            name, description, entity_type
        );
-        let embedding = generate_embedding(ai_client, &embedding_input).await?;
+        let embedding = generate_embedding(ai_client, &embedding_input, db_client).await?;
+
+        let now = Utc::now();

        db_client
            .client
@@ -110,7 +195,7 @@ impl KnowledgeEntity {
            .bind(("table", Self::table_name()))
            .bind(("id", id.to_string()))
            .bind(("name", name.to_string()))
-            .bind(("updated_at", Utc::now()))
+            .bind(("updated_at", surrealdb::Datetime::from(now)))
            .bind(("entity_type", entity_type.to_owned()))
            .bind(("embedding", embedding))
            .bind(("description", description.to_string()))
@@ -118,6 +203,115 @@ impl KnowledgeEntity {

        Ok(())
    }
+
+    /// Re-creates embeddings for all knowledge entities in the database.
+    ///
+    /// This is a costly operation that should be run in the background. It follows the same
+    /// pattern as the text chunk update:
+    /// 1. Re-defines the vector index with the new dimensions.
+    /// 2. Fetches all existing entities.
+    /// 3. Sequentially regenerates the embedding for each and updates the record.
+    pub async fn update_all_embeddings(
+        db: &SurrealDbClient,
+        openai_client: &Client<OpenAIConfig>,
+        new_model: &str,
+        new_dimensions: u32,
+    ) -> Result<(), AppError> {
+        info!(
+            "Starting re-embedding process for all knowledge entities. New dimensions: {}",
+            new_dimensions
+        );
+
+        // Fetch all entities first
+        let all_entities: Vec<KnowledgeEntity> = db.select(Self::table_name()).await?;
+        let total_entities = all_entities.len();
+        if total_entities == 0 {
+            info!("No knowledge entities to update. Just updating the idx");
+
+            let mut transaction_query = String::from("BEGIN TRANSACTION;");
+            transaction_query
+                .push_str("REMOVE INDEX idx_embedding_entities ON TABLE knowledge_entity;");
+            transaction_query.push_str(&format!(
+                "DEFINE INDEX idx_embedding_entities ON TABLE knowledge_entity FIELDS embedding HNSW DIMENSION {};",
+                new_dimensions
+            ));
+            transaction_query.push_str("COMMIT TRANSACTION;");
+
+            db.query(transaction_query).await?;
+            return Ok(());
+        }
+        info!("Found {} entities to process.", total_entities);
+
+        // Generate all new embeddings in memory
+        let mut new_embeddings: HashMap<String, Vec<f32>> = HashMap::new();
+        info!("Generating new embeddings for all entities...");
+        for entity in all_entities.iter() {
+            let embedding_input = format!(
+                "name: {}, description: {}, type: {:?}",
+                entity.name, entity.description, entity.entity_type
+            );
+            let retry_strategy = ExponentialBackoff::from_millis(100).map(jitter).take(3);
+
+            let embedding = Retry::spawn(retry_strategy, || {
+                crate::utils::embedding::generate_embedding_with_params(
+                    openai_client,
+                    &embedding_input,
+                    new_model,
+                    new_dimensions,
+                )
+            })
+            .await?;
+
+            // Check embedding lengths
+            if embedding.len() != new_dimensions as usize {
+                let err_msg = format!(
+                "CRITICAL: Generated embedding for entity {} has incorrect dimension ({}). Expected {}. Aborting.",
+                entity.id, embedding.len(), new_dimensions
+            );
+                error!("{}", err_msg);
+                return Err(AppError::InternalError(err_msg));
+            }
+            new_embeddings.insert(entity.id.clone(), embedding);
+        }
+        info!("Successfully generated all new embeddings.");
+
+        // Perform DB updates in a single transaction
+        info!("Applying schema and data changes in a transaction...");
+        let mut transaction_query = String::from("BEGIN TRANSACTION;");
+
+        // Add all update statements
+        for (id, embedding) in new_embeddings {
+            // We must properly serialize the vector for the SurrealQL query string
+            let embedding_str = format!(
+                "[{}]",
+                embedding
+                    .iter()
+                    .map(|f| f.to_string())
+                    .collect::<Vec<_>>()
+                    .join(",")
+            );
+            transaction_query.push_str(&format!(
+            "UPDATE type::thing('knowledge_entity', '{}') SET embedding = {}, updated_at = time::now();",
+            id, embedding_str
+        ));
+        }
+
+        // Re-create the index after updating the data that it will index
+        transaction_query
+            .push_str("REMOVE INDEX idx_embedding_entities ON TABLE knowledge_entity;");
+        transaction_query.push_str(&format!(
+        "DEFINE INDEX idx_embedding_entities ON TABLE knowledge_entity FIELDS embedding HNSW DIMENSION {};",
+        new_dimensions
+    ));
+
+        transaction_query.push_str("COMMIT TRANSACTION;");
+
+        // Execute the entire atomic operation
+        db.query(transaction_query).await?;
+
+        info!("Re-embedding process for knowledge entities completed successfully.");
+        Ok(())
+    }
 }

 #[cfg(test)]
--- a/common/src/storage/types/knowledge_relationship.rs
+++ b/common/src/storage/types/knowledge_relationship.rs
@@ -75,13 +75,36 @@ impl KnowledgeRelationship {

    pub async fn delete_relationship_by_id(
        id: &str,
+        user_id: &str,
        db_client: &SurrealDbClient,
    ) -> Result<(), AppError> {
-        let query = format!("DELETE relates_to:`{}`", id);
+        let mut authorized_result = db_client
+            .query(format!(
+                "SELECT * FROM relates_to WHERE id = relates_to:`{}` AND metadata.user_id = '{}'",
+                id, user_id
+            ))
+            .await?;
+        let authorized: Vec<KnowledgeRelationship> = authorized_result.take(0).unwrap_or_default();

-        db_client.query(query).await?;
+        if authorized.is_empty() {
+            let mut exists_result = db_client
+                .query(format!("SELECT * FROM relates_to:`{}`", id))
+                .await?;
+            let existing: Option<KnowledgeRelationship> = exists_result.take(0)?;

-        Ok(())
+            if existing.is_some() {
+                Err(AppError::Auth(
+                    "Not authorized to delete relationship".into(),
+                ))
+            } else {
+                Err(AppError::NotFound(format!("Relationship {} not found", id)))
+            }
+        } else {
+            db_client
+                .query(format!("DELETE relates_to:`{}`", id))
+                .await?;
+            Ok(())
+        }
    }
 }

@@ -161,7 +184,7 @@ mod tests {
        let relationship = KnowledgeRelationship::new(
            entity1_id.clone(),
            entity2_id.clone(),
-            user_id,
+            user_id.clone(),
            source_id.clone(),
            relationship_type,
        );
@@ -209,7 +232,7 @@ mod tests {
        let relationship = KnowledgeRelationship::new(
            entity1_id.clone(),
            entity2_id.clone(),
-            user_id,
+            user_id.clone(),
            source_id.clone(),
            relationship_type,
        );
@@ -220,20 +243,107 @@ mod tests {
            .await
            .expect("Failed to store relationship");

+        // Ensure relationship exists before deletion attempt
+        let mut existing_before_delete = db
+            .query(format!(
+                "SELECT * FROM relates_to WHERE metadata.user_id = '{}' AND metadata.source_id = '{}'",
+                user_id, source_id
+            ))
+            .await
+            .expect("Query failed");
+        let before_results: Vec<KnowledgeRelationship> =
+            existing_before_delete.take(0).unwrap_or_default();
+        assert!(
+            !before_results.is_empty(),
+            "Relationship should exist before deletion"
+        );
+
        // Delete the relationship by ID
-        KnowledgeRelationship::delete_relationship_by_id(&relationship.id, &db)
+        KnowledgeRelationship::delete_relationship_by_id(&relationship.id, &user_id, &db)
            .await
            .expect("Failed to delete relationship by ID");

        // Query to verify the relationship was deleted
-        let query = format!("SELECT * FROM relates_to WHERE id = '{}'", relationship.id);
-        let mut result = db.query(query).await.expect("Query failed");
+        let mut result = db
+            .query(format!(
+                "SELECT * FROM relates_to WHERE metadata.user_id = '{}' AND metadata.source_id = '{}'",
+                user_id, source_id
+            ))
+            .await
+            .expect("Query failed");
        let results: Vec<KnowledgeRelationship> = result.take(0).unwrap_or_default();

        // Verify the relationship no longer exists
        assert!(results.is_empty(), "Relationship should be deleted");
    }

+    #[tokio::test]
+    async fn test_delete_relationship_by_id_unauthorized() {
+        let namespace = "test_ns";
+        let database = &Uuid::new_v4().to_string();
+        let db = SurrealDbClient::memory(namespace, database)
+            .await
+            .expect("Failed to start in-memory surrealdb");
+
+        let entity1_id = create_test_entity("Entity 1", &db).await;
+        let entity2_id = create_test_entity("Entity 2", &db).await;
+
+        let owner_user_id = "owner-user".to_string();
+        let source_id = "source123".to_string();
+
+        let relationship = KnowledgeRelationship::new(
+            entity1_id.clone(),
+            entity2_id.clone(),
+            owner_user_id.clone(),
+            source_id,
+            "references".to_string(),
+        );
+
+        relationship
+            .store_relationship(&db)
+            .await
+            .expect("Failed to store relationship");
+
+        let mut before_attempt = db
+            .query(format!(
+                "SELECT * FROM relates_to WHERE metadata.user_id = '{}'",
+                owner_user_id
+            ))
+            .await
+            .expect("Query failed");
+        let before_results: Vec<KnowledgeRelationship> = before_attempt.take(0).unwrap_or_default();
+        assert!(
+            !before_results.is_empty(),
+            "Relationship should exist before unauthorized delete attempt"
+        );
+
+        let result = KnowledgeRelationship::delete_relationship_by_id(
+            &relationship.id,
+            "different-user",
+            &db,
+        )
+        .await;
+
+        match result {
+            Err(AppError::Auth(_)) => {}
+            _ => panic!("Expected authorization error when deleting someone else's relationship"),
+        }
+
+        let mut after_attempt = db
+            .query(format!(
+                "SELECT * FROM relates_to WHERE metadata.user_id = '{}'",
+                owner_user_id
+            ))
+            .await
+            .expect("Query failed");
+        let results: Vec<KnowledgeRelationship> = after_attempt.take(0).unwrap_or_default();
+
+        assert!(
+            !results.is_empty(),
+            "Relationship should still exist after unauthorized delete attempt"
+        );
+    }
+
    #[tokio::test]
    async fn test_delete_relationships_by_source_id() {
        // Setup in-memory database for testing
--- a/common/src/storage/types/mod.rs
+++ b/common/src/storage/types/mod.rs
@@ -7,6 +7,7 @@ pub mod ingestion_task;
 pub mod knowledge_entity;
 pub mod knowledge_relationship;
 pub mod message;
+pub mod scratchpad;
 pub mod system_prompts;
 pub mod system_settings;
 pub mod text_chunk;
@@ -83,6 +84,32 @@ macro_rules! stored_object {
            Ok(DateTime::<Utc>::from(dt))
        }

+        #[allow(dead_code)]
+        fn serialize_option_datetime<S>(
+            date: &Option<DateTime<Utc>>,
+            serializer: S,
+        ) -> Result<S::Ok, S::Error>
+        where
+            S: serde::Serializer,
+        {
+            match date {
+                Some(dt) => serializer
+                    .serialize_some(&Into::<surrealdb::sql::Datetime>::into(*dt)),
+                None => serializer.serialize_none(),
+            }
+        }
+
+        #[allow(dead_code)]
+        fn deserialize_option_datetime<'de, D>(
+            deserializer: D,
+        ) -> Result<Option<DateTime<Utc>>, D::Error>
+        where
+            D: serde::Deserializer<'de>,
+        {
+            let value = Option::<surrealdb::sql::Datetime>::deserialize(deserializer)?;
+            Ok(value.map(DateTime::<Utc>::from))
+        }
+

        #[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
        pub struct $name {
@@ -92,7 +119,7 @@ macro_rules! stored_object {
            pub created_at: DateTime<Utc>,
            #[serde(serialize_with = "serialize_datetime", deserialize_with = "deserialize_datetime", default)]
            pub updated_at: DateTime<Utc>,
-            $(pub $field: $ty),*
+            $( $(#[$attr])* pub $field: $ty),*
        }

        impl StoredObject for $name {
--- a/common/src/storage/types/scratchpad.rs
+++ b/common/src/storage/types/scratchpad.rs
@@ -0,0 +1,462 @@
+use chrono::Utc as ChronoUtc;
+use surrealdb::opt::PatchOp;
+use uuid::Uuid;
+
+use crate::{error::AppError, storage::db::SurrealDbClient, stored_object};
+
+stored_object!(Scratchpad, "scratchpad", {
+    user_id: String,
+    title: String,
+    content: String,
+    #[serde(serialize_with = "serialize_datetime", deserialize_with="deserialize_datetime")]
+    last_saved_at: DateTime<Utc>,
+    is_dirty: bool,
+    #[serde(default)]
+    is_archived: bool,
+    #[serde(
+        serialize_with = "serialize_option_datetime",
+        deserialize_with = "deserialize_option_datetime",
+        default
+    )]
+    archived_at: Option<DateTime<Utc>>,
+    #[serde(
+        serialize_with = "serialize_option_datetime",
+        deserialize_with = "deserialize_option_datetime",
+        default
+    )]
+    ingested_at: Option<DateTime<Utc>>
+});
+
+impl Scratchpad {
+    pub fn new(user_id: String, title: String) -> Self {
+        let now = ChronoUtc::now();
+        Self {
+            id: Uuid::new_v4().to_string(),
+            created_at: now,
+            updated_at: now,
+            user_id,
+            title,
+            content: String::new(),
+            last_saved_at: now,
+            is_dirty: false,
+            is_archived: false,
+            archived_at: None,
+            ingested_at: None,
+        }
+    }
+
+    pub async fn get_by_user(user_id: &str, db: &SurrealDbClient) -> Result<Vec<Self>, AppError> {
+        let scratchpads: Vec<Scratchpad> = db.client
+            .query("SELECT * FROM type::table($table_name) WHERE user_id = $user_id AND (is_archived = false OR is_archived IS NONE) ORDER BY updated_at DESC")
+            .bind(("table_name", Self::table_name()))
+            .bind(("user_id", user_id.to_string()))
+            .await?
+            .take(0)?;
+
+        Ok(scratchpads)
+    }
+
+    pub async fn get_archived_by_user(
+        user_id: &str,
+        db: &SurrealDbClient,
+    ) -> Result<Vec<Self>, AppError> {
+        let scratchpads: Vec<Scratchpad> = db.client
+            .query("SELECT * FROM type::table($table_name) WHERE user_id = $user_id AND is_archived = true ORDER BY archived_at DESC, updated_at DESC")
+            .bind(("table_name", Self::table_name()))
+            .bind(("user_id", user_id.to_string()))
+            .await?
+            .take(0)?;
+
+        Ok(scratchpads)
+    }
+
+    pub async fn get_by_id(
+        id: &str,
+        user_id: &str,
+        db: &SurrealDbClient,
+    ) -> Result<Self, AppError> {
+        let scratchpad: Option<Scratchpad> = db.get_item(id).await?;
+
+        let scratchpad =
+            scratchpad.ok_or_else(|| AppError::NotFound("Scratchpad not found".to_string()))?;
+
+        if scratchpad.user_id != user_id {
+            return Err(AppError::Auth(
+                "You don't have access to this scratchpad".to_string(),
+            ));
+        }
+
+        Ok(scratchpad)
+    }
+
+    pub async fn update_content(
+        id: &str,
+        user_id: &str,
+        new_content: &str,
+        db: &SurrealDbClient,
+    ) -> Result<Self, AppError> {
+        // First verify ownership
+        let scratchpad = Self::get_by_id(id, user_id, db).await?;
+
+        if scratchpad.is_archived {
+            return Ok(scratchpad);
+        }
+
+        let now = ChronoUtc::now();
+        let _updated: Option<Self> = db
+            .update((Self::table_name(), id))
+            .patch(PatchOp::replace("/content", new_content.to_string()))
+            .patch(PatchOp::replace(
+                "/updated_at",
+                surrealdb::Datetime::from(now),
+            ))
+            .patch(PatchOp::replace(
+                "/last_saved_at",
+                surrealdb::Datetime::from(now),
+            ))
+            .patch(PatchOp::replace("/is_dirty", false))
+            .await?;
+
+        // Return the updated scratchpad
+        Self::get_by_id(id, user_id, db).await
+    }
+
+    pub async fn update_title(
+        id: &str,
+        user_id: &str,
+        new_title: &str,
+        db: &SurrealDbClient,
+    ) -> Result<(), AppError> {
+        // First verify ownership
+        let _scratchpad = Self::get_by_id(id, user_id, db).await?;
+
+        let _updated: Option<Self> = db
+            .update((Self::table_name(), id))
+            .patch(PatchOp::replace("/title", new_title.to_string()))
+            .patch(PatchOp::replace(
+                "/updated_at",
+                surrealdb::Datetime::from(ChronoUtc::now()),
+            ))
+            .await?;
+
+        Ok(())
+    }
+
+    pub async fn delete(id: &str, user_id: &str, db: &SurrealDbClient) -> Result<(), AppError> {
+        // First verify ownership
+        let _scratchpad = Self::get_by_id(id, user_id, db).await?;
+
+        let _: Option<Self> = db.client.delete((Self::table_name(), id)).await?;
+
+        Ok(())
+    }
+
+    pub async fn archive(
+        id: &str,
+        user_id: &str,
+        db: &SurrealDbClient,
+        mark_ingested: bool,
+    ) -> Result<Self, AppError> {
+        // Verify ownership
+        let scratchpad = Self::get_by_id(id, user_id, db).await?;
+
+        if scratchpad.is_archived {
+            if mark_ingested && scratchpad.ingested_at.is_none() {
+                // Ensure ingested_at is set if required
+                let surreal_now = surrealdb::Datetime::from(ChronoUtc::now());
+                let _updated: Option<Self> = db
+                    .update((Self::table_name(), id))
+                    .patch(PatchOp::replace("/ingested_at", surreal_now))
+                    .await?;
+                return Self::get_by_id(id, user_id, db).await;
+            }
+            return Ok(scratchpad);
+        }
+
+        let now = ChronoUtc::now();
+        let surreal_now = surrealdb::Datetime::from(now);
+        let mut update = db
+            .update((Self::table_name(), id))
+            .patch(PatchOp::replace("/is_archived", true))
+            .patch(PatchOp::replace("/archived_at", surreal_now.clone()))
+            .patch(PatchOp::replace("/updated_at", surreal_now.clone()));
+
+        update = if mark_ingested {
+            update.patch(PatchOp::replace("/ingested_at", surreal_now))
+        } else {
+            update.patch(PatchOp::remove("/ingested_at"))
+        };
+
+        let _updated: Option<Self> = update.await?;
+
+        Self::get_by_id(id, user_id, db).await
+    }
+
+    pub async fn restore(id: &str, user_id: &str, db: &SurrealDbClient) -> Result<Self, AppError> {
+        // Verify ownership
+        let scratchpad = Self::get_by_id(id, user_id, db).await?;
+
+        if !scratchpad.is_archived {
+            return Ok(scratchpad);
+        }
+
+        let now = ChronoUtc::now();
+        let surreal_now = surrealdb::Datetime::from(now);
+        let _updated: Option<Self> = db
+            .update((Self::table_name(), id))
+            .patch(PatchOp::replace("/is_archived", false))
+            .patch(PatchOp::remove("/archived_at"))
+            .patch(PatchOp::remove("/ingested_at"))
+            .patch(PatchOp::replace("/updated_at", surreal_now))
+            .await?;
+
+        Self::get_by_id(id, user_id, db).await
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[tokio::test]
+    async fn test_create_scratchpad() {
+        // Setup in-memory database for testing
+        let namespace = "test_ns";
+        let database = &Uuid::new_v4().to_string();
+        let db = SurrealDbClient::memory(namespace, database)
+            .await
+            .expect("Failed to start in-memory surrealdb");
+
+        db.apply_migrations()
+            .await
+            .expect("Failed to apply migrations");
+
+        // Create a new scratchpad
+        let user_id = "test_user";
+        let title = "Test Scratchpad";
+        let scratchpad = Scratchpad::new(user_id.to_string(), title.to_string());
+
+        // Verify scratchpad properties
+        assert_eq!(scratchpad.user_id, user_id);
+        assert_eq!(scratchpad.title, title);
+        assert_eq!(scratchpad.content, "");
+        assert!(!scratchpad.is_dirty);
+        assert!(!scratchpad.is_archived);
+        assert!(scratchpad.archived_at.is_none());
+        assert!(scratchpad.ingested_at.is_none());
+        assert!(!scratchpad.id.is_empty());
+
+        // Store the scratchpad
+        let result = db.store_item(scratchpad.clone()).await;
+        assert!(result.is_ok());
+
+        // Verify it can be retrieved
+        let retrieved: Option<Scratchpad> = db
+            .get_item(&scratchpad.id)
+            .await
+            .expect("Failed to retrieve scratchpad");
+        assert!(retrieved.is_some());
+
+        let retrieved = retrieved.unwrap();
+        assert_eq!(retrieved.id, scratchpad.id);
+        assert_eq!(retrieved.user_id, user_id);
+        assert_eq!(retrieved.title, title);
+        assert!(!retrieved.is_archived);
+        assert!(retrieved.archived_at.is_none());
+        assert!(retrieved.ingested_at.is_none());
+    }
+
+    #[tokio::test]
+    async fn test_get_by_user() {
+        let namespace = "test_ns";
+        let database = &Uuid::new_v4().to_string();
+        let db = SurrealDbClient::memory(namespace, database)
+            .await
+            .expect("Failed to start in-memory surrealdb");
+
+        db.apply_migrations()
+            .await
+            .expect("Failed to apply migrations");
+
+        let user_id = "test_user";
+
+        // Create multiple scratchpads
+        let scratchpad1 = Scratchpad::new(user_id.to_string(), "First".to_string());
+        let scratchpad2 = Scratchpad::new(user_id.to_string(), "Second".to_string());
+        let scratchpad3 = Scratchpad::new("other_user".to_string(), "Other".to_string());
+
+        // Store them
+        let scratchpad1_id = scratchpad1.id.clone();
+        let scratchpad2_id = scratchpad2.id.clone();
+        db.store_item(scratchpad1).await.unwrap();
+        db.store_item(scratchpad2).await.unwrap();
+        db.store_item(scratchpad3).await.unwrap();
+
+        // Archive one of the user's scratchpads
+        Scratchpad::archive(&scratchpad2_id, user_id, &db, false)
+            .await
+            .unwrap();
+
+        // Get scratchpads for user_id
+        let user_scratchpads = Scratchpad::get_by_user(user_id, &db).await.unwrap();
+        assert_eq!(user_scratchpads.len(), 1);
+        assert_eq!(user_scratchpads[0].id, scratchpad1_id);
+
+        // Verify they belong to the user
+        for scratchpad in &user_scratchpads {
+            assert_eq!(scratchpad.user_id, user_id);
+        }
+
+        let archived = Scratchpad::get_archived_by_user(user_id, &db)
+            .await
+            .unwrap();
+        assert_eq!(archived.len(), 1);
+        assert_eq!(archived[0].id, scratchpad2_id);
+        assert!(archived[0].is_archived);
+        assert!(archived[0].ingested_at.is_none());
+    }
+
+    #[tokio::test]
+    async fn test_archive_and_restore() {
+        let namespace = "test_ns";
+        let database = &Uuid::new_v4().to_string();
+        let db = SurrealDbClient::memory(namespace, database)
+            .await
+            .expect("Failed to start in-memory surrealdb");
+
+        db.apply_migrations()
+            .await
+            .expect("Failed to apply migrations");
+
+        let user_id = "test_user";
+        let scratchpad = Scratchpad::new(user_id.to_string(), "Test".to_string());
+        let scratchpad_id = scratchpad.id.clone();
+        db.store_item(scratchpad).await.unwrap();
+
+        let archived = Scratchpad::archive(&scratchpad_id, user_id, &db, true)
+            .await
+            .expect("Failed to archive");
+        assert!(archived.is_archived);
+        assert!(archived.archived_at.is_some());
+        assert!(archived.ingested_at.is_some());
+
+        let restored = Scratchpad::restore(&scratchpad_id, user_id, &db)
+            .await
+            .expect("Failed to restore");
+        assert!(!restored.is_archived);
+        assert!(restored.archived_at.is_none());
+        assert!(restored.ingested_at.is_none());
+    }
+
+    #[tokio::test]
+    async fn test_update_content() {
+        let namespace = "test_ns";
+        let database = &Uuid::new_v4().to_string();
+        let db = SurrealDbClient::memory(namespace, database)
+            .await
+            .expect("Failed to start in-memory surrealdb");
+
+        db.apply_migrations()
+            .await
+            .expect("Failed to apply migrations");
+
+        let user_id = "test_user";
+        let scratchpad = Scratchpad::new(user_id.to_string(), "Test".to_string());
+        let scratchpad_id = scratchpad.id.clone();
+
+        db.store_item(scratchpad).await.unwrap();
+
+        let new_content = "Updated content";
+        let updated = Scratchpad::update_content(&scratchpad_id, user_id, new_content, &db)
+            .await
+            .unwrap();
+
+        assert_eq!(updated.content, new_content);
+        assert!(!updated.is_dirty);
+    }
+
+    #[tokio::test]
+    async fn test_update_content_unauthorized() {
+        let namespace = "test_ns";
+        let database = &Uuid::new_v4().to_string();
+        let db = SurrealDbClient::memory(namespace, database)
+            .await
+            .expect("Failed to start in-memory surrealdb");
+
+        db.apply_migrations()
+            .await
+            .expect("Failed to apply migrations");
+
+        let owner_id = "owner";
+        let other_user = "other_user";
+        let scratchpad = Scratchpad::new(owner_id.to_string(), "Test".to_string());
+        let scratchpad_id = scratchpad.id.clone();
+
+        db.store_item(scratchpad).await.unwrap();
+
+        let result = Scratchpad::update_content(&scratchpad_id, other_user, "Hacked", &db).await;
+        assert!(result.is_err());
+        match result {
+            Err(AppError::Auth(_)) => {}
+            _ => panic!("Expected Auth error"),
+        }
+    }
+
+    #[tokio::test]
+    async fn test_delete_scratchpad() {
+        let namespace = "test_ns";
+        let database = &Uuid::new_v4().to_string();
+        let db = SurrealDbClient::memory(namespace, database)
+            .await
+            .expect("Failed to start in-memory surrealdb");
+
+        db.apply_migrations()
+            .await
+            .expect("Failed to apply migrations");
+
+        let user_id = "test_user";
+        let scratchpad = Scratchpad::new(user_id.to_string(), "Test".to_string());
+        let scratchpad_id = scratchpad.id.clone();
+
+        db.store_item(scratchpad).await.unwrap();
+
+        // Delete should succeed
+        let result = Scratchpad::delete(&scratchpad_id, user_id, &db).await;
+        assert!(result.is_ok());
+
+        // Verify it's gone
+        let retrieved: Option<Scratchpad> = db.get_item(&scratchpad_id).await.unwrap();
+        assert!(retrieved.is_none());
+    }
+
+    #[tokio::test]
+    async fn test_delete_unauthorized() {
+        let namespace = "test_ns";
+        let database = &Uuid::new_v4().to_string();
+        let db = SurrealDbClient::memory(namespace, database)
+            .await
+            .expect("Failed to start in-memory surrealdb");
+
+        db.apply_migrations()
+            .await
+            .expect("Failed to apply migrations");
+
+        let owner_id = "owner";
+        let other_user = "other_user";
+        let scratchpad = Scratchpad::new(owner_id.to_string(), "Test".to_string());
+        let scratchpad_id = scratchpad.id.clone();
+
+        db.store_item(scratchpad).await.unwrap();
+
+        let result = Scratchpad::delete(&scratchpad_id, other_user, &db).await;
+        assert!(result.is_err());
+        match result {
+            Err(AppError::Auth(_)) => {}
+            _ => panic!("Expected Auth error"),
+        }
+
+        // Verify it still exists
+        let retrieved: Option<Scratchpad> = db.get_item(&scratchpad_id).await.unwrap();
+        assert!(retrieved.is_some());
+    }
+}
--- a/common/src/storage/types/system_prompts.rs
+++ b/common/src/storage/types/system_prompts.rs
@@ -54,3 +54,10 @@ Guidelines:
 7. Only create relationships between existing KnowledgeEntities.
 8. Entities that exist already in the database should NOT be created again. If there is only a minor overlap, skip creating a new entity.
 9. A new relationship MUST include a newly created KnowledgeEntity."#;
+
+pub static DEFAULT_IMAGE_PROCESSING_PROMPT: &str = r#"Analyze this image and respond based on its primary content:
+- If the image is mainly text (document, screenshot, sign), transcribe the text verbatim.
+- If the image is mainly visual (photograph, art, landscape), provide a concise description of the scene.
+- For hybrid images (diagrams, ads), briefly describe the visual, then transcribe the text under a "Text:" heading.
+
+Respond directly with the analysis."#;
--- a/common/src/storage/types/system_settings.rs
+++ b/common/src/storage/types/system_settings.rs
@@ -11,8 +11,13 @@ pub struct SystemSettings {
    pub require_email_verification: bool,
    pub query_model: String,
    pub processing_model: String,
+    pub embedding_model: String,
+    pub embedding_dimensions: u32,
    pub query_system_prompt: String,
    pub ingestion_system_prompt: String,
+    pub image_processing_model: String,
+    pub image_processing_prompt: String,
+    pub voice_processing_model: String,
 }

 impl StoredObject for SystemSettings {
@@ -44,28 +49,64 @@ impl SystemSettings {
            "Something went wrong updating the settings".into(),
        ))
    }
-
-    pub fn new() -> Self {
-        Self {
-            id: "current".to_string(),
-            query_system_prompt: crate::storage::types::system_prompts::DEFAULT_QUERY_SYSTEM_PROMPT
-                .to_string(),
-            ingestion_system_prompt:
-                crate::storage::types::system_prompts::DEFAULT_INGRESS_ANALYSIS_SYSTEM_PROMPT
-                    .to_string(),
-            query_model: "gpt-4o-mini".to_string(),
-            processing_model: "gpt-4o-mini".to_string(),
-            registrations_enabled: true,
-            require_email_verification: false,
-        }
-    }
 }

 #[cfg(test)]
 mod tests {
+    use crate::storage::types::{knowledge_entity::KnowledgeEntity, text_chunk::TextChunk};
+    use async_openai::Client;
+
    use super::*;
    use uuid::Uuid;

+    async fn get_hnsw_index_dimension(
+        db: &SurrealDbClient,
+        table_name: &str,
+        index_name: &str,
+    ) -> u32 {
+        let query = format!("INFO FOR TABLE {table_name};");
+        let mut response = db
+            .client
+            .query(query)
+            .await
+            .expect("Failed to fetch table info");
+
+        let info: Option<serde_json::Value> = response
+            .take(0)
+            .expect("Failed to extract table info response");
+
+        let info = info.expect("Table info result missing");
+
+        let indexes = info
+            .get("indexes")
+            .or_else(|| {
+                info.get("tables")
+                    .and_then(|tables| tables.get(table_name))
+                    .and_then(|table| table.get("indexes"))
+            })
+            .unwrap_or_else(|| panic!("Indexes collection missing in table info: {info:#?}"));
+
+        let definition = indexes
+            .get(index_name)
+            .and_then(|definition| definition.as_str())
+            .unwrap_or_else(|| panic!("Index definition not found in table info: {info:#?}"));
+
+        let dimension_part = definition
+            .split("DIMENSION")
+            .nth(1)
+            .expect("Index definition missing DIMENSION clause");
+
+        let dimension_token = dimension_part
+            .split_whitespace()
+            .next()
+            .expect("Dimension value missing in definition")
+            .trim_end_matches(';');
+
+        dimension_token
+            .parse::<u32>()
+            .expect("Dimension value is not a valid number")
+    }
+
    #[tokio::test]
    async fn test_settings_initialization() {
        // Setup in-memory database for testing
@@ -89,6 +130,7 @@ mod tests {
        assert_eq!(settings.require_email_verification, false);
        assert_eq!(settings.query_model, "gpt-4o-mini");
        assert_eq!(settings.processing_model, "gpt-4o-mini");
+        assert_eq!(settings.image_processing_model, "gpt-4o-mini");
        // Dont test these for now, having a hard time getting the formatting exactly the same
        // assert_eq!(
        //     settings.query_system_prompt,
@@ -157,7 +199,7 @@ mod tests {
            .expect("Failed to apply migrations");

        // Create updated settings
-        let mut updated_settings = SystemSettings::new();
+        let mut updated_settings = SystemSettings::get_current(&db).await.unwrap();
        updated_settings.id = "current".to_string();
        updated_settings.registrations_enabled = false;
        updated_settings.require_email_verification = true;
@@ -206,21 +248,130 @@ mod tests {
    }

    #[tokio::test]
-    async fn test_new_method() {
-        let settings = SystemSettings::new();
+    async fn test_migration_after_changing_embedding_length() {
+        let db = SurrealDbClient::memory("test", &Uuid::new_v4().to_string())
+            .await
+            .expect("Failed to start DB");
+
+        // Apply initial migrations. This sets up the text_chunk index with DIMENSION 1536.
+        db.apply_migrations()
+            .await
+            .expect("Initial migration failed");
+
+        let initial_chunk = TextChunk::new(
+            "source1".into(),
+            "This chunk has the original dimension".into(),
+            vec![0.1; 1536],
+            "user1".into(),
+        );
+
+        db.store_item(initial_chunk.clone())
+            .await
+            .expect("Failed to store initial chunk");
+
+        async fn simulate_reembedding(
+            db: &SurrealDbClient,
+            target_dimension: usize,
+            initial_chunk: TextChunk,
+        ) {
+            db.query("REMOVE INDEX idx_embedding_chunks ON TABLE text_chunk;")
+                .await
+                .unwrap();
+            let define_index_query = format!(
+                         "DEFINE INDEX idx_embedding_chunks ON TABLE text_chunk FIELDS embedding HNSW DIMENSION {};",
+                         target_dimension
+                     );
+            db.query(define_index_query)
+                .await
+                .expect("Re-defining index should succeed");
+
+            let new_embedding = vec![0.5; target_dimension];
+            let sql = "UPDATE type::thing('text_chunk', $id) SET embedding = $embedding;";
+
+            let update_result = db
+                .client
+                .query(sql)
+                .bind(("id", initial_chunk.id.clone()))
+                .bind(("embedding", new_embedding))
+                .await;
+
+            assert!(update_result.is_ok());
+        }
+
+        simulate_reembedding(&db, 768, initial_chunk).await;
+
+        let migration_result = db.apply_migrations().await;
+
+        assert!(migration_result.is_ok(), "Migrations should not fail");
+    }
+
+    #[tokio::test]
+    async fn test_should_change_embedding_length_on_indexes_when_switching_length() {
+        let db = SurrealDbClient::memory("test", &Uuid::new_v4().to_string())
+            .await
+            .expect("Failed to start DB");
+
+        // Apply initial migrations. This sets up the text_chunk index with DIMENSION 1536.
+        db.apply_migrations()
+            .await
+            .expect("Initial migration failed");
+
+        let mut current_settings = SystemSettings::get_current(&db)
+            .await
+            .expect("Failed to load current settings");
+
+        let initial_chunk_dimension =
+            get_hnsw_index_dimension(&db, "text_chunk", "idx_embedding_chunks").await;

-        assert!(settings.id.len() > 0);
-        assert_eq!(settings.registrations_enabled, true);
-        assert_eq!(settings.require_email_verification, false);
-        assert_eq!(settings.query_model, "gpt-4o-mini");
-        assert_eq!(settings.processing_model, "gpt-4o-mini");
        assert_eq!(
-            settings.query_system_prompt,
-            crate::storage::types::system_prompts::DEFAULT_QUERY_SYSTEM_PROMPT
+            initial_chunk_dimension, current_settings.embedding_dimensions,
+            "embedding size should match initial system settings"
+        );
+
+        let new_dimension = 768;
+        let new_model = "new-test-embedding-model".to_string();
+
+        current_settings.embedding_dimensions = new_dimension;
+        current_settings.embedding_model = new_model.clone();
+
+        let updated_settings = SystemSettings::update(&db, current_settings)
+            .await
+            .expect("Failed to update settings");
+
+        assert_eq!(
+            updated_settings.embedding_dimensions, new_dimension,
+            "Settings should reflect the new embedding dimension"
+        );
+
+        let openai_client = Client::new();
+
+        TextChunk::update_all_embeddings(&db, &openai_client, &new_model, new_dimension)
+            .await
+            .expect("TextChunk re-embedding should succeed on fresh DB");
+        KnowledgeEntity::update_all_embeddings(&db, &openai_client, &new_model, new_dimension)
+            .await
+            .expect("KnowledgeEntity re-embedding should succeed on fresh DB");
+
+        let text_chunk_dimension =
+            get_hnsw_index_dimension(&db, "text_chunk", "idx_embedding_chunks").await;
+        let knowledge_dimension =
+            get_hnsw_index_dimension(&db, "knowledge_entity", "idx_embedding_entities").await;
+
+        assert_eq!(
+            text_chunk_dimension, new_dimension,
+            "text_chunk index dimension should update"
        );
        assert_eq!(
-            settings.ingestion_system_prompt,
-            crate::storage::types::system_prompts::DEFAULT_INGRESS_ANALYSIS_SYSTEM_PROMPT
+            knowledge_dimension, new_dimension,
+            "knowledge_entity index dimension should update"
+        );
+
+        let persisted_settings = SystemSettings::get_current(&db)
+            .await
+            .expect("Failed to reload updated settings");
+        assert_eq!(
+            persisted_settings.embedding_dimensions, new_dimension,
+            "Settings should persist new embedding dimension"
        );
    }
 }
--- a/common/src/storage/types/text_chunk.rs
+++ b/common/src/storage/types/text_chunk.rs
@@ -1,4 +1,13 @@
+use std::collections::HashMap;
+
 use crate::{error::AppError, storage::db::SurrealDbClient, stored_object};
+use async_openai::{config::OpenAIConfig, Client};
+use tokio_retry::{
+    strategy::{jitter, ExponentialBackoff},
+    Retry,
+};
+
+use tracing::{error, info};
 use uuid::Uuid;

 stored_object!(TextChunk, "text_chunk", {
@@ -35,6 +44,109 @@ impl TextChunk {

        Ok(())
    }
+
+    /// Re-creates embeddings for all text chunks using a safe, atomic transaction.
+    ///
+    /// This is a costly operation that should be run in the background. It performs these steps:
+    /// 1. **Fetches All Chunks**: Loads all existing text_chunk records into memory.
+    /// 2. **Generates All Embeddings**: Creates new embeddings for every chunk. If any fails or
+    ///    has the wrong dimension, the entire operation is aborted before any DB changes are made.
+    /// 3. **Executes Atomic Transaction**: All data updates and the index recreation are
+    ///    performed in a single, all-or-nothing database transaction.
+    pub async fn update_all_embeddings(
+        db: &SurrealDbClient,
+        openai_client: &Client<OpenAIConfig>,
+        new_model: &str,
+        new_dimensions: u32,
+    ) -> Result<(), AppError> {
+        info!(
+            "Starting re-embedding process for all text chunks. New dimensions: {}",
+            new_dimensions
+        );
+
+        // Fetch all chunks first
+        let all_chunks: Vec<TextChunk> = db.select(Self::table_name()).await?;
+        let total_chunks = all_chunks.len();
+        if total_chunks == 0 {
+            info!("No text chunks to update. Just updating the idx");
+
+            let mut transaction_query = String::from("BEGIN TRANSACTION;");
+            transaction_query.push_str("REMOVE INDEX idx_embedding_chunks ON TABLE text_chunk;");
+            transaction_query.push_str(&format!(
+            "DEFINE INDEX idx_embedding_chunks ON TABLE text_chunk FIELDS embedding HNSW DIMENSION {};",
+            new_dimensions));
+            transaction_query.push_str("COMMIT TRANSACTION;");
+
+            db.query(transaction_query).await?;
+
+            return Ok(());
+        }
+        info!("Found {} chunks to process.", total_chunks);
+
+        // Generate all new embeddings in memory
+        let mut new_embeddings: HashMap<String, Vec<f32>> = HashMap::new();
+        info!("Generating new embeddings for all chunks...");
+        for chunk in all_chunks.iter() {
+            let retry_strategy = ExponentialBackoff::from_millis(100).map(jitter).take(3);
+
+            let embedding = Retry::spawn(retry_strategy, || {
+                crate::utils::embedding::generate_embedding_with_params(
+                    openai_client,
+                    &chunk.chunk,
+                    new_model,
+                    new_dimensions,
+                )
+            })
+            .await?;
+
+            // Safety check: ensure the generated embedding has the correct dimension.
+            if embedding.len() != new_dimensions as usize {
+                let err_msg = format!(
+                    "CRITICAL: Generated embedding for chunk {} has incorrect dimension ({}). Expected {}. Aborting.",
+                    chunk.id, embedding.len(), new_dimensions
+                );
+                error!("{}", err_msg);
+                return Err(AppError::InternalError(err_msg));
+            }
+            new_embeddings.insert(chunk.id.clone(), embedding);
+        }
+        info!("Successfully generated all new embeddings.");
+
+        // Perform DB updates in a single transaction
+        info!("Applying schema and data changes in a transaction...");
+        let mut transaction_query = String::from("BEGIN TRANSACTION;");
+
+        // Add all update statements
+        for (id, embedding) in new_embeddings {
+            let embedding_str = format!(
+                "[{}]",
+                embedding
+                    .iter()
+                    .map(|f| f.to_string())
+                    .collect::<Vec<_>>()
+                    .join(",")
+            );
+            transaction_query.push_str(&format!(
+                "UPDATE type::thing('text_chunk', '{}') SET embedding = {}, updated_at = time::now();",
+                id, embedding_str
+            ));
+        }
+
+        // Re-create the index inside the same transaction
+        transaction_query.push_str("REMOVE INDEX idx_embedding_chunks ON TABLE text_chunk;");
+        transaction_query.push_str(&format!(
+            "DEFINE INDEX idx_embedding_chunks ON TABLE text_chunk FIELDS embedding HNSW DIMENSION {};",
+            new_dimensions
+        ));
+
+        transaction_query.push_str("COMMIT TRANSACTION;");
+
+        // Execute the entire atomic operation
+        db.query(transaction_query).await?;
+
+        info!("Re-embedding process for text chunks completed successfully.");
+        Ok(())
+    }
 }

 #[cfg(test)]
--- a/common/src/storage/types/text_content.rs
+++ b/common/src/storage/types/text_content.rs
@@ -101,12 +101,35 @@ impl TextContent {
            .patch(PatchOp::replace("/context", context))
            .patch(PatchOp::replace("/category", category))
            .patch(PatchOp::replace("/text", text))
-            .patch(PatchOp::replace("/updated_at", now))
+            .patch(PatchOp::replace(
+                "/updated_at",
+                surrealdb::Datetime::from(now),
+            ))
            .await?;

        Ok(())
    }

+    pub async fn has_other_with_file(
+        file_id: &str,
+        exclude_id: &str,
+        db: &SurrealDbClient,
+    ) -> Result<bool, AppError> {
+        let mut response = db
+            .client
+            .query(
+                "SELECT VALUE id FROM type::table($table_name) WHERE file_info.id = $file_id AND id != type::thing($table_name, $exclude_id) LIMIT 1",
+            )
+            .bind(("table_name", TextContent::table_name()))
+            .bind(("file_id", file_id.to_owned()))
+            .bind(("exclude_id", exclude_id.to_owned()))
+            .await?;
+
+        let existing: Option<surrealdb::sql::Thing> = response.take(0)?;
+
+        Ok(existing.is_some())
+    }
+
    pub async fn search(
        db: &SurrealDbClient,
        search_terms: &str,
@@ -273,4 +296,64 @@ mod tests {
        assert_eq!(updated_content.text, new_text);
        assert!(updated_content.updated_at > text_content.updated_at);
    }
+
+    #[tokio::test]
+    async fn test_has_other_with_file_detects_shared_usage() {
+        let namespace = "test_ns";
+        let database = &Uuid::new_v4().to_string();
+        let db = SurrealDbClient::memory(namespace, database)
+            .await
+            .expect("Failed to start in-memory surrealdb");
+
+        let user_id = "user123".to_string();
+        let file_info = FileInfo {
+            id: "file-1".to_string(),
+            created_at: chrono::Utc::now(),
+            updated_at: chrono::Utc::now(),
+            sha256: "sha-test".to_string(),
+            path: "user123/file-1/test.txt".to_string(),
+            file_name: "test.txt".to_string(),
+            mime_type: "text/plain".to_string(),
+            user_id: user_id.clone(),
+        };
+
+        let content_a = TextContent::new(
+            "First".to_string(),
+            Some("ctx-a".to_string()),
+            "category".to_string(),
+            Some(file_info.clone()),
+            None,
+            user_id.clone(),
+        );
+        let content_b = TextContent::new(
+            "Second".to_string(),
+            Some("ctx-b".to_string()),
+            "category".to_string(),
+            Some(file_info.clone()),
+            None,
+            user_id.clone(),
+        );
+
+        db.store_item(content_a.clone())
+            .await
+            .expect("Failed to store first content");
+        db.store_item(content_b.clone())
+            .await
+            .expect("Failed to store second content");
+
+        let has_other = TextContent::has_other_with_file(&file_info.id, &content_a.id, &db)
+            .await
+            .expect("Failed to check for shared file usage");
+        assert!(has_other);
+
+        let _removed: Option<TextContent> = db
+            .delete_item(&content_b.id)
+            .await
+            .expect("Failed to delete second content");
+
+        let has_other_after = TextContent::has_other_with_file(&file_info.id, &content_a.id, &db)
+            .await
+            .expect("Failed to check shared usage after delete");
+        assert!(!has_other_after);
+    }
 }
--- a/common/src/storage/types/user.rs
+++ b/common/src/storage/types/user.rs
@@ -1,14 +1,21 @@
 use crate::{error::AppError, storage::db::SurrealDbClient, stored_object};
 use async_trait::async_trait;
 use axum_session_auth::Authentication;
+use chrono_tz::Tz;
 use surrealdb::{engine::any::Any, Surreal};
 use uuid::Uuid;

+use super::text_chunk::TextChunk;
 use super::{
-    conversation::Conversation, ingestion_task::IngestionTask, knowledge_entity::KnowledgeEntity,
-    knowledge_relationship::KnowledgeRelationship, system_settings::SystemSettings,
+    conversation::Conversation,
+    ingestion_task::{IngestionTask, TaskState},
+    knowledge_entity::{KnowledgeEntity, KnowledgeEntityType},
+    knowledge_relationship::KnowledgeRelationship,
+    system_settings::SystemSettings,
    text_content::TextContent,
 };
+use chrono::Duration;
+use futures::try_join;

 #[derive(Deserialize)]
 pub struct CategoryResponse {
@@ -49,9 +56,6 @@ impl Authentication<User, String, Surreal<Any>> for User {
 }

 fn validate_timezone(input: &str) -> String {
-    use chrono_tz::Tz;
-
-    // Check if it's a valid IANA timezone identifier
    match input.parse::<Tz>() {
        Ok(_) => input.to_owned(),
        Err(_) => {
@@ -61,7 +65,93 @@ fn validate_timezone(input: &str) -> String {
    }
 }

+#[derive(Serialize, Deserialize, Debug, Clone)]
+pub struct DashboardStats {
+    pub total_documents: i64,
+    pub new_documents_week: i64,
+    pub total_entities: i64,
+    pub new_entities_week: i64,
+    pub total_conversations: i64,
+    pub new_conversations_week: i64,
+    pub total_text_chunks: i64,
+    pub new_text_chunks_week: i64,
+}
+
+#[derive(Deserialize)]
+struct CountResult {
+    count: i64,
+}
+
 impl User {
+    async fn count_total<T: crate::storage::types::StoredObject>(
+        db: &SurrealDbClient,
+        user_id: &str,
+    ) -> Result<i64, AppError> {
+        let result: Option<CountResult> = db
+            .client
+            .query("SELECT count() as count FROM type::table($table) WHERE user_id = $user_id GROUP ALL")
+            .bind(("table", T::table_name()))
+            .bind(("user_id", user_id.to_string()))
+            .await?
+            .take(0)?;
+        Ok(result.map(|r| r.count).unwrap_or(0))
+    }
+
+    async fn count_since<T: crate::storage::types::StoredObject>(
+        db: &SurrealDbClient,
+        user_id: &str,
+        since: chrono::DateTime<chrono::Utc>,
+    ) -> Result<i64, AppError> {
+        let result: Option<CountResult> = db
+            .client
+            .query(
+                "SELECT count() as count FROM type::table($table) WHERE user_id = $user_id AND created_at >= $since GROUP ALL",
+            )
+            .bind(("table", T::table_name()))
+            .bind(("user_id", user_id.to_string()))
+            .bind(("since", surrealdb::Datetime::from(since)))
+            .await?
+            .take(0)?;
+        Ok(result.map(|r| r.count).unwrap_or(0))
+    }
+
+    pub async fn get_dashboard_stats(
+        user_id: &str,
+        db: &SurrealDbClient,
+    ) -> Result<DashboardStats, AppError> {
+        let since = chrono::Utc::now() - Duration::days(7);
+
+        let (
+            total_documents,
+            new_documents_week,
+            total_entities,
+            new_entities_week,
+            total_conversations,
+            new_conversations_week,
+            total_text_chunks,
+            new_text_chunks_week,
+        ) = try_join!(
+            Self::count_total::<TextContent>(db, user_id),
+            Self::count_since::<TextContent>(db, user_id, since),
+            Self::count_total::<KnowledgeEntity>(db, user_id),
+            Self::count_since::<KnowledgeEntity>(db, user_id, since),
+            Self::count_total::<Conversation>(db, user_id),
+            Self::count_since::<Conversation>(db, user_id, since),
+            Self::count_total::<TextChunk>(db, user_id),
+            Self::count_since::<TextChunk>(db, user_id, since)
+        )?;
+
+        Ok(DashboardStats {
+            total_documents,
+            new_documents_week,
+            total_entities,
+            new_entities_week,
+            total_conversations,
+            new_conversations_week,
+            total_text_chunks,
+            new_text_chunks_week,
+        })
+    }
    pub async fn create_new(
        email: String,
        password: String,
@@ -78,7 +168,7 @@ impl User {
        let now = Utc::now();
        let id = Uuid::new_v4().to_string();

-        let user: Option<User> = db
+        let user: Option<Self> = db
            .client
            .query(
                "LET $count = (SELECT count() FROM type::table($table))[0].count;
@@ -95,8 +185,8 @@ impl User {
            .bind(("id", id))
            .bind(("email", email))
            .bind(("password", password))
-            .bind(("created_at", now))
-            .bind(("updated_at", now))
+            .bind(("created_at", surrealdb::Datetime::from(now)))
+            .bind(("updated_at", surrealdb::Datetime::from(now)))
            .bind(("timezone", validated_tz))
            .await?
            .take(1)?;
@@ -127,7 +217,7 @@ impl User {
        password: &str,
        db: &SurrealDbClient,
    ) -> Result<Self, AppError> {
-        let user: Option<User> = db
+        let user: Option<Self> = db
            .client
            .query(
                "SELECT * FROM user 
@@ -145,7 +235,7 @@ impl User {
        email: &str,
        db: &SurrealDbClient,
    ) -> Result<Option<Self>, AppError> {
-        let user: Option<User> = db
+        let user: Option<Self> = db
            .client
            .query("SELECT * FROM user WHERE email = $email LIMIT 1")
            .bind(("email", email.to_string()))
@@ -159,7 +249,7 @@ impl User {
        api_key: &str,
        db: &SurrealDbClient,
    ) -> Result<Option<Self>, AppError> {
-        let user: Option<User> = db
+        let user: Option<Self> = db
            .client
            .query("SELECT * FROM user WHERE api_key = $api_key LIMIT 1")
            .bind(("api_key", api_key.to_string()))
@@ -174,7 +264,7 @@ impl User {
        let api_key = format!("sk_{}", Uuid::new_v4().to_string().replace("-", ""));

        // Update the user record with the new API key
-        let user: Option<User> = db
+        let user: Option<Self> = db
            .client
            .query(
                "UPDATE type::thing('user', $id) 
@@ -195,7 +285,7 @@ impl User {
    }

    pub async fn revoke_api_key(id: &str, db: &SurrealDbClient) -> Result<(), AppError> {
-        let user: Option<User> = db
+        let user: Option<Self> = db
            .client
            .query(
                "UPDATE type::thing('user', $id) 
@@ -266,7 +356,10 @@ impl User {
        // Extract the entity types from the response
        let entity_types: Vec<String> = response
            .into_iter()
-            .map(|item| format!("{:?}", item.entity_type))
+            .map(|item| {
+                let normalized = KnowledgeEntityType::from(item.entity_type);
+                format!("{:?}", normalized)
+            })
            .collect();

        Ok(entity_types)
@@ -356,7 +449,7 @@ impl User {
        db: &SurrealDbClient,
    ) -> Result<(), AppError> {
        db.query("UPDATE type::thing('user', $user_id) SET timezone = $timezone")
-            .bind(("table_name", User::table_name()))
+            .bind(("table_name", Self::table_name()))
            .bind(("user_id", user_id.to_string()))
            .bind(("timezone", timezone.to_string()))
            .await?;
@@ -442,19 +535,43 @@ impl User {
        let jobs: Vec<IngestionTask> = db
            .query(
                "SELECT * FROM type::table($table) 
-             WHERE user_id = $user_id 
-             AND (
-                status = 'Created' 
-                OR (
-                    status.InProgress != NONE 
-                    AND status.InProgress.attempts < $max_attempts
-                )
-             )
-             ORDER BY created_at DESC",
+                 WHERE user_id = $user_id 
+                   AND (
+                       state IN $active_states
+                       OR (state = $failed_state AND attempts < max_attempts)
+                   )
+                 ORDER BY scheduled_at ASC, created_at DESC",
+            )
+            .bind(("table", IngestionTask::table_name()))
+            .bind(("user_id", user_id.to_owned()))
+            .bind((
+                "active_states",
+                vec![
+                    TaskState::Pending.as_str(),
+                    TaskState::Reserved.as_str(),
+                    TaskState::Processing.as_str(),
+                ],
+            ))
+            .bind(("failed_state", TaskState::Failed.as_str()))
+            .await?
+            .take(0)?;
+
+        Ok(jobs)
+    }
+
+    /// Gets all ingestion tasks for the specified user ordered by newest first
+    pub async fn get_all_ingestion_tasks(
+        user_id: &str,
+        db: &SurrealDbClient,
+    ) -> Result<Vec<IngestionTask>, AppError> {
+        let jobs: Vec<IngestionTask> = db
+            .query(
+                "SELECT * FROM type::table($table)
+                 WHERE user_id = $user_id
+                 ORDER BY created_at DESC",
            )
            .bind(("table", IngestionTask::table_name()))
            .bind(("user_id", user_id.to_owned()))
-            .bind(("max_attempts", 3))
            .await?
            .take(0)?;

@@ -511,6 +628,9 @@ impl User {
 #[cfg(test)]
 mod tests {
    use super::*;
+    use crate::storage::types::ingestion_payload::IngestionPayload;
+    use crate::storage::types::ingestion_task::{IngestionTask, TaskState, MAX_ATTEMPTS};
+    use std::collections::HashSet;

    // Helper function to set up a test database with SystemSettings
    async fn setup_test_db() -> SurrealDbClient {
@@ -596,6 +716,122 @@ mod tests {
        assert!(nonexistent.is_err());
    }

+    #[tokio::test]
+    async fn test_get_unfinished_ingestion_tasks_filters_correctly() {
+        let db = setup_test_db().await;
+        let user_id = "unfinished_user";
+        let other_user_id = "other_user";
+
+        let payload = IngestionPayload::Text {
+            text: "Test".to_string(),
+            context: "Context".to_string(),
+            category: "Category".to_string(),
+            user_id: user_id.to_string(),
+        };
+
+        let created_task = IngestionTask::new(payload.clone(), user_id.to_string());
+        db.store_item(created_task.clone())
+            .await
+            .expect("Failed to store created task");
+
+        let mut processing_task = IngestionTask::new(payload.clone(), user_id.to_string());
+        processing_task.state = TaskState::Processing;
+        processing_task.attempts = 1;
+        db.store_item(processing_task.clone())
+            .await
+            .expect("Failed to store processing task");
+
+        let mut failed_retry_task = IngestionTask::new(payload.clone(), user_id.to_string());
+        failed_retry_task.state = TaskState::Failed;
+        failed_retry_task.attempts = 1;
+        failed_retry_task.scheduled_at = chrono::Utc::now() - chrono::Duration::minutes(5);
+        db.store_item(failed_retry_task.clone())
+            .await
+            .expect("Failed to store retryable failed task");
+
+        let mut failed_blocked_task = IngestionTask::new(payload.clone(), user_id.to_string());
+        failed_blocked_task.state = TaskState::Failed;
+        failed_blocked_task.attempts = MAX_ATTEMPTS;
+        failed_blocked_task.error_message = Some("Too many failures".into());
+        db.store_item(failed_blocked_task.clone())
+            .await
+            .expect("Failed to store blocked task");
+
+        let mut completed_task = IngestionTask::new(payload.clone(), user_id.to_string());
+        completed_task.state = TaskState::Succeeded;
+        db.store_item(completed_task.clone())
+            .await
+            .expect("Failed to store completed task");
+
+        let other_payload = IngestionPayload::Text {
+            text: "Other".to_string(),
+            context: "Context".to_string(),
+            category: "Category".to_string(),
+            user_id: other_user_id.to_string(),
+        };
+        let other_task = IngestionTask::new(other_payload, other_user_id.to_string());
+        db.store_item(other_task)
+            .await
+            .expect("Failed to store other user task");
+
+        let unfinished = User::get_unfinished_ingestion_tasks(user_id, &db)
+            .await
+            .expect("Failed to fetch unfinished tasks");
+
+        let unfinished_ids: HashSet<String> =
+            unfinished.iter().map(|task| task.id.clone()).collect();
+
+        assert!(unfinished_ids.contains(&created_task.id));
+        assert!(unfinished_ids.contains(&processing_task.id));
+        assert!(unfinished_ids.contains(&failed_retry_task.id));
+        assert!(!unfinished_ids.contains(&failed_blocked_task.id));
+        assert!(!unfinished_ids.contains(&completed_task.id));
+        assert_eq!(unfinished_ids.len(), 3);
+    }
+
+    #[tokio::test]
+    async fn test_get_all_ingestion_tasks_returns_sorted() {
+        let db = setup_test_db().await;
+        let user_id = "archive_user";
+        let other_user_id = "other_user";
+
+        let payload = IngestionPayload::Text {
+            text: "One".to_string(),
+            context: "Context".to_string(),
+            category: "Category".to_string(),
+            user_id: user_id.to_string(),
+        };
+
+        // Oldest task
+        let mut first = IngestionTask::new(payload.clone(), user_id.to_string());
+        first.created_at = first.created_at - chrono::Duration::minutes(1);
+        first.updated_at = first.created_at;
+        first.state = TaskState::Succeeded;
+        db.store_item(first.clone()).await.expect("store first");
+
+        // Latest task
+        let mut second = IngestionTask::new(payload.clone(), user_id.to_string());
+        second.state = TaskState::Processing;
+        db.store_item(second.clone()).await.expect("store second");
+
+        let other_payload = IngestionPayload::Text {
+            text: "Other".to_string(),
+            context: "Context".to_string(),
+            category: "Category".to_string(),
+            user_id: other_user_id.to_string(),
+        };
+        let other_task = IngestionTask::new(other_payload, other_user_id.to_string());
+        db.store_item(other_task).await.expect("store other");
+
+        let tasks = User::get_all_ingestion_tasks(user_id, &db)
+            .await
+            .expect("fetch all tasks");
+
+        assert_eq!(tasks.len(), 2);
+        assert_eq!(tasks[0].id, second.id); // newest first
+        assert_eq!(tasks[1].id, first.id);
+    }
+
    #[tokio::test]
    async fn test_find_by_email() {
        // Setup test database
@@ -816,4 +1052,56 @@ mod tests {
        let most_recent = conversations.iter().max_by_key(|c| c.created_at).unwrap();
        assert_eq!(retrieved[0].id, most_recent.id);
    }
+
+    #[tokio::test]
+    async fn test_get_latest_text_contents_returns_last_five() {
+        let db = setup_test_db().await;
+        let user_id = "latest_text_user";
+
+        let mut inserted_ids = Vec::new();
+        let base_time = chrono::Utc::now() - chrono::Duration::minutes(60);
+
+        for i in 0..12 {
+            let mut item = TextContent::new(
+                format!("Text {}", i),
+                Some(format!("Context {}", i)),
+                "Category".to_string(),
+                None,
+                None,
+                user_id.to_string(),
+            );
+
+            let timestamp = base_time + chrono::Duration::minutes(i);
+            item.created_at = timestamp;
+            item.updated_at = timestamp;
+
+            db.store_item(item.clone())
+                .await
+                .expect("Failed to store text content");
+
+            inserted_ids.push(item.id.clone());
+        }
+
+        let latest = User::get_latest_text_contents(user_id, &db)
+            .await
+            .expect("Failed to fetch latest text contents");
+
+        assert_eq!(latest.len(), 5, "Expected exactly five items");
+
+        let mut expected_ids = inserted_ids[inserted_ids.len() - 5..].to_vec();
+        expected_ids.reverse();
+
+        let returned_ids: Vec<String> = latest.iter().map(|item| item.id.clone()).collect();
+        assert_eq!(
+            returned_ids, expected_ids,
+            "Latest items did not match expectation"
+        );
+
+        for window in latest.windows(2) {
+            assert!(
+                window[0].created_at >= window[1].created_at,
+                "Results are not ordered by created_at descending"
+            );
+        }
+    }
 }
--- a/common/src/utils/config.rs
+++ b/common/src/utils/config.rs
@@ -1,6 +1,30 @@
 use config::{Config, ConfigError, Environment, File};
 use serde::Deserialize;

+#[derive(Clone, Deserialize, Debug)]
+#[serde(rename_all = "lowercase")]
+pub enum StorageKind {
+    Local,
+}
+
+fn default_storage_kind() -> StorageKind {
+    StorageKind::Local
+}
+
+/// Selects the strategy used for PDF ingestion.
+#[derive(Clone, Deserialize, Debug)]
+#[serde(rename_all = "kebab-case")]
+pub enum PdfIngestMode {
+    /// Only rely on classic text extraction (no LLM fallbacks).
+    Classic,
+    /// Prefer fast text extraction, but fall back to the LLM rendering path when needed.
+    LlmFirst,
+}
+
+fn default_pdf_ingest_mode() -> PdfIngestMode {
+    PdfIngestMode::LlmFirst
+}
+
 #[derive(Clone, Deserialize, Debug)]
 pub struct AppConfig {
    pub openai_api_key: String,
@@ -11,17 +35,28 @@ pub struct AppConfig {
    pub surrealdb_database: String,
    #[serde(default = "default_data_dir")]
    pub data_dir: String,
+    pub http_port: u16,
+    #[serde(default = "default_base_url")]
+    pub openai_base_url: String,
+    #[serde(default = "default_storage_kind")]
+    pub storage: StorageKind,
+    #[serde(default = "default_pdf_ingest_mode")]
+    pub pdf_ingest_mode: PdfIngestMode,
 }

 fn default_data_dir() -> String {
    "./data".to_string()
 }

+fn default_base_url() -> String {
+    "https://api.openai.com/v1".to_string()
+}
+
 pub fn get_config() -> Result<AppConfig, ConfigError> {
    let config = Config::builder()
        .add_source(File::with_name("config").required(false))
        .add_source(Environment::default())
        .build()?;

-    Ok(config.try_deserialize()?)
+    config.try_deserialize()
 }
--- a/common/src/utils/embedding.rs
+++ b/common/src/utils/embedding.rs
@@ -1,6 +1,10 @@
 use async_openai::types::CreateEmbeddingRequestArgs;
+use tracing::debug;

-use crate::error::AppError;
+use crate::{
+    error::AppError,
+    storage::{db::SurrealDbClient, types::system_settings::SystemSettings},
+};
 /// Generates an embedding vector for the given input text using OpenAI's embedding model.
 ///
 /// This function takes a text input and converts it into a numerical vector representation (embedding)
@@ -27,9 +31,13 @@ use crate::error::AppError;
 pub async fn generate_embedding(
    client: &async_openai::Client<async_openai::config::OpenAIConfig>,
    input: &str,
+    db: &SurrealDbClient,
 ) -> Result<Vec<f32>, AppError> {
+    let model = SystemSettings::get_current(db).await?;
+
    let request = CreateEmbeddingRequestArgs::default()
-        .model("text-embedding-3-small")
+        .model(model.embedding_model)
+        .dimensions(model.embedding_dimensions)
        .input([input])
        .build()?;

@@ -46,3 +54,36 @@ pub async fn generate_embedding(

    Ok(embedding)
 }
+
+/// Generates an embedding vector using a specific model and dimension.
+///
+/// This is used for the re-embedding process where the model and dimensions
+/// are known ahead of time and shouldn't be repeatedly fetched from settings.
+pub async fn generate_embedding_with_params(
+    client: &async_openai::Client<async_openai::config::OpenAIConfig>,
+    input: &str,
+    model: &str,
+    dimensions: u32,
+) -> Result<Vec<f32>, AppError> {
+    let request = CreateEmbeddingRequestArgs::default()
+        .model(model)
+        .input([input])
+        .dimensions(dimensions)
+        .build()?;
+
+    let response = client.embeddings().create(request).await?;
+
+    let embedding = response
+        .data
+        .first()
+        .ok_or_else(|| AppError::LLMParsing("No embedding data received from API".into()))?
+        .embedding
+        .clone();
+
+    debug!(
+        "Embedding was created with {:?} dimensions",
+        embedding.len()
+    );
+
+    Ok(embedding)
+}
--- a/common/src/utils/template_engine.rs
+++ b/common/src/utils/template_engine.rs
@@ -59,13 +59,13 @@ impl TemplateEngine {
        match self {
            // Only compile this arm for debug builds
            #[cfg(debug_assertions)]
-            TemplateEngine::AutoReload(reloader) => {
+            Self::AutoReload(reloader) => {
                let env = reloader.acquire_env()?;
                env.get_template(name)?.render(ctx)
            }
            // Only compile this arm for release builds
            #[cfg(not(debug_assertions))]
-            TemplateEngine::Embedded(env) => env.get_template(name)?.render(ctx),
+            Self::Embedded(env) => env.get_template(name)?.render(ctx),
        }
    }

@@ -78,19 +78,17 @@ impl TemplateEngine {
        match self {
            // Only compile this arm for debug builds
            #[cfg(debug_assertions)]
-            TemplateEngine::AutoReload(reloader) => {
-                let env = reloader.acquire_env()?;
-                let template = env.get_template(template_name)?;
-                let mut state = template.eval_to_state(context)?;
-                state.render_block(block_name)
-            }
+            Self::AutoReload(reloader) => reloader
+                .acquire_env()?
+                .get_template(template_name)?
+                .eval_to_state(context)?
+                .render_block(block_name),
            // Only compile this arm for release builds
            #[cfg(not(debug_assertions))]
-            TemplateEngine::Embedded(env) => {
-                let template = env.get_template(template_name)?;
-                let mut state = template.eval_to_state(context)?;
-                state.render_block(block_name)
-            }
+            Self::Embedded(env) => env
+                .get_template(template_name)?
+                .eval_to_state(context)?
+                .render_block(block_name),
        }
    }
 }
--- a/composite-retrieval/Cargo.toml
+++ b/composite-retrieval/Cargo.toml
@@ -4,6 +4,9 @@ version = "0.1.0"
 edition = "2021"
 license = "AGPL-3.0-or-later"

+[lints]
+workspace = true
+
 [dependencies]
 tokio = { workspace = true }
 serde = { workspace = true }
@@ -18,3 +21,4 @@ async-openai = { workspace = true }
 uuid = { workspace = true }
 
 common = { path = "../common", features = ["test-utils"] }
+state-machines = { workspace = true }
--- a/composite-retrieval/src/answer_retrieval.rs
+++ b/composite-retrieval/src/answer_retrieval.rs
@@ -11,16 +11,15 @@ use common::{
    storage::{
        db::SurrealDbClient,
        types::{
-            knowledge_entity::KnowledgeEntity,
            message::{format_history, Message},
            system_settings::SystemSettings,
        },
    },
 };
 use serde::Deserialize;
-use serde_json::{json, Value};
+use serde_json::Value;

-use crate::retrieve_entities;
+use crate::{retrieve_entities, retrieved_entities_to_json};

 use super::answer_retrieval_helper::get_query_response_schema;

@@ -43,8 +42,8 @@ pub struct LLMResponseFormat {
 ///
 /// # Arguments
 ///
-/// * `surreal_db_client` - Client for SurrealDB interactions
-/// * `openai_client` - Client for OpenAI API calls
+/// * `surreal_db_client` - Client for `SurrealDB` interactions
+/// * `openai_client` - Client for `OpenAI` API calls
 /// * `query` - The user's query string
 /// * `user_id` - The user's id
 ///
@@ -66,7 +65,7 @@ pub async fn get_answer_with_references(
    let entities = retrieve_entities(surreal_db_client, openai_client, query, user_id).await?;
    let settings = SystemSettings::get_current(surreal_db_client).await?;

-    let entities_json = format_entities_json(&entities);
+    let entities_json = retrieved_entities_to_json(&entities);
    let user_message = create_user_message(&entities_json, query);

    let request = create_chat_request(user_message, &settings)?;
@@ -84,33 +83,17 @@ pub async fn get_answer_with_references(
    })
 }

-pub fn format_entities_json(entities: &[KnowledgeEntity]) -> Value {
-    json!(entities
-        .iter()
-        .map(|entity| {
-            json!({
-                "KnowledgeEntity": {
-                    "id": entity.id,
-                    "name": entity.name,
-                    "description": entity.description
-                }
-            })
-        })
-        .collect::<Vec<_>>())
-}
-
 pub fn create_user_message(entities_json: &Value, query: &str) -> String {
    format!(
-        r#"
+        r"
        Context Information:
        ==================
-        {}
+        {entities_json}

        User Question:
        ==================
-        {}
-        "#,
-        entities_json, query
+        {query}
+        "
    )
 }

@@ -120,7 +103,7 @@ pub fn create_user_message_with_history(
    query: &str,
 ) -> String {
    format!(
-        r#"
+        r"
        Chat history:
        ==================
        {}
@@ -132,7 +115,7 @@ pub fn create_user_message_with_history(
        User Question:
        ==================
        {}
-        "#,
+        ",
        format_history(history),
        entities_json,
        query
@@ -154,8 +137,6 @@ pub fn create_chat_request(

    CreateChatCompletionRequestArgs::default()
        .model(&settings.query_model)
-        .temperature(0.2)
-        .max_tokens(3048u32)
        .messages([
            ChatCompletionRequestSystemMessage::from(settings.query_system_prompt.clone()).into(),
            ChatCompletionRequestUserMessage::from(user_message).into(),
@@ -176,7 +157,7 @@ pub async fn process_llm_response(
        ))
        .and_then(|content| {
            serde_json::from_str::<LLMResponseFormat>(content).map_err(|e| {
-                AppError::LLMParsing(format!("Failed to parse LLM response into analysis: {}", e))
+                AppError::LLMParsing(format!("Failed to parse LLM response into analysis: {e}"))
            })
        })
 }
--- a/composite-retrieval/src/fts.rs
+++ b/composite-retrieval/src/fts.rs
@@ -0,0 +1,265 @@
+use std::collections::HashMap;
+
+use serde::Deserialize;
+use tracing::debug;
+
+use common::{
+    error::AppError,
+    storage::{db::SurrealDbClient, types::StoredObject},
+};
+
+use crate::scoring::Scored;
+use common::storage::types::file_info::deserialize_flexible_id;
+use surrealdb::sql::Thing;
+
+#[derive(Debug, Deserialize)]
+struct FtsScoreRow {
+    #[serde(deserialize_with = "deserialize_flexible_id")]
+    id: String,
+    fts_score: Option<f32>,
+}
+
+/// Executes a full-text search query against SurrealDB and returns scored results.
+///
+/// The function expects FTS indexes to exist for the provided table. Currently supports
+/// `knowledge_entity` (name + description) and `text_chunk` (chunk).
+pub async fn find_items_by_fts<T>(
+    take: usize,
+    query: &str,
+    db_client: &SurrealDbClient,
+    table: &str,
+    user_id: &str,
+) -> Result<Vec<Scored<T>>, AppError>
+where
+    T: for<'de> serde::Deserialize<'de> + StoredObject,
+{
+    let (filter_clause, score_clause) = match table {
+        "knowledge_entity" => (
+            "(name @0@ $terms OR description @1@ $terms)",
+            "(IF search::score(0) != NONE THEN search::score(0) ELSE 0 END) + \
+             (IF search::score(1) != NONE THEN search::score(1) ELSE 0 END)",
+        ),
+        "text_chunk" => (
+            "(chunk @0@ $terms)",
+            "IF search::score(0) != NONE THEN search::score(0) ELSE 0 END",
+        ),
+        _ => {
+            return Err(AppError::Validation(format!(
+                "FTS not configured for table '{table}'"
+            )))
+        }
+    };
+
+    let sql = format!(
+        "SELECT id, {score_clause} AS fts_score \
+         FROM {table} \
+         WHERE {filter_clause} \
+           AND user_id = $user_id \
+         ORDER BY fts_score DESC \
+         LIMIT $limit",
+        table = table,
+        filter_clause = filter_clause,
+        score_clause = score_clause
+    );
+
+    debug!(
+        table = table,
+        limit = take,
+        "Executing FTS query with filter clause: {}",
+        filter_clause
+    );
+
+    let mut response = db_client
+        .query(sql)
+        .bind(("terms", query.to_owned()))
+        .bind(("user_id", user_id.to_owned()))
+        .bind(("limit", take as i64))
+        .await?;
+
+    let score_rows: Vec<FtsScoreRow> = response.take(0)?;
+
+    if score_rows.is_empty() {
+        return Ok(Vec::new());
+    }
+
+    let ids: Vec<String> = score_rows.iter().map(|row| row.id.clone()).collect();
+    let thing_ids: Vec<Thing> = ids
+        .iter()
+        .map(|id| Thing::from((table, id.as_str())))
+        .collect();
+
+    let mut items_response = db_client
+        .query("SELECT * FROM type::table($table) WHERE id IN $things AND user_id = $user_id")
+        .bind(("table", table.to_owned()))
+        .bind(("things", thing_ids.clone()))
+        .bind(("user_id", user_id.to_owned()))
+        .await?;
+
+    let items: Vec<T> = items_response.take(0)?;
+
+    let mut item_map: HashMap<String, T> = items
+        .into_iter()
+        .map(|item| (item.get_id().to_owned(), item))
+        .collect();
+
+    let mut results = Vec::with_capacity(score_rows.len());
+    for row in score_rows {
+        if let Some(item) = item_map.remove(&row.id) {
+            let score = row.fts_score.unwrap_or_default();
+            results.push(Scored::new(item).with_fts_score(score));
+        }
+    }
+
+    Ok(results)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use common::storage::types::{
+        knowledge_entity::{KnowledgeEntity, KnowledgeEntityType},
+        text_chunk::TextChunk,
+        StoredObject,
+    };
+    use uuid::Uuid;
+
+    fn dummy_embedding() -> Vec<f32> {
+        vec![0.0; 1536]
+    }
+
+    #[tokio::test]
+    async fn fts_preserves_single_field_score_for_name() {
+        let namespace = "fts_test_ns";
+        let database = &Uuid::new_v4().to_string();
+        let db = SurrealDbClient::memory(namespace, database)
+            .await
+            .expect("failed to create in-memory surreal");
+
+        db.apply_migrations()
+            .await
+            .expect("failed to apply migrations");
+
+        let user_id = "user_fts";
+        let entity = KnowledgeEntity::new(
+            "source_a".into(),
+            "Rustacean handbook".into(),
+            "completely unrelated description".into(),
+            KnowledgeEntityType::Document,
+            None,
+            dummy_embedding(),
+            user_id.into(),
+        );
+
+        db.store_item(entity.clone())
+            .await
+            .expect("failed to insert entity");
+
+        db.rebuild_indexes()
+            .await
+            .expect("failed to rebuild indexes");
+
+        let results = find_items_by_fts::<KnowledgeEntity>(
+            5,
+            "rustacean",
+            &db,
+            KnowledgeEntity::table_name(),
+            user_id,
+        )
+        .await
+        .expect("fts query failed");
+
+        assert!(!results.is_empty(), "expected at least one FTS result");
+        assert!(
+            results[0].scores.fts.is_some(),
+            "expected an FTS score when only the name matched"
+        );
+    }
+
+    #[tokio::test]
+    async fn fts_preserves_single_field_score_for_description() {
+        let namespace = "fts_test_ns_desc";
+        let database = &Uuid::new_v4().to_string();
+        let db = SurrealDbClient::memory(namespace, database)
+            .await
+            .expect("failed to create in-memory surreal");
+
+        db.apply_migrations()
+            .await
+            .expect("failed to apply migrations");
+
+        let user_id = "user_fts_desc";
+        let entity = KnowledgeEntity::new(
+            "source_b".into(),
+            "neutral name".into(),
+            "Detailed notes about async runtimes".into(),
+            KnowledgeEntityType::Document,
+            None,
+            dummy_embedding(),
+            user_id.into(),
+        );
+
+        db.store_item(entity.clone())
+            .await
+            .expect("failed to insert entity");
+
+        db.rebuild_indexes()
+            .await
+            .expect("failed to rebuild indexes");
+
+        let results = find_items_by_fts::<KnowledgeEntity>(
+            5,
+            "async",
+            &db,
+            KnowledgeEntity::table_name(),
+            user_id,
+        )
+        .await
+        .expect("fts query failed");
+
+        assert!(!results.is_empty(), "expected at least one FTS result");
+        assert!(
+            results[0].scores.fts.is_some(),
+            "expected an FTS score when only the description matched"
+        );
+    }
+
+    #[tokio::test]
+    async fn fts_preserves_scores_for_text_chunks() {
+        let namespace = "fts_test_ns_chunks";
+        let database = &Uuid::new_v4().to_string();
+        let db = SurrealDbClient::memory(namespace, database)
+            .await
+            .expect("failed to create in-memory surreal");
+
+        db.apply_migrations()
+            .await
+            .expect("failed to apply migrations");
+
+        let user_id = "user_fts_chunk";
+        let chunk = TextChunk::new(
+            "source_chunk".into(),
+            "GraphQL documentation reference".into(),
+            dummy_embedding(),
+            user_id.into(),
+        );
+
+        db.store_item(chunk.clone())
+            .await
+            .expect("failed to insert chunk");
+
+        db.rebuild_indexes()
+            .await
+            .expect("failed to rebuild indexes");
+
+        let results =
+            find_items_by_fts::<TextChunk>(5, "graphql", &db, TextChunk::table_name(), user_id)
+                .await
+                .expect("fts query failed");
+
+        assert!(!results.is_empty(), "expected at least one FTS result");
+        assert!(
+            results[0].scores.fts.is_some(),
+            "expected an FTS score when chunk field matched"
+        );
+    }
+}
--- a/composite-retrieval/src/graph.rs
+++ b/composite-retrieval/src/graph.rs
@@ -1,7 +1,14 @@
-use surrealdb::Error;
-use tracing::debug;
+use std::collections::{HashMap, HashSet};

-use common::storage::{db::SurrealDbClient, types::knowledge_entity::KnowledgeEntity};
+use surrealdb::{sql::Thing, Error};
+
+use common::storage::{
+    db::SurrealDbClient,
+    types::{
+        knowledge_entity::KnowledgeEntity, knowledge_relationship::KnowledgeRelationship,
+        StoredObject,
+    },
+};

 /// Retrieves database entries that match a specific source identifier.
 ///
@@ -13,7 +20,7 @@ use common::storage::{db::SurrealDbClient, types::knowledge_entity::KnowledgeEnt
 ///
 /// * `source_id` - The identifier to search for in the database
 /// * `table_name` - The name of the table to search in
-/// * `db_client` - The SurrealDB client instance for database operations
+/// * `db_client` - The `SurrealDB` client instance for database operations
 ///
 /// # Type Parameters
 ///
@@ -31,18 +38,21 @@ use common::storage::{db::SurrealDbClient, types::knowledge_entity::KnowledgeEnt
 /// * The database query fails to execute
 /// * The results cannot be deserialized into type `T`
 pub async fn find_entities_by_source_ids<T>(
-    source_id: Vec<String>,
-    table_name: String,
+    source_ids: Vec<String>,
+    table_name: &str,
+    user_id: &str,
    db: &SurrealDbClient,
 ) -> Result<Vec<T>, Error>
 where
    T: for<'de> serde::Deserialize<'de>,
 {
-    let query = "SELECT * FROM type::table($table) WHERE source_id IN $source_ids";
+    let query =
+        "SELECT * FROM type::table($table) WHERE source_id IN $source_ids AND user_id = $user_id";

    db.query(query)
-        .bind(("table", table_name))
-        .bind(("source_ids", source_id))
+        .bind(("table", table_name.to_owned()))
+        .bind(("source_ids", source_ids))
+        .bind(("user_id", user_id.to_owned()))
        .await?
        .take(0)
 }
@@ -50,16 +60,92 @@ where
 /// Find entities by their relationship to the id
 pub async fn find_entities_by_relationship_by_id(
    db: &SurrealDbClient,
-    entity_id: String,
+    entity_id: &str,
+    user_id: &str,
+    limit: usize,
 ) -> Result<Vec<KnowledgeEntity>, Error> {
-    let query = format!(
-        "SELECT *, <-> relates_to <-> knowledge_entity AS related FROM knowledge_entity:`{}`",
-        entity_id
-    );
+    let mut relationships_response = db
+        .query(
+            "
+            SELECT * FROM relates_to
+            WHERE metadata.user_id = $user_id
+              AND (in = type::thing('knowledge_entity', $entity_id)
+                   OR out = type::thing('knowledge_entity', $entity_id))
+            ",
+        )
+        .bind(("entity_id", entity_id.to_owned()))
+        .bind(("user_id", user_id.to_owned()))
+        .await?;

-    debug!("{}", query);
+    let relationships: Vec<KnowledgeRelationship> = relationships_response.take(0)?;
+    if relationships.is_empty() {
+        return Ok(Vec::new());
+    }

-    db.query(query).await?.take(0)
+    let mut neighbor_ids: Vec<String> = Vec::new();
+    let mut seen: HashSet<String> = HashSet::new();
+    for rel in relationships {
+        if rel.in_ == entity_id {
+            if seen.insert(rel.out.clone()) {
+                neighbor_ids.push(rel.out);
+            }
+        } else if rel.out == entity_id {
+            if seen.insert(rel.in_.clone()) {
+                neighbor_ids.push(rel.in_);
+            }
+        } else {
+            if seen.insert(rel.in_.clone()) {
+                neighbor_ids.push(rel.in_.clone());
+            }
+            if seen.insert(rel.out.clone()) {
+                neighbor_ids.push(rel.out);
+            }
+        }
+    }
+
+    neighbor_ids.retain(|id| id != entity_id);
+
+    if neighbor_ids.is_empty() {
+        return Ok(Vec::new());
+    }
+
+    if limit > 0 && neighbor_ids.len() > limit {
+        neighbor_ids.truncate(limit);
+    }
+
+    let thing_ids: Vec<Thing> = neighbor_ids
+        .iter()
+        .map(|id| Thing::from((KnowledgeEntity::table_name(), id.as_str())))
+        .collect();
+
+    let mut neighbors_response = db
+        .query("SELECT * FROM type::table($table) WHERE id IN $things AND user_id = $user_id")
+        .bind(("table", KnowledgeEntity::table_name().to_owned()))
+        .bind(("things", thing_ids))
+        .bind(("user_id", user_id.to_owned()))
+        .await?;
+
+    let neighbors: Vec<KnowledgeEntity> = neighbors_response.take(0)?;
+    if neighbors.is_empty() {
+        return Ok(Vec::new());
+    }
+
+    let mut neighbor_map: HashMap<String, KnowledgeEntity> = neighbors
+        .into_iter()
+        .map(|entity| (entity.id.clone(), entity))
+        .collect();
+
+    let mut ordered = Vec::new();
+    for id in neighbor_ids {
+        if let Some(entity) = neighbor_map.remove(&id) {
+            ordered.push(entity);
+        }
+        if limit > 0 && ordered.len() >= limit {
+            break;
+        }
+    }
+
+    Ok(ordered)
 }

 #[cfg(test)]
@@ -149,7 +235,7 @@ mod tests {
        // Test finding entities by multiple source_ids
        let source_ids = vec![source_id1.clone(), source_id2.clone()];
        let found_entities: Vec<KnowledgeEntity> =
-            find_entities_by_source_ids(source_ids, KnowledgeEntity::table_name().to_string(), &db)
+            find_entities_by_source_ids(source_ids, KnowledgeEntity::table_name(), &user_id, &db)
                .await
                .expect("Failed to find entities by source_ids");

@@ -180,7 +266,8 @@ mod tests {
        let single_source_id = vec![source_id1.clone()];
        let found_entities: Vec<KnowledgeEntity> = find_entities_by_source_ids(
            single_source_id,
-            KnowledgeEntity::table_name().to_string(),
+            KnowledgeEntity::table_name(),
+            &user_id,
            &db,
        )
        .await
@@ -205,7 +292,8 @@ mod tests {
        let non_existent_source_id = vec!["non_existent_source".to_string()];
        let found_entities: Vec<KnowledgeEntity> = find_entities_by_source_ids(
            non_existent_source_id,
-            KnowledgeEntity::table_name().to_string(),
+            KnowledgeEntity::table_name(),
+            &user_id,
            &db,
        )
        .await
@@ -330,11 +418,15 @@ mod tests {
            .expect("Failed to store relationship 2");

        // Test finding entities related to the central entity
-        let related_entities = find_entities_by_relationship_by_id(&db, central_entity.id.clone())
-            .await
-            .expect("Failed to find entities by relationship");
+        let related_entities =
+            find_entities_by_relationship_by_id(&db, &central_entity.id, &user_id, usize::MAX)
+                .await
+                .expect("Failed to find entities by relationship");

        // Check that we found relationships
-        assert!(related_entities.len() > 0, "Should find related entities");
+        assert!(
+            related_entities.len() >= 2,
+            "Should find related entities in both directions"
+        );
    }
 }
--- a/composite-retrieval/src/lib.rs
+++ b/composite-retrieval/src/lib.rs
@@ -1,6 +1,9 @@
 pub mod answer_retrieval;
 pub mod answer_retrieval_helper;
+pub mod fts;
 pub mod graph;
+pub mod pipeline;
+pub mod scoring;
 pub mod vector;

 use common::{
@@ -10,81 +13,249 @@ use common::{
        types::{knowledge_entity::KnowledgeEntity, text_chunk::TextChunk},
    },
 };
-use futures::future::{try_join, try_join_all};
-use graph::{find_entities_by_relationship_by_id, find_entities_by_source_ids};
-use std::collections::HashMap;
-use vector::find_items_by_vector_similarity;
+use tracing::instrument;

-/// Performs a comprehensive knowledge entity retrieval using multiple search strategies
-/// to find the most relevant entities for a given query.
-///
-/// # Strategy
-/// The function employs a three-pronged approach to knowledge retrieval:
-/// 1. Direct vector similarity search on knowledge entities
-/// 2. Text chunk similarity search with source entity lookup
-/// 3. Graph relationship traversal from related entities
-///
-/// This combined approach ensures both semantic similarity matches and structurally
-/// related content are included in the results.
-///
-/// # Arguments
-/// * `db_client` - SurrealDB client for database operations
-/// * `openai_client` - OpenAI client for vector embeddings generation
-/// * `query` - The search query string to find relevant knowledge entities
-/// * 'user_id' - The user id of the current user
-///
-/// # Returns
-/// * `Result<Vec<KnowledgeEntity>, AppError>` - A deduplicated vector of relevant
-///   knowledge entities, or an error if the retrieval process fails
+pub use pipeline::{retrieved_entities_to_json, RetrievalConfig, RetrievalTuning};
+
+// Captures a supporting chunk plus its fused retrieval score for downstream prompts.
+#[derive(Debug, Clone)]
+pub struct RetrievedChunk {
+    pub chunk: TextChunk,
+    pub score: f32,
+}
+
+// Final entity representation returned to callers, enriched with ranked chunks.
+#[derive(Debug, Clone)]
+pub struct RetrievedEntity {
+    pub entity: KnowledgeEntity,
+    pub score: f32,
+    pub chunks: Vec<RetrievedChunk>,
+}
+
+// Primary orchestrator for the process of retrieving KnowledgeEntitities related to a input_text
+#[instrument(skip_all, fields(user_id))]
 pub async fn retrieve_entities(
    db_client: &SurrealDbClient,
    openai_client: &async_openai::Client<async_openai::config::OpenAIConfig>,
-    query: &str,
+    input_text: &str,
    user_id: &str,
-) -> Result<Vec<KnowledgeEntity>, AppError> {
-    let (items_from_knowledge_entity_similarity, closest_chunks) = try_join(
-        find_items_by_vector_similarity(
-            10,
-            query,
-            db_client,
-            "knowledge_entity",
-            openai_client,
-            user_id,
-        ),
-        find_items_by_vector_similarity(5, query, db_client, "text_chunk", openai_client, user_id),
+) -> Result<Vec<RetrievedEntity>, AppError> {
+    pipeline::run_pipeline(
+        db_client,
+        openai_client,
+        input_text,
+        user_id,
+        RetrievalConfig::default(),
    )
-    .await?;
-
-    let source_ids = closest_chunks
-        .iter()
-        .map(|chunk: &TextChunk| chunk.source_id.clone())
-        .collect::<Vec<String>>();
-
-    let items_from_text_chunk_similarity: Vec<KnowledgeEntity> =
-        find_entities_by_source_ids(source_ids, "knowledge_entity".to_string(), db_client).await?;
-
-    let items_from_relationships_futures: Vec<_> = items_from_text_chunk_similarity
-        .clone()
-        .into_iter()
-        .map(|entity| find_entities_by_relationship_by_id(db_client, entity.id.clone()))
-        .collect();
-
-    let items_from_relationships = try_join_all(items_from_relationships_futures)
-        .await?
-        .into_iter()
-        .flatten()
-        .collect::<Vec<KnowledgeEntity>>();
-
-    let entities: Vec<KnowledgeEntity> = items_from_knowledge_entity_similarity
-        .into_iter()
-        .chain(items_from_text_chunk_similarity.into_iter())
-        .chain(items_from_relationships.into_iter())
-        .fold(HashMap::new(), |mut map, entity| {
-            map.insert(entity.id.clone(), entity);
-            map
-        })
-        .into_values()
-        .collect();
-
-    Ok(entities)
+    .await
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use async_openai::Client;
+    use common::storage::types::{
+        knowledge_entity::{KnowledgeEntity, KnowledgeEntityType},
+        knowledge_relationship::KnowledgeRelationship,
+        text_chunk::TextChunk,
+    };
+    use pipeline::RetrievalConfig;
+    use uuid::Uuid;
+
+    fn test_embedding() -> Vec<f32> {
+        vec![0.9, 0.1, 0.0]
+    }
+
+    fn entity_embedding_high() -> Vec<f32> {
+        vec![0.8, 0.2, 0.0]
+    }
+
+    fn entity_embedding_low() -> Vec<f32> {
+        vec![0.1, 0.9, 0.0]
+    }
+
+    fn chunk_embedding_primary() -> Vec<f32> {
+        vec![0.85, 0.15, 0.0]
+    }
+
+    fn chunk_embedding_secondary() -> Vec<f32> {
+        vec![0.2, 0.8, 0.0]
+    }
+
+    async fn setup_test_db() -> SurrealDbClient {
+        let namespace = "test_ns";
+        let database = &Uuid::new_v4().to_string();
+        let db = SurrealDbClient::memory(namespace, database)
+            .await
+            .expect("Failed to start in-memory surrealdb");
+
+        db.apply_migrations()
+            .await
+            .expect("Failed to apply migrations");
+
+        db.query(
+            "BEGIN TRANSACTION;
+             REMOVE INDEX IF EXISTS idx_embedding_chunks ON TABLE text_chunk;
+             DEFINE INDEX idx_embedding_chunks ON TABLE text_chunk FIELDS embedding HNSW DIMENSION 3;
+             REMOVE INDEX IF EXISTS idx_embedding_entities ON TABLE knowledge_entity;
+             DEFINE INDEX idx_embedding_entities ON TABLE knowledge_entity FIELDS embedding HNSW DIMENSION 3;
+             COMMIT TRANSACTION;",
+        )
+        .await
+        .expect("Failed to configure indices");
+
+        db
+    }
+
+    #[tokio::test]
+    async fn test_retrieve_entities_with_embedding_basic_flow() {
+        let db = setup_test_db().await;
+        let user_id = "test_user";
+        let entity = KnowledgeEntity::new(
+            "source_1".into(),
+            "Rust async guide".into(),
+            "Detailed notes about async runtimes".into(),
+            KnowledgeEntityType::Document,
+            None,
+            entity_embedding_high(),
+            user_id.into(),
+        );
+        let chunk = TextChunk::new(
+            entity.source_id.clone(),
+            "Tokio uses cooperative scheduling for fairness.".into(),
+            chunk_embedding_primary(),
+            user_id.into(),
+        );
+
+        db.store_item(entity.clone())
+            .await
+            .expect("Failed to store entity");
+        db.store_item(chunk.clone())
+            .await
+            .expect("Failed to store chunk");
+
+        let openai_client = Client::new();
+        let results = pipeline::run_pipeline_with_embedding(
+            &db,
+            &openai_client,
+            test_embedding(),
+            "Rust concurrency async tasks",
+            user_id,
+            RetrievalConfig::default(),
+        )
+        .await
+        .expect("Hybrid retrieval failed");
+
+        assert!(
+            !results.is_empty(),
+            "Expected at least one retrieval result"
+        );
+        let top = &results[0];
+        assert!(
+            top.entity.name.contains("Rust"),
+            "Expected Rust entity to be ranked first"
+        );
+        assert!(
+            !top.chunks.is_empty(),
+            "Expected Rust entity to include supporting chunks"
+        );
+    }
+
+    #[tokio::test]
+    async fn test_graph_relationship_enriches_results() {
+        let db = setup_test_db().await;
+        let user_id = "graph_user";
+
+        let primary = KnowledgeEntity::new(
+            "primary_source".into(),
+            "Async Rust patterns".into(),
+            "Explores async runtimes and scheduling strategies.".into(),
+            KnowledgeEntityType::Document,
+            None,
+            entity_embedding_high(),
+            user_id.into(),
+        );
+        let neighbor = KnowledgeEntity::new(
+            "neighbor_source".into(),
+            "Tokio Scheduler Deep Dive".into(),
+            "Details on Tokio's cooperative scheduler.".into(),
+            KnowledgeEntityType::Document,
+            None,
+            entity_embedding_low(),
+            user_id.into(),
+        );
+
+        db.store_item(primary.clone())
+            .await
+            .expect("Failed to store primary entity");
+        db.store_item(neighbor.clone())
+            .await
+            .expect("Failed to store neighbor entity");
+
+        let primary_chunk = TextChunk::new(
+            primary.source_id.clone(),
+            "Rust async tasks use Tokio's cooperative scheduler.".into(),
+            chunk_embedding_primary(),
+            user_id.into(),
+        );
+        let neighbor_chunk = TextChunk::new(
+            neighbor.source_id.clone(),
+            "Tokio's scheduler manages task fairness across executors.".into(),
+            chunk_embedding_secondary(),
+            user_id.into(),
+        );
+
+        db.store_item(primary_chunk)
+            .await
+            .expect("Failed to store primary chunk");
+        db.store_item(neighbor_chunk)
+            .await
+            .expect("Failed to store neighbor chunk");
+
+        let openai_client = Client::new();
+        let relationship = KnowledgeRelationship::new(
+            primary.id.clone(),
+            neighbor.id.clone(),
+            user_id.into(),
+            "relationship_source".into(),
+            "references".into(),
+        );
+        relationship
+            .store_relationship(&db)
+            .await
+            .expect("Failed to store relationship");
+
+        let results = pipeline::run_pipeline_with_embedding(
+            &db,
+            &openai_client,
+            test_embedding(),
+            "Rust concurrency async tasks",
+            user_id,
+            RetrievalConfig::default(),
+        )
+        .await
+        .expect("Hybrid retrieval failed");
+
+        let mut neighbor_entry = None;
+        for entity in &results {
+            if entity.entity.id == neighbor.id {
+                neighbor_entry = Some(entity.clone());
+            }
+        }
+
+        let neighbor_entry =
+            neighbor_entry.expect("Graph-enriched neighbor should appear in results");
+
+        assert!(
+            neighbor_entry.score > 0.2,
+            "Graph-enriched entity should have a meaningful fused score"
+        );
+        assert!(
+            neighbor_entry
+                .chunks
+                .iter()
+                .all(|chunk| chunk.chunk.source_id == neighbor.source_id),
+            "Neighbor entity should surface its own supporting chunks"
+        );
+    }
 }
--- a/composite-retrieval/src/pipeline/config.rs
+++ b/composite-retrieval/src/pipeline/config.rs
@@ -0,0 +1,61 @@
+use serde::{Deserialize, Serialize};
+
+/// Tunable parameters that govern each retrieval stage.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct RetrievalTuning {
+    pub entity_vector_take: usize,
+    pub chunk_vector_take: usize,
+    pub entity_fts_take: usize,
+    pub chunk_fts_take: usize,
+    pub score_threshold: f32,
+    pub fallback_min_results: usize,
+    pub token_budget_estimate: usize,
+    pub avg_chars_per_token: usize,
+    pub max_chunks_per_entity: usize,
+    pub graph_traversal_seed_limit: usize,
+    pub graph_neighbor_limit: usize,
+    pub graph_score_decay: f32,
+    pub graph_seed_min_score: f32,
+    pub graph_vector_inheritance: f32,
+}
+
+impl Default for RetrievalTuning {
+    fn default() -> Self {
+        Self {
+            entity_vector_take: 15,
+            chunk_vector_take: 20,
+            entity_fts_take: 10,
+            chunk_fts_take: 20,
+            score_threshold: 0.35,
+            fallback_min_results: 10,
+            token_budget_estimate: 2800,
+            avg_chars_per_token: 4,
+            max_chunks_per_entity: 4,
+            graph_traversal_seed_limit: 5,
+            graph_neighbor_limit: 6,
+            graph_score_decay: 0.75,
+            graph_seed_min_score: 0.4,
+            graph_vector_inheritance: 0.6,
+        }
+    }
+}
+
+/// Wrapper containing tuning plus future flags for per-request overrides.
+#[derive(Debug, Clone)]
+pub struct RetrievalConfig {
+    pub tuning: RetrievalTuning,
+}
+
+impl RetrievalConfig {
+    pub fn new(tuning: RetrievalTuning) -> Self {
+        Self { tuning }
+    }
+}
+
+impl Default for RetrievalConfig {
+    fn default() -> Self {
+        Self {
+            tuning: RetrievalTuning::default(),
+        }
+    }
+}
--- a/composite-retrieval/src/pipeline/mod.rs
+++ b/composite-retrieval/src/pipeline/mod.rs
@@ -0,0 +1,100 @@
+mod config;
+mod stages;
+mod state;
+
+pub use config::{RetrievalConfig, RetrievalTuning};
+
+use crate::RetrievedEntity;
+use async_openai::Client;
+use common::{error::AppError, storage::db::SurrealDbClient};
+use tracing::info;
+
+/// Drives the retrieval pipeline from embedding through final assembly.
+pub async fn run_pipeline(
+    db_client: &SurrealDbClient,
+    openai_client: &Client<async_openai::config::OpenAIConfig>,
+    input_text: &str,
+    user_id: &str,
+    config: RetrievalConfig,
+) -> Result<Vec<RetrievedEntity>, AppError> {
+    let machine = state::ready();
+    let input_chars = input_text.chars().count();
+    let input_preview: String = input_text.chars().take(120).collect();
+    let input_preview_clean = input_preview.replace('\n', " ");
+    let preview_len = input_preview_clean.chars().count();
+    info!(
+        %user_id,
+        input_chars,
+        preview_truncated = input_chars > preview_len,
+        preview = %input_preview_clean,
+        "Starting ingestion retrieval pipeline"
+    );
+    let mut ctx = stages::PipelineContext::new(
+        db_client,
+        openai_client,
+        input_text.to_owned(),
+        user_id.to_owned(),
+        config,
+    );
+    let machine = stages::embed(machine, &mut ctx).await?;
+    let machine = stages::collect_candidates(machine, &mut ctx).await?;
+    let machine = stages::expand_graph(machine, &mut ctx).await?;
+    let machine = stages::attach_chunks(machine, &mut ctx).await?;
+    let results = stages::assemble(machine, &mut ctx)?;
+
+    Ok(results)
+}
+
+#[cfg(test)]
+pub async fn run_pipeline_with_embedding(
+    db_client: &SurrealDbClient,
+    openai_client: &Client<async_openai::config::OpenAIConfig>,
+    query_embedding: Vec<f32>,
+    input_text: &str,
+    user_id: &str,
+    config: RetrievalConfig,
+) -> Result<Vec<RetrievedEntity>, AppError> {
+    let machine = state::ready();
+    let mut ctx = stages::PipelineContext::with_embedding(
+        db_client,
+        openai_client,
+        query_embedding,
+        input_text.to_owned(),
+        user_id.to_owned(),
+        config,
+    );
+    let machine = stages::embed(machine, &mut ctx).await?;
+    let machine = stages::collect_candidates(machine, &mut ctx).await?;
+    let machine = stages::expand_graph(machine, &mut ctx).await?;
+    let machine = stages::attach_chunks(machine, &mut ctx).await?;
+    let results = stages::assemble(machine, &mut ctx)?;
+
+    Ok(results)
+}
+
+/// Helper exposed for tests to convert retrieved entities into downstream prompt JSON.
+pub fn retrieved_entities_to_json(entities: &[RetrievedEntity]) -> serde_json::Value {
+    serde_json::json!(entities
+        .iter()
+        .map(|entry| {
+            serde_json::json!({
+                "KnowledgeEntity": {
+                    "id": entry.entity.id,
+                    "name": entry.entity.name,
+                    "description": entry.entity.description,
+                    "score": round_score(entry.score),
+                    "chunks": entry.chunks.iter().map(|chunk| {
+                        serde_json::json!({
+                            "score": round_score(chunk.score),
+                            "content": chunk.chunk.chunk
+                        })
+                    }).collect::<Vec<_>>()
+                }
+            })
+        })
+        .collect::<Vec<_>>())
+}
+
+fn round_score(value: f32) -> f64 {
+    (f64::from(value) * 1000.0).round() / 1000.0
+}
--- a/composite-retrieval/src/pipeline/stages/mod.rs
+++ b/composite-retrieval/src/pipeline/stages/mod.rs
@@ -0,0 +1,599 @@
+use async_openai::Client;
+use common::{
+    error::AppError,
+    storage::{
+        db::SurrealDbClient,
+        types::{knowledge_entity::KnowledgeEntity, text_chunk::TextChunk, StoredObject},
+    },
+    utils::embedding::generate_embedding,
+};
+use futures::{stream::FuturesUnordered, StreamExt};
+use state_machines::core::GuardError;
+use std::collections::{HashMap, HashSet};
+use tracing::{debug, instrument, warn};
+
+use crate::{
+    fts::find_items_by_fts,
+    graph::{find_entities_by_relationship_by_id, find_entities_by_source_ids},
+    scoring::{
+        clamp_unit, fuse_scores, merge_scored_by_id, min_max_normalize, sort_by_fused_desc,
+        FusionWeights, Scored,
+    },
+    vector::find_items_by_vector_similarity_with_embedding,
+    RetrievedChunk, RetrievedEntity,
+};
+
+use super::{
+    config::RetrievalConfig,
+    state::{
+        CandidatesLoaded, ChunksAttached, Embedded, GraphExpanded, HybridRetrievalMachine, Ready,
+    },
+};
+
+pub struct PipelineContext<'a> {
+    pub db_client: &'a SurrealDbClient,
+    pub openai_client: &'a Client<async_openai::config::OpenAIConfig>,
+    pub input_text: String,
+    pub user_id: String,
+    pub config: RetrievalConfig,
+    pub query_embedding: Option<Vec<f32>>,
+    pub entity_candidates: HashMap<String, Scored<KnowledgeEntity>>,
+    pub chunk_candidates: HashMap<String, Scored<TextChunk>>,
+    pub filtered_entities: Vec<Scored<KnowledgeEntity>>,
+    pub chunk_values: Vec<Scored<TextChunk>>,
+}
+
+impl<'a> PipelineContext<'a> {
+    pub fn new(
+        db_client: &'a SurrealDbClient,
+        openai_client: &'a Client<async_openai::config::OpenAIConfig>,
+        input_text: String,
+        user_id: String,
+        config: RetrievalConfig,
+    ) -> Self {
+        Self {
+            db_client,
+            openai_client,
+            input_text,
+            user_id,
+            config,
+            query_embedding: None,
+            entity_candidates: HashMap::new(),
+            chunk_candidates: HashMap::new(),
+            filtered_entities: Vec::new(),
+            chunk_values: Vec::new(),
+        }
+    }
+
+    #[cfg(test)]
+    pub fn with_embedding(
+        db_client: &'a SurrealDbClient,
+        openai_client: &'a Client<async_openai::config::OpenAIConfig>,
+        query_embedding: Vec<f32>,
+        input_text: String,
+        user_id: String,
+        config: RetrievalConfig,
+    ) -> Self {
+        let mut ctx = Self::new(db_client, openai_client, input_text, user_id, config);
+        ctx.query_embedding = Some(query_embedding);
+        ctx
+    }
+
+    fn ensure_embedding(&self) -> Result<&Vec<f32>, AppError> {
+        self.query_embedding.as_ref().ok_or_else(|| {
+            AppError::InternalError(
+                "query embedding missing before candidate collection".to_string(),
+            )
+        })
+    }
+}
+
+#[instrument(level = "trace", skip_all)]
+pub async fn embed(
+    machine: HybridRetrievalMachine<(), Ready>,
+    ctx: &mut PipelineContext<'_>,
+) -> Result<HybridRetrievalMachine<(), Embedded>, AppError> {
+    let embedding_cached = ctx.query_embedding.is_some();
+    if embedding_cached {
+        debug!("Reusing cached query embedding for hybrid retrieval");
+    } else {
+        debug!("Generating query embedding for hybrid retrieval");
+        let embedding =
+            generate_embedding(ctx.openai_client, &ctx.input_text, ctx.db_client).await?;
+        ctx.query_embedding = Some(embedding);
+    }
+
+    machine
+        .embed()
+        .map_err(|(_, guard)| map_guard_error("embed", guard))
+}
+
+#[instrument(level = "trace", skip_all)]
+pub async fn collect_candidates(
+    machine: HybridRetrievalMachine<(), Embedded>,
+    ctx: &mut PipelineContext<'_>,
+) -> Result<HybridRetrievalMachine<(), CandidatesLoaded>, AppError> {
+    debug!("Collecting initial candidates via vector and FTS search");
+    let embedding = ctx.ensure_embedding()?.clone();
+    let tuning = &ctx.config.tuning;
+
+    let weights = FusionWeights::default();
+
+    let (vector_entities, vector_chunks, mut fts_entities, mut fts_chunks) = tokio::try_join!(
+        find_items_by_vector_similarity_with_embedding(
+            tuning.entity_vector_take,
+            embedding.clone(),
+            ctx.db_client,
+            "knowledge_entity",
+            &ctx.user_id,
+        ),
+        find_items_by_vector_similarity_with_embedding(
+            tuning.chunk_vector_take,
+            embedding,
+            ctx.db_client,
+            "text_chunk",
+            &ctx.user_id,
+        ),
+        find_items_by_fts(
+            tuning.entity_fts_take,
+            &ctx.input_text,
+            ctx.db_client,
+            "knowledge_entity",
+            &ctx.user_id,
+        ),
+        find_items_by_fts(
+            tuning.chunk_fts_take,
+            &ctx.input_text,
+            ctx.db_client,
+            "text_chunk",
+            &ctx.user_id
+        ),
+    )?;
+
+    debug!(
+        vector_entities = vector_entities.len(),
+        vector_chunks = vector_chunks.len(),
+        fts_entities = fts_entities.len(),
+        fts_chunks = fts_chunks.len(),
+        "Hybrid retrieval initial candidate counts"
+    );
+
+    normalize_fts_scores(&mut fts_entities);
+    normalize_fts_scores(&mut fts_chunks);
+
+    merge_scored_by_id(&mut ctx.entity_candidates, vector_entities);
+    merge_scored_by_id(&mut ctx.entity_candidates, fts_entities);
+    merge_scored_by_id(&mut ctx.chunk_candidates, vector_chunks);
+    merge_scored_by_id(&mut ctx.chunk_candidates, fts_chunks);
+
+    apply_fusion(&mut ctx.entity_candidates, weights);
+    apply_fusion(&mut ctx.chunk_candidates, weights);
+
+    machine
+        .collect_candidates()
+        .map_err(|(_, guard)| map_guard_error("collect_candidates", guard))
+}
+
+#[instrument(level = "trace", skip_all)]
+pub async fn expand_graph(
+    machine: HybridRetrievalMachine<(), CandidatesLoaded>,
+    ctx: &mut PipelineContext<'_>,
+) -> Result<HybridRetrievalMachine<(), GraphExpanded>, AppError> {
+    debug!("Expanding candidates using graph relationships");
+    let tuning = &ctx.config.tuning;
+    let weights = FusionWeights::default();
+
+    if ctx.entity_candidates.is_empty() {
+        return machine
+            .expand_graph()
+            .map_err(|(_, guard)| map_guard_error("expand_graph", guard));
+    }
+
+    let graph_seeds = seeds_from_candidates(
+        &ctx.entity_candidates,
+        tuning.graph_seed_min_score,
+        tuning.graph_traversal_seed_limit,
+    );
+
+    if graph_seeds.is_empty() {
+        return machine
+            .expand_graph()
+            .map_err(|(_, guard)| map_guard_error("expand_graph", guard));
+    }
+
+    let mut futures = FuturesUnordered::new();
+    for seed in graph_seeds {
+        let db = ctx.db_client;
+        let user = ctx.user_id.clone();
+        futures.push(async move {
+            let neighbors = find_entities_by_relationship_by_id(
+                db,
+                &seed.id,
+                &user,
+                tuning.graph_neighbor_limit,
+            )
+            .await;
+            (seed, neighbors)
+        });
+    }
+
+    while let Some((seed, neighbors_result)) = futures.next().await {
+        let neighbors = neighbors_result.map_err(AppError::from)?;
+        if neighbors.is_empty() {
+            continue;
+        }
+
+        for neighbor in neighbors {
+            if neighbor.id == seed.id {
+                continue;
+            }
+
+            let graph_score = clamp_unit(seed.fused * tuning.graph_score_decay);
+            let entry = ctx
+                .entity_candidates
+                .entry(neighbor.id.clone())
+                .or_insert_with(|| Scored::new(neighbor.clone()));
+
+            entry.item = neighbor;
+
+            let inherited_vector = clamp_unit(graph_score * tuning.graph_vector_inheritance);
+            let vector_existing = entry.scores.vector.unwrap_or(0.0);
+            if inherited_vector > vector_existing {
+                entry.scores.vector = Some(inherited_vector);
+            }
+
+            let existing_graph = entry.scores.graph.unwrap_or(f32::MIN);
+            if graph_score > existing_graph || entry.scores.graph.is_none() {
+                entry.scores.graph = Some(graph_score);
+            }
+
+            let fused = fuse_scores(&entry.scores, weights);
+            entry.update_fused(fused);
+        }
+    }
+
+    machine
+        .expand_graph()
+        .map_err(|(_, guard)| map_guard_error("expand_graph", guard))
+}
+
+#[instrument(level = "trace", skip_all)]
+pub async fn attach_chunks(
+    machine: HybridRetrievalMachine<(), GraphExpanded>,
+    ctx: &mut PipelineContext<'_>,
+) -> Result<HybridRetrievalMachine<(), ChunksAttached>, AppError> {
+    debug!("Attaching chunks to surviving entities");
+    let tuning = &ctx.config.tuning;
+    let weights = FusionWeights::default();
+
+    let chunk_by_source = group_chunks_by_source(&ctx.chunk_candidates);
+
+    backfill_entities_from_chunks(
+        &mut ctx.entity_candidates,
+        &chunk_by_source,
+        ctx.db_client,
+        &ctx.user_id,
+        weights,
+    )
+    .await?;
+
+    boost_entities_with_chunks(&mut ctx.entity_candidates, &chunk_by_source, weights);
+
+    let mut entity_results: Vec<Scored<KnowledgeEntity>> =
+        ctx.entity_candidates.values().cloned().collect();
+    sort_by_fused_desc(&mut entity_results);
+
+    let mut filtered_entities: Vec<Scored<KnowledgeEntity>> = entity_results
+        .iter()
+        .filter(|candidate| candidate.fused >= tuning.score_threshold)
+        .cloned()
+        .collect();
+
+    if filtered_entities.len() < tuning.fallback_min_results {
+        filtered_entities = entity_results
+            .into_iter()
+            .take(tuning.fallback_min_results)
+            .collect();
+    }
+
+    ctx.filtered_entities = filtered_entities;
+
+    let mut chunk_results: Vec<Scored<TextChunk>> =
+        ctx.chunk_candidates.values().cloned().collect();
+    sort_by_fused_desc(&mut chunk_results);
+
+    let mut chunk_by_id: HashMap<String, Scored<TextChunk>> = HashMap::new();
+    for chunk in chunk_results {
+        chunk_by_id.insert(chunk.item.id.clone(), chunk);
+    }
+
+    enrich_chunks_from_entities(
+        &mut chunk_by_id,
+        &ctx.filtered_entities,
+        ctx.db_client,
+        &ctx.user_id,
+        weights,
+    )
+    .await?;
+
+    let mut chunk_values: Vec<Scored<TextChunk>> = chunk_by_id.into_values().collect();
+    sort_by_fused_desc(&mut chunk_values);
+
+    ctx.chunk_values = chunk_values;
+
+    machine
+        .attach_chunks()
+        .map_err(|(_, guard)| map_guard_error("attach_chunks", guard))
+}
+
+#[instrument(level = "trace", skip_all)]
+pub fn assemble(
+    machine: HybridRetrievalMachine<(), ChunksAttached>,
+    ctx: &mut PipelineContext<'_>,
+) -> Result<Vec<RetrievedEntity>, AppError> {
+    debug!("Assembling final retrieved entities");
+    let tuning = &ctx.config.tuning;
+
+    let mut chunk_by_source: HashMap<String, Vec<Scored<TextChunk>>> = HashMap::new();
+    for chunk in ctx.chunk_values.drain(..) {
+        chunk_by_source
+            .entry(chunk.item.source_id.clone())
+            .or_default()
+            .push(chunk);
+    }
+
+    for chunk_list in chunk_by_source.values_mut() {
+        sort_by_fused_desc(chunk_list);
+    }
+
+    let mut token_budget_remaining = tuning.token_budget_estimate;
+    let mut results = Vec::new();
+
+    for entity in &ctx.filtered_entities {
+        let mut selected_chunks = Vec::new();
+        if let Some(candidates) = chunk_by_source.get_mut(&entity.item.source_id) {
+            let mut per_entity_count = 0;
+            candidates.sort_by(|a, b| {
+                b.fused
+                    .partial_cmp(&a.fused)
+                    .unwrap_or(std::cmp::Ordering::Equal)
+            });
+
+            for candidate in candidates.iter() {
+                if per_entity_count >= tuning.max_chunks_per_entity {
+                    break;
+                }
+                let estimated_tokens =
+                    estimate_tokens(&candidate.item.chunk, tuning.avg_chars_per_token);
+                if estimated_tokens > token_budget_remaining {
+                    continue;
+                }
+
+                token_budget_remaining = token_budget_remaining.saturating_sub(estimated_tokens);
+                per_entity_count += 1;
+
+                selected_chunks.push(RetrievedChunk {
+                    chunk: candidate.item.clone(),
+                    score: candidate.fused,
+                });
+            }
+        }
+
+        results.push(RetrievedEntity {
+            entity: entity.item.clone(),
+            score: entity.fused,
+            chunks: selected_chunks,
+        });
+
+        if token_budget_remaining == 0 {
+            break;
+        }
+    }
+
+    machine
+        .assemble()
+        .map_err(|(_, guard)| map_guard_error("assemble", guard))?;
+    Ok(results)
+}
+
+fn map_guard_error(stage: &'static str, err: GuardError) -> AppError {
+    AppError::InternalError(format!(
+        "state machine guard '{stage}' failed: guard={}, event={}, kind={:?}",
+        err.guard, err.event, err.kind
+    ))
+}
+fn normalize_fts_scores<T>(results: &mut [Scored<T>]) {
+    let raw_scores: Vec<f32> = results
+        .iter()
+        .map(|candidate| candidate.scores.fts.unwrap_or(0.0))
+        .collect();
+
+    let normalized = min_max_normalize(&raw_scores);
+    for (candidate, normalized_score) in results.iter_mut().zip(normalized.into_iter()) {
+        candidate.scores.fts = Some(normalized_score);
+        candidate.update_fused(0.0);
+    }
+}
+
+fn apply_fusion<T>(candidates: &mut HashMap<String, Scored<T>>, weights: FusionWeights)
+where
+    T: StoredObject,
+{
+    for candidate in candidates.values_mut() {
+        let fused = fuse_scores(&candidate.scores, weights);
+        candidate.update_fused(fused);
+    }
+}
+
+fn group_chunks_by_source(
+    chunks: &HashMap<String, Scored<TextChunk>>,
+) -> HashMap<String, Vec<Scored<TextChunk>>> {
+    let mut by_source: HashMap<String, Vec<Scored<TextChunk>>> = HashMap::new();
+
+    for chunk in chunks.values() {
+        by_source
+            .entry(chunk.item.source_id.clone())
+            .or_default()
+            .push(chunk.clone());
+    }
+    by_source
+}
+
+async fn backfill_entities_from_chunks(
+    entity_candidates: &mut HashMap<String, Scored<KnowledgeEntity>>,
+    chunk_by_source: &HashMap<String, Vec<Scored<TextChunk>>>,
+    db_client: &SurrealDbClient,
+    user_id: &str,
+    weights: FusionWeights,
+) -> Result<(), AppError> {
+    let mut missing_sources = Vec::new();
+
+    for source_id in chunk_by_source.keys() {
+        if !entity_candidates
+            .values()
+            .any(|entity| entity.item.source_id == *source_id)
+        {
+            missing_sources.push(source_id.clone());
+        }
+    }
+
+    if missing_sources.is_empty() {
+        return Ok(());
+    }
+
+    let related_entities: Vec<KnowledgeEntity> = find_entities_by_source_ids(
+        missing_sources.clone(),
+        "knowledge_entity",
+        user_id,
+        db_client,
+    )
+    .await
+    .unwrap_or_default();
+
+    if related_entities.is_empty() {
+        warn!("expected related entities for missing chunk sources, but none were found");
+    }
+
+    for entity in related_entities {
+        if let Some(chunks) = chunk_by_source.get(&entity.source_id) {
+            let best_chunk_score = chunks
+                .iter()
+                .map(|chunk| chunk.fused)
+                .fold(0.0f32, f32::max);
+
+            let mut scored = Scored::new(entity.clone()).with_vector_score(best_chunk_score);
+            let fused = fuse_scores(&scored.scores, weights);
+            scored.update_fused(fused);
+            entity_candidates.insert(entity.id.clone(), scored);
+        }
+    }
+
+    Ok(())
+}
+
+fn boost_entities_with_chunks(
+    entity_candidates: &mut HashMap<String, Scored<KnowledgeEntity>>,
+    chunk_by_source: &HashMap<String, Vec<Scored<TextChunk>>>,
+    weights: FusionWeights,
+) {
+    for entity in entity_candidates.values_mut() {
+        if let Some(chunks) = chunk_by_source.get(&entity.item.source_id) {
+            let best_chunk_score = chunks
+                .iter()
+                .map(|chunk| chunk.fused)
+                .fold(0.0f32, f32::max);
+
+            if best_chunk_score > 0.0 {
+                let boosted = entity.scores.vector.unwrap_or(0.0).max(best_chunk_score);
+                entity.scores.vector = Some(boosted);
+                let fused = fuse_scores(&entity.scores, weights);
+                entity.update_fused(fused);
+            }
+        }
+    }
+}
+
+async fn enrich_chunks_from_entities(
+    chunk_candidates: &mut HashMap<String, Scored<TextChunk>>,
+    entities: &[Scored<KnowledgeEntity>],
+    db_client: &SurrealDbClient,
+    user_id: &str,
+    weights: FusionWeights,
+) -> Result<(), AppError> {
+    let mut source_ids: HashSet<String> = HashSet::new();
+    for entity in entities {
+        source_ids.insert(entity.item.source_id.clone());
+    }
+
+    if source_ids.is_empty() {
+        return Ok(());
+    }
+
+    let chunks = find_entities_by_source_ids::<TextChunk>(
+        source_ids.into_iter().collect(),
+        "text_chunk",
+        user_id,
+        db_client,
+    )
+    .await?;
+
+    let mut entity_score_lookup: HashMap<String, f32> = HashMap::new();
+    for entity in entities {
+        entity_score_lookup.insert(entity.item.source_id.clone(), entity.fused);
+    }
+
+    for chunk in chunks {
+        let entry = chunk_candidates
+            .entry(chunk.id.clone())
+            .or_insert_with(|| Scored::new(chunk.clone()).with_vector_score(0.0));
+
+        let entity_score = entity_score_lookup
+            .get(&chunk.source_id)
+            .copied()
+            .unwrap_or(0.0);
+
+        entry.scores.vector = Some(entry.scores.vector.unwrap_or(0.0).max(entity_score * 0.8));
+        let fused = fuse_scores(&entry.scores, weights);
+        entry.update_fused(fused);
+        entry.item = chunk;
+    }
+
+    Ok(())
+}
+
+fn estimate_tokens(text: &str, avg_chars_per_token: usize) -> usize {
+    let chars = text.chars().count().max(1);
+    (chars / avg_chars_per_token).max(1)
+}
+
+#[derive(Clone)]
+struct GraphSeed {
+    id: String,
+    fused: f32,
+}
+
+fn seeds_from_candidates(
+    entity_candidates: &HashMap<String, Scored<KnowledgeEntity>>,
+    min_score: f32,
+    limit: usize,
+) -> Vec<GraphSeed> {
+    let mut seeds: Vec<GraphSeed> = entity_candidates
+        .values()
+        .filter(|entity| entity.fused >= min_score)
+        .map(|entity| GraphSeed {
+            id: entity.item.id.clone(),
+            fused: entity.fused,
+        })
+        .collect();
+
+    seeds.sort_by(|a, b| {
+        b.fused
+            .partial_cmp(&a.fused)
+            .unwrap_or(std::cmp::Ordering::Equal)
+    });
+    if seeds.len() > limit {
+        seeds.truncate(limit);
+    }
+
+    seeds
+}
--- a/composite-retrieval/src/pipeline/state.rs
+++ b/composite-retrieval/src/pipeline/state.rs
@@ -0,0 +1,25 @@
+use state_machines::state_machine;
+
+state_machine! {
+    name: HybridRetrievalMachine,
+    state: HybridRetrievalState,
+    initial: Ready,
+    states: [Ready, Embedded, CandidatesLoaded, GraphExpanded, ChunksAttached, Completed, Failed],
+    events {
+        embed { transition: { from: Ready, to: Embedded } }
+        collect_candidates { transition: { from: Embedded, to: CandidatesLoaded } }
+        expand_graph { transition: { from: CandidatesLoaded, to: GraphExpanded } }
+        attach_chunks { transition: { from: GraphExpanded, to: ChunksAttached } }
+        assemble { transition: { from: ChunksAttached, to: Completed } }
+        abort {
+            transition: { from: Ready, to: Failed }
+            transition: { from: CandidatesLoaded, to: Failed }
+            transition: { from: GraphExpanded, to: Failed }
+            transition: { from: ChunksAttached, to: Failed }
+        }
+    }
+}
+
+pub fn ready() -> HybridRetrievalMachine<(), Ready> {
+    HybridRetrievalMachine::new(())
+}
--- a/composite-retrieval/src/scoring.rs
+++ b/composite-retrieval/src/scoring.rs
@@ -0,0 +1,183 @@
+use std::cmp::Ordering;
+
+use common::storage::types::StoredObject;
+
+/// Holds optional subscores gathered from different retrieval signals.
+#[derive(Debug, Clone, Copy, Default)]
+pub struct Scores {
+    pub fts: Option<f32>,
+    pub vector: Option<f32>,
+    pub graph: Option<f32>,
+}
+
+/// Generic wrapper combining an item with its accumulated retrieval scores.
+#[derive(Debug, Clone)]
+pub struct Scored<T> {
+    pub item: T,
+    pub scores: Scores,
+    pub fused: f32,
+}
+
+impl<T> Scored<T> {
+    pub fn new(item: T) -> Self {
+        Self {
+            item,
+            scores: Scores::default(),
+            fused: 0.0,
+        }
+    }
+
+    pub const fn with_vector_score(mut self, score: f32) -> Self {
+        self.scores.vector = Some(score);
+        self
+    }
+
+    pub const fn with_fts_score(mut self, score: f32) -> Self {
+        self.scores.fts = Some(score);
+        self
+    }
+
+    pub const fn with_graph_score(mut self, score: f32) -> Self {
+        self.scores.graph = Some(score);
+        self
+    }
+
+    pub const fn update_fused(&mut self, fused: f32) {
+        self.fused = fused;
+    }
+}
+
+/// Weights used for linear score fusion.
+#[derive(Debug, Clone, Copy)]
+pub struct FusionWeights {
+    pub vector: f32,
+    pub fts: f32,
+    pub graph: f32,
+    pub multi_bonus: f32,
+}
+
+impl Default for FusionWeights {
+    fn default() -> Self {
+        Self {
+            vector: 0.5,
+            fts: 0.3,
+            graph: 0.2,
+            multi_bonus: 0.02,
+        }
+    }
+}
+
+pub const fn clamp_unit(value: f32) -> f32 {
+    value.clamp(0.0, 1.0)
+}
+
+pub fn distance_to_similarity(distance: f32) -> f32 {
+    if !distance.is_finite() {
+        return 0.0;
+    }
+    clamp_unit(1.0 / (1.0 + distance.max(0.0)))
+}
+
+pub fn min_max_normalize(scores: &[f32]) -> Vec<f32> {
+    if scores.is_empty() {
+        return Vec::new();
+    }
+
+    let mut min = f32::MAX;
+    let mut max = f32::MIN;
+
+    for s in scores {
+        if !s.is_finite() {
+            continue;
+        }
+        if *s < min {
+            min = *s;
+        }
+        if *s > max {
+            max = *s;
+        }
+    }
+
+    if !min.is_finite() || !max.is_finite() {
+        return scores.iter().map(|_| 0.0).collect();
+    }
+
+    if (max - min).abs() < f32::EPSILON {
+        return vec![1.0; scores.len()];
+    }
+
+    scores
+        .iter()
+        .map(|score| {
+            if score.is_finite() {
+                clamp_unit((score - min) / (max - min))
+            } else {
+                0.0
+            }
+        })
+        .collect()
+}
+
+pub fn fuse_scores(scores: &Scores, weights: FusionWeights) -> f32 {
+    let vector = scores.vector.unwrap_or(0.0);
+    let fts = scores.fts.unwrap_or(0.0);
+    let graph = scores.graph.unwrap_or(0.0);
+
+    let mut fused = graph.mul_add(
+        weights.graph,
+        vector.mul_add(weights.vector, fts * weights.fts),
+    );
+
+    let signals_present = scores
+        .vector
+        .iter()
+        .chain(scores.fts.iter())
+        .chain(scores.graph.iter())
+        .count();
+    if signals_present >= 2 {
+        fused += weights.multi_bonus;
+    }
+
+    clamp_unit(fused)
+}
+
+pub fn merge_scored_by_id<T>(
+    target: &mut std::collections::HashMap<String, Scored<T>>,
+    incoming: Vec<Scored<T>>,
+) where
+    T: StoredObject + Clone,
+{
+    for scored in incoming {
+        let id = scored.item.get_id().to_owned();
+        target
+            .entry(id)
+            .and_modify(|existing| {
+                if let Some(score) = scored.scores.vector {
+                    existing.scores.vector = Some(score);
+                }
+                if let Some(score) = scored.scores.fts {
+                    existing.scores.fts = Some(score);
+                }
+                if let Some(score) = scored.scores.graph {
+                    existing.scores.graph = Some(score);
+                }
+            })
+            .or_insert_with(|| Scored {
+                item: scored.item.clone(),
+                scores: scored.scores,
+                fused: scored.fused,
+            });
+    }
+}
+
+pub fn sort_by_fused_desc<T>(items: &mut [Scored<T>])
+where
+    T: StoredObject,
+{
+    items.sort_by(|a, b| {
+        b.fused
+            .partial_cmp(&a.fused)
+            .unwrap_or(Ordering::Equal)
+            .then_with(|| a.item.get_id().cmp(b.item.get_id()))
+    });
+}
--- a/composite-retrieval/src/vector.rs
+++ b/composite-retrieval/src/vector.rs
@@ -1,6 +1,15 @@
-use surrealdb::{engine::any::Any, Surreal};
+use std::collections::HashMap;

-use common::{error::AppError, utils::embedding::generate_embedding};
+use common::storage::types::file_info::deserialize_flexible_id;
+use common::{
+    error::AppError,
+    storage::{db::SurrealDbClient, types::StoredObject},
+    utils::embedding::generate_embedding,
+};
+use serde::Deserialize;
+use surrealdb::sql::Thing;
+
+use crate::scoring::{clamp_unit, distance_to_similarity, Scored};

 /// Compares vectors and retrieves a number of items from the specified table.
 ///
@@ -24,24 +33,125 @@ use common::{error::AppError, utils::embedding::generate_embedding};
 ///
 /// * `T` - The type to deserialize the query results into. Must implement `serde::Deserialize`.
 pub async fn find_items_by_vector_similarity<T>(
-    take: u8,
+    take: usize,
    input_text: &str,
-    db_client: &Surreal<Any>,
+    db_client: &SurrealDbClient,
    table: &str,
    openai_client: &async_openai::Client<async_openai::config::OpenAIConfig>,
    user_id: &str,
-) -> Result<Vec<T>, AppError>
+) -> Result<Vec<Scored<T>>, AppError>
 where
-    T: for<'de> serde::Deserialize<'de>,
+    T: for<'de> serde::Deserialize<'de> + StoredObject,
 {
    // Generate embeddings
-    let input_embedding = generate_embedding(openai_client, input_text).await?;
-
-    // Construct the query
-    let closest_query = format!("SELECT *, vector::distance::knn() AS distance FROM {} WHERE user_id = '{}' AND embedding <|{},40|> {:?} ORDER BY distance", table, user_id, take, input_embedding);
-
-    // Perform query and deserialize to struct
-    let closest_entities: Vec<T> = db_client.query(closest_query).await?.take(0)?;
-
-    Ok(closest_entities)
+    let input_embedding = generate_embedding(openai_client, input_text, db_client).await?;
+    find_items_by_vector_similarity_with_embedding(take, input_embedding, db_client, table, user_id)
+        .await
+}
+
+#[derive(Debug, Deserialize)]
+struct DistanceRow {
+    #[serde(deserialize_with = "deserialize_flexible_id")]
+    id: String,
+    distance: Option<f32>,
+}
+
+pub async fn find_items_by_vector_similarity_with_embedding<T>(
+    take: usize,
+    query_embedding: Vec<f32>,
+    db_client: &SurrealDbClient,
+    table: &str,
+    user_id: &str,
+) -> Result<Vec<Scored<T>>, AppError>
+where
+    T: for<'de> serde::Deserialize<'de> + StoredObject,
+{
+    let embedding_literal = serde_json::to_string(&query_embedding)
+        .map_err(|err| AppError::InternalError(format!("Failed to serialize embedding: {err}")))?;
+    let closest_query = format!(
+        "SELECT id, vector::distance::knn() AS distance \
+         FROM {table} \
+         WHERE user_id = $user_id AND embedding <|{take},40|> {embedding} \
+         LIMIT $limit",
+        table = table,
+        take = take,
+        embedding = embedding_literal
+    );
+
+    let mut response = db_client
+        .query(closest_query)
+        .bind(("user_id", user_id.to_owned()))
+        .bind(("limit", take as i64))
+        .await?;
+
+    let distance_rows: Vec<DistanceRow> = response.take(0)?;
+
+    if distance_rows.is_empty() {
+        return Ok(Vec::new());
+    }
+
+    let ids: Vec<String> = distance_rows.iter().map(|row| row.id.clone()).collect();
+    let thing_ids: Vec<Thing> = ids
+        .iter()
+        .map(|id| Thing::from((table, id.as_str())))
+        .collect();
+
+    let mut items_response = db_client
+        .query("SELECT * FROM type::table($table) WHERE id IN $things AND user_id = $user_id")
+        .bind(("table", table.to_owned()))
+        .bind(("things", thing_ids.clone()))
+        .bind(("user_id", user_id.to_owned()))
+        .await?;
+
+    let items: Vec<T> = items_response.take(0)?;
+
+    let mut item_map: HashMap<String, T> = items
+        .into_iter()
+        .map(|item| (item.get_id().to_owned(), item))
+        .collect();
+
+    let mut min_distance = f32::MAX;
+    let mut max_distance = f32::MIN;
+
+    for row in &distance_rows {
+        if let Some(distance) = row.distance {
+            if distance.is_finite() {
+                if distance < min_distance {
+                    min_distance = distance;
+                }
+                if distance > max_distance {
+                    max_distance = distance;
+                }
+            }
+        }
+    }
+
+    let normalize = min_distance.is_finite()
+        && max_distance.is_finite()
+        && (max_distance - min_distance).abs() > f32::EPSILON;
+
+    let mut scored = Vec::with_capacity(distance_rows.len());
+    for row in distance_rows {
+        if let Some(item) = item_map.remove(&row.id) {
+            let similarity = row
+                .distance
+                .map(|distance| {
+                    if normalize {
+                        let span = max_distance - min_distance;
+                        if span.abs() < f32::EPSILON {
+                            1.0
+                        } else {
+                            let normalized = 1.0 - ((distance - min_distance) / span);
+                            clamp_unit(normalized)
+                        }
+                    } else {
+                        distance_to_similarity(distance)
+                    }
+                })
+                .unwrap_or_default();
+            scored.push(Scored::new(item).with_vector_score(similarity));
+        }
+    }
+
+    Ok(scored)
 }
--- a/dist-workspace.toml
+++ b/dist-workspace.toml
@@ -17,3 +17,4 @@ allow-dirty = ["ci"]
 [dist.github-custom-runners]
 x86_64-unknown-linux-gnu = "ubuntu-22.04"
 x86_64-unknown-linux-musl = "ubuntu-22.04"
+x86_64-pc-windows-msvc = "windows-latest"
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -14,7 +14,8 @@ services:
      SURREALDB_NAMESPACE: "test"
      OPENAI_API_KEY: "sk-key"
      DATA_DIR: "./data"
-      # RUST_LOG: "info"
+      HTTP_PORT: 3000
+      RUST_LOG: "info"
    depends_on:
      - surrealdb 
    networks:
--- a/html-router/Cargo.toml
+++ b/html-router/Cargo.toml
@@ -4,6 +4,9 @@ version = "0.1.0"
 edition = "2021"
 license = "AGPL-3.0-or-later"

+[lints]
+workspace = true
+
 [dependencies]
 tokio = { workspace = true }
 serde = { workspace = true }
@@ -26,11 +29,13 @@ minijinja-embed = { workspace = true }
 minijinja-contrib = {workspace = true }
 axum-htmx = { workspace = true } 
 async-stream = { workspace = true } 
-plotly = { workspace = true } 
 tower-http = { workspace = true } 
 chrono-tz = { workspace = true } 
 tower-serve-static = { workspace = true } 
 tokio-util = { workspace = true } 
+chrono = { workspace = true }
+url = { workspace = true }
+uuid = { workspace = true }

 common = { path = "../common" }
 composite-retrieval = { path = "../composite-retrieval" }
--- a/html-router/app.css
+++ b/html-router/app.css
@@ -1,29 +1,108 @@
-@import 'tailwindcss' source(none);
+@import 'tailwindcss';
+
+@source './templates/**/*.html';

@plugin "daisyui" {
-  exclude: rootscrollbargutter;
+  logs: false;
+  themes: false;
+  include: [ "properties",
+    "scrollbar",
+    "rootscrolllock",
+    "rootcolor",
+    "svg",
+    "button",
+    "menu",
+    "navbar",
+    "drawer",
+    "modal",
+    "chat",
+    "card",
+    "loading",
+    "validator",
+    "fileinput",
+    "alert",
+    "swap"
+    ];
 }

@plugin "@tailwindcss/typography";

-@config './tailwind.config.js';
-
-/*
-  The default border color has changed to `currentColor` in Tailwind CSS v4,
-  so we've added these compatibility styles to make sure everything still
-  looks the same as it did with Tailwind CSS v3.
-
-  If we ever want to remove these styles, we need to add an explicit border
-  color utility to any element that depends on these defaults.
-*/
-
@view-transition {
  navigation: auto;
 }

@layer base {
+  :root {
+    --nb-shadow: 4px 4px 0 0 #000;
+    --nb-shadow-hover: 6px 6px 0 0 #000;
+  }
+
+  [data-theme="light"] {
+    color-scheme: light;
+    --color-base-100: oklch(98.42% 0.012 96.42);
+    --color-base-200: oklch(94.52% 0.0122 96.43);
+    --color-base-300: oklch(90.96% 0.0125 91.53);
+    --color-base-content: oklch(17.76% 0 89.88);
+    --color-primary: oklch(20.77% 0.0398 265.75);
+    --color-primary-content: oklch(100% 0 89.88);
+    --color-secondary: oklch(54.61% 0.2152 262.88);
+    --color-secondary-content: oklch(100% 0 89.88);
+    --color-accent: oklch(72% 0.19 80);
+    --color-accent-content: oklch(21% 0.035 80);
+    --color-neutral: oklch(17.76% 0 89.88);
+    --color-neutral-content: oklch(96.99% 0.0013 106.42);
+    --color-info: oklch(60.89% 0.1109 221.72);
+    --color-info-content: oklch(96.99% 0.0013 106.42);
+    --color-success: oklch(62.71% 0.1699 149.21);
+    --color-success-content: oklch(96.99% 0.0013 106.42);
+    --color-warning: oklch(79.52% 0.1617 86.05);
+    --color-warning-content: oklch(17.76% 0 89.88);
+    --color-error: oklch(57.71% 0.2152 27.33);
+    --color-error-content: oklch(96.99% 0.0013 106.42);
+    --radius-selector: 0rem;
+    --radius-field: 0rem;
+    --radius-box: 0rem;
+    --size-selector: 0.25rem;
+    --size-field: 0.25rem;
+    --border: 2px;
+  }
+
+  [data-theme="dark"] {
+    color-scheme: dark;
+    --color-base-100: oklch(22% 0.015 255);
+    --color-base-200: oklch(18% 0.014 253);
+    --color-base-300: oklch(14% 0.012 251);
+    --color-base-content: oklch(97.2% 0.02 255);
+    --color-primary: oklch(58% 0.233 277.12);
+    --color-primary-content: oklch(96% 0.018 272.31);
+    --color-secondary: oklch(65% 0.241 354.31);
+    --color-secondary-content: oklch(94% 0.028 342.26);
+    --color-accent: oklch(78% 0.22 80);
+    --color-accent-content: oklch(20% 0.035 80);
+    --color-neutral: oklch(26% 0.02 255);
+    --color-neutral-content: oklch(97% 0.03 255);
+    --color-info: oklch(74% 0.16 232.66);
+    --color-info-content: oklch(29% 0.066 243.16);
+    --color-success: oklch(76% 0.177 163.22);
+    --color-success-content: oklch(37% 0.077 168.94);
+    --color-warning: oklch(82% 0.189 84.43);
+    --color-warning-content: oklch(41% 0.112 45.9);
+    --color-error: oklch(71% 0.194 13.43);
+    --color-error-content: oklch(27% 0.105 12.09);
+    --radius-selector: 0rem;
+    --radius-field: 0rem;
+    --radius-box: 0rem;
+    --size-selector: 0.25rem;
+    --size-field: 0.25rem;
+    --border: 2px;
+  }
+
  body {
-    @apply font-satoshi;
+    background-color: var(--color-base-100);
+    color: var(--color-base-content);
+    font-family: 'Satoshi', sans-serif;
+    -webkit-font-smoothing: antialiased;
+    @apply selection:bg-yellow-300/40 selection:text-neutral;
  }

  html {
@@ -37,6 +116,581 @@
  ::file-selector-button {
    border-color: var(--color-gray-200, currentColor);
  }
+
+  .container {
+    padding-inline: 10px;
+  }
+
+  @media (min-width: 640px) {
+    .container {
+      padding-inline: 2rem;
+    }
+  }
+
+  @media (min-width: 1024px) {
+    .container {
+      padding-inline: 4rem;
+    }
+  }
+
+  @media (min-width: 1280px) {
+    .container {
+      padding-inline: 5rem;
+    }
+  }
+
+  @media (min-width: 1536px) {
+    .container {
+      padding-inline: 6rem;
+    }
+  }
+
+  .custom-scrollbar {
+    scrollbar-width: thin;
+    scrollbar-color: rgba(0, 0, 0, 0.2) transparent;
+  }
+
+  .custom-scrollbar::-webkit-scrollbar {
+    width: 4px;
+  }
+
+  .custom-scrollbar::-webkit-scrollbar-track {
+    background: transparent;
+  }
+
+  .custom-scrollbar::-webkit-scrollbar-thumb {
+    background-color: rgba(0, 0, 0, 0.2);
+    border-radius: 3px;
+  }
+
+  .hide-scrollbar {
+    -ms-overflow-style: none;
+    scrollbar-width: none;
+  }
+
+  .hide-scrollbar::-webkit-scrollbar {
+    display: none;
+  }
+
+  form.htmx-request {
+    opacity: 0.5;
+  }
+}
+
+/* Neobrutalist helpers influenced by Tufte principles */
+@layer components {
+
+  /* Offset, hard-edge shadow; minimal ink with strong contrast */
+  .nb-shadow {
+    box-shadow: var(--nb-shadow);
+    transition: transform 150ms, box-shadow 150ms;
+  }
+
+  .nb-shadow-hover {
+    transform: translate(-1px, -1px);
+    box-shadow: var(--nb-shadow-hover);
+  }
+
+  .nb-card {
+    @apply bg-base-100 border-2 border-neutral p-4;
+    box-shadow: var(--nb-shadow);
+    transition: transform 150ms, box-shadow 150ms;
+  }
+
+  .nb-card:hover {
+    transform: translate(-1px, -1px);
+    box-shadow: var(--nb-shadow-hover);
+  }
+
+  .nb-panel {
+    @apply border-2 border-neutral;
+    background-color: var(--nb-panel-bg, var(--color-base-200));
+    box-shadow: var(--nb-shadow);
+    transition: transform 150ms, box-shadow 150ms;
+  }
+
+  .nb-panel:hover {
+    transform: translate(-1px, -1px);
+    box-shadow: var(--nb-shadow-hover);
+  }
+
+  .nb-panel-canvas {
+    --nb-panel-bg: var(--color-base-100);
+  }
+
+  .nb-canvas {
+    background-color: var(--color-base-100);
+  }
+
+  .nb-btn {
+    @apply btn rounded-none border-2 border-neutral text-base-content;
+    --btn-color: var(--color-base-100);
+    --btn-fg: var(--color-base-content);
+    --btn-noise: none;
+    background-image: none;
+    box-shadow: var(--nb-shadow);
+    transition: transform 150ms, box-shadow 150ms;
+  }
+
+  .nb-btn:hover {
+    transform: translate(-1px, -1px);
+    box-shadow: var(--nb-shadow-hover);
+  }
+
+  .nb-link {
+    @apply underline underline-offset-2 decoration-neutral hover:decoration-4;
+  }
+
+  .nb-stat {
+    @apply bg-base-100 border-2 border-neutral p-5 flex flex-col gap-1;
+    box-shadow: var(--nb-shadow);
+    transition: transform 150ms, box-shadow 150ms;
+  }
+
+  /* Hairline rules and quiet gridlines for Tufte feel */
+  .u-hairline {
+    @apply border-t border-neutral/20;
+  }
+
+  .prose-tufte {
+    @apply prose prose-neutral;
+    max-width: min(90ch, 100%);
+    line-height: 1.7;
+  }
+
+  .prose-tufte-compact {
+    @apply prose prose-neutral;
+    max-width: min(90ch, 100%);
+    font-size: 0.875rem;
+    line-height: 1.6;
+  }
+
+  [data-theme="dark"] .prose-tufte,
+  [data-theme="dark"] .prose-tufte-compact {
+    color: var(--color-base-content);
+    --tw-prose-body: var(--color-base-content);
+    --tw-prose-headings: var(--color-base-content);
+    --tw-prose-lead: rgba(255, 255, 255, 0.78);
+    --tw-prose-links: var(--color-accent);
+    --tw-prose-bold: var(--color-base-content);
+    --tw-prose-counters: rgba(255, 255, 255, 0.7);
+    --tw-prose-bullets: rgba(255, 255, 255, 0.35);
+    --tw-prose-hr: rgba(255, 255, 255, 0.2);
+    --tw-prose-quotes: var(--color-base-content);
+    --tw-prose-quote-borders: rgba(255, 255, 255, 0.25);
+    --tw-prose-captions: rgba(255, 255, 255, 0.65);
+    --tw-prose-code: var(--color-base-content);
+    --tw-prose-pre-code: inherit;
+    --tw-prose-pre-bg: rgba(255, 255, 255, 0.07);
+    --tw-prose-th-borders: rgba(255, 255, 255, 0.25);
+    --tw-prose-td-borders: rgba(255, 255, 255, 0.2);
+  }
+
+  [data-theme="dark"] .prose-tufte a,
+  [data-theme="dark"] .prose-tufte-compact a {
+    color: var(--color-accent);
+  }
+
+  /* Encourage a consistent card look app-wide */
+  .card {
+    @apply border-2 border-neutral rounded-none;
+    box-shadow: var(--nb-shadow);
+    transition: transform 150ms, box-shadow 150ms;
+  }
+
+  .card:hover {
+    transform: translate(-1px, -1px);
+    box-shadow: var(--nb-shadow-hover);
+  }
+
+  /* Input styling with good dark/light contrast */
+  .nb-input {
+    @apply rounded-none border-2 border-neutral bg-base-100 text-base-content placeholder:text-base-content/60 px-3 py-[0.5rem];
+    box-shadow: var(--nb-shadow);
+    transition: transform 150ms, box-shadow 150ms, border-color 150ms;
+  }
+
+  .nb-input:hover {
+    transform: translate(-1px, -1px);
+    box-shadow: var(--nb-shadow-hover);
+  }
+
+  .nb-input:focus {
+    outline: none;
+    box-shadow: var(--nb-shadow-hover);
+  }
+
+  /* Select styling parallels inputs */
+  .nb-select {
+    @apply rounded-none border-2 border-neutral bg-base-100 text-base-content px-3 py-[0.5rem];
+    box-shadow: var(--nb-shadow);
+    transition: transform 150ms, box-shadow 150ms, border-color 150ms;
+  }
+
+  .nb-select:hover {
+    transform: translate(-1px, -1px);
+    box-shadow: var(--nb-shadow-hover);
+  }
+
+  .nb-select:focus {
+    outline: none;
+    box-shadow: var(--nb-shadow-hover);
+  }
+
+  /* Compact variants */
+  .nb-input-sm {
+    @apply text-sm px-2 py-[0.25rem];
+  }
+
+  .nb-select-sm {
+    @apply text-sm px-2 py-[0.25rem];
+  }
+
+  .nb-cta {
+    --btn-color: var(--color-accent);
+    --btn-fg: var(--color-accent-content);
+    --btn-noise: none;
+    background-image: none;
+    background-color: var(--color-accent);
+    color: var(--color-accent-content);
+  }
+
+  .nb-cta:hover {
+    background-color: var(--color-accent);
+    color: var(--color-accent-content);
+    filter: saturate(1.1) brightness(1.05);
+  }
+
+  /* Badges */
+  .nb-badge {
+    @apply inline-flex items-center uppercase tracking-wide text-[10px] px-2 py-0.5 bg-base-100 border-2 border-neutral rounded-none;
+    box-shadow: 3px 3px 0 0 #000;
+  }
+
+  .nb-masonry {
+    column-count: 1;
+    column-gap: 1rem;
+  }
+
+  .nb-masonry>* {
+    break-inside: avoid;
+    display: block;
+  }
+
+  @media (min-width: 768px) {
+    .nb-masonry {
+      column-count: 2;
+    }
+  }
+
+  @media (min-width: 1536px) {
+    .nb-masonry {
+      column-count: 3;
+    }
+  }
+
+  /* Chat bubbles neobrutalist */
+  .chat .chat-bubble {
+    @apply rounded-none border-2 border-neutral bg-base-100 text-neutral;
+    box-shadow: var(--nb-shadow);
+    transition: transform 150ms, box-shadow 150ms;
+  }
+
+  /* Remove DaisyUI tail so our rectangle keeps clean borders/shadows */
+  .chat .chat-bubble::before,
+  .chat .chat-bubble::after {
+    display: none !important;
+    content: none !important;
+  }
+
+  .chat.chat-start .chat-bubble {
+    @apply bg-secondary text-secondary-content;
+  }
+
+  .chat.chat-end .chat-bubble {
+    @apply bg-base-100 text-neutral;
+  }
+
+  /* Tables */
+  .nb-table {
+    @apply w-full;
+    border-collapse: separate;
+    border-spacing: 0;
+  }
+
+  .nb-table thead th {
+    @apply uppercase tracking-wide text-xs border-b-2 border-neutral;
+  }
+
+  .nb-table th,
+  .nb-table td {
+    @apply p-3;
+  }
+
+  .nb-table tbody tr+tr td {
+    @apply border-t border-neutral/30;
+  }
+
+  .nb-table tbody tr:hover {
+    @apply bg-base-200/40;
+  }
+
+  .nb-table tbody tr:hover td:first-child {
+    box-shadow: inset 3px 0 0 0 #000;
+  }
+
+  .kg-overlay {
+    @apply absolute top-4 left-4 right-4 z-10 flex flex-col items-stretch gap-2;
+    max-width: min(420px, calc(100% - 2rem));
+  }
+
+  .kg-control-row {
+    @apply flex flex-wrap items-center gap-2;
+  }
+
+  .kg-control-row-primary {
+    @apply justify-start;
+  }
+
+  .kg-control-row-secondary {
+    @apply justify-center;
+  }
+
+  .kg-search-input {
+    @apply pl-2;
+    height: 2rem;
+    width: 100%;
+    max-width: 320px;
+    min-width: 0;
+  }
+
+  .kg-control-row-primary .kg-search-input {
+    flex: 1 1 auto;
+  }
+
+  .kg-search-btn {
+    flex: 0 0 auto;
+  }
+
+  .kg-toggle {
+    @apply transition-colors;
+  }
+
+  .kg-toggle-active {
+    --btn-color: var(--color-accent);
+    --btn-fg: var(--color-accent-content);
+    --btn-noise: none;
+    background-image: none;
+    background-color: var(--color-accent);
+    color: var(--color-accent-content);
+  }
+
+  .kg-toggle-active:hover {
+    background-color: var(--color-accent);
+    color: var(--color-accent-content);
+    filter: saturate(1.1) brightness(1.05);
+  }
+
+  @media (min-width: 768px) {
+    .kg-overlay {
+      right: auto;
+      max-width: none;
+      width: auto;
+    }
+  }
+
+  .kg-legend {
+    @apply absolute bottom-2 left-2 z-10 flex flex-wrap gap-4;
+  }
+
+  .kg-legend-card {
+    @apply p-2;
+  }
+
+  .kg-legend-heading {
+    @apply mb-1 text-xs opacity-70;
+  }
+
+  .kg-legend-row {
+    @apply flex items-center gap-2 text-xs;
+  }
+
+  /* Checkboxes */
+  .nb-checkbox {
+    @apply appearance-none inline-block align-middle rounded-none border-2 border-neutral bg-base-100;
+    width: 1rem;
+    height: 1rem;
+    box-shadow: var(--nb-shadow);
+    transition: transform 150ms, box-shadow 150ms, border-color 150ms, background-color 150ms;
+    background-repeat: no-repeat;
+    background-position: center;
+    background-size: 80% 80%;
+    cursor: pointer;
+  }
+
+  .nb-checkbox:hover {
+    transform: translate(-1px, -1px);
+    box-shadow: 5px 5px 0 0 #000;
+  }
+
+  .nb-checkbox:focus-visible {
+    outline: 2px solid #000;
+    outline-offset: 2px;
+  }
+
+  .nb-checkbox:active {
+    transform: translate(0, 0);
+    box-shadow: 3px 3px 0 0 #000;
+  }
+
+  /* Tick mark in light mode (black) */
+  .nb-checkbox:checked {
+    background-image: url("data:image/svg+xml;utf8,<svg xmlns='http://www.w3.org/2000/svg' width='24' height='24' viewBox='0 0 24 24' fill='none' stroke='%23000' stroke-width='3' stroke-linecap='round' stroke-linejoin='round'><polyline points='20 6 9 17 4 12'/></svg>");
+  }
+
+  /* Tick mark in dark mode (white) */
+  [data-theme="dark"] .nb-checkbox:checked {
+    background-image: url("data:image/svg+xml;utf8,<svg xmlns='http://www.w3.org/2000/svg' width='24' height='24' viewBox='0 0 24 24' fill='none' stroke='%23fff' stroke-width='3' stroke-linecap='round' stroke-linejoin='round'><polyline points='20 6 9 17 4 12'/></svg>");
+  }
+
+  /* Compact size */
+  .nb-checkbox-sm {
+    width: 0.875rem;
+    height: 0.875rem;
+  }
+
+  /* Placeholder style for smaller, quieter helper text */
+  .nb-input::placeholder {
+    font-size: 0.75rem;
+    letter-spacing: 0.02em;
+    opacity: 0.75;
+  }
+
+  .markdown-content {
+    line-height: 1.5;
+    word-wrap: break-word;
+  }
+
+  .markdown-content p {
+    margin-bottom: 0.75em;
+  }
+
+  .markdown-content p:last-child {
+    margin-bottom: 0;
+  }
+
+  .markdown-content ul,
+  .markdown-content ol {
+    margin-top: 0.5em;
+    margin-bottom: 0.75em;
+    padding-left: 2em;
+  }
+
+  .markdown-content li {
+    margin-bottom: 0.25em;
+  }
+
+  .markdown-content pre {
+    background-color: var(--color-base-200);
+    color: var(--color-base-content);
+    padding: 0.75em 1em;
+    border-radius: 4px;
+    border: 1px solid rgba(0, 0, 0, 0.08);
+    overflow-x: auto;
+  }
+
+  .markdown-content pre code {
+    background-color: transparent;
+    color: inherit;
+    padding: 0;
+    border-radius: 0;
+    display: block;
+    line-height: inherit;
+  }
+
+  .markdown-content :not(pre) > code {
+    background-color: rgba(0, 0, 0, 0.05);
+    color: var(--color-base-content);
+    padding: 0.15em 0.4em;
+    border-radius: 3px;
+    font-size: 0.9em;
+  }
+
+  .markdown-content table {
+    border-collapse: collapse;
+    margin: 0.75em 0;
+    width: 100%;
+  }
+
+  .markdown-content th,
+  .markdown-content td {
+    border: 1px solid rgba(0, 0, 0, 0.15);
+    padding: 6px 12px;
+    text-align: left;
+  }
+
+  [data-theme="dark"] .markdown-content th,
+  [data-theme="dark"] .markdown-content td {
+    border-color: rgba(255, 255, 255, 0.25);
+  }
+
+  .markdown-content blockquote {
+    border-left: 4px solid rgba(0, 0, 0, 0.15);
+    padding-left: 10px;
+    margin: 0.5em 0 0.5em 0.5em;
+    color: rgba(0, 0, 0, 0.6);
+  }
+
+  [data-theme="dark"] .markdown-content blockquote {
+    border-color: rgba(255, 255, 255, 0.3);
+    color: rgba(255, 255, 255, 0.8);
+  }
+
+  .markdown-content hr {
+    border: none;
+    border-top: 1px solid rgba(0, 0, 0, 0.15);
+    margin: 0.75em 0;
+  }
+
+  [data-theme="dark"] .markdown-content hr {
+    border-top-color: rgba(255, 255, 255, 0.2);
+  }
+
+  [data-theme="dark"] .markdown-content pre {
+    background-color: var(--color-base-200);
+    border-color: rgba(255, 255, 255, 0.12);
+    color: var(--color-base-content);
+  }
+
+  [data-theme="dark"] .markdown-content :not(pre) > code {
+    background-color: rgba(255, 255, 255, 0.12);
+    color: var(--color-base-content);
+  }
+
+  .brand-mark {
+    letter-spacing: 0.02em;
+  }
+
+  .reference-tooltip {
+    @apply bg-base-100 text-base-content border-2 border-neutral p-3 text-sm w-72 max-w-xs;
+    position: fixed;
+    z-index: 9999;
+    box-shadow: var(--nb-shadow);
+  }
+}
+
+/* Theme-aware placeholder contrast tweaks */
+@layer base {
+
+  /* Light theme keeps default neutral tone via utilities */
+  [data-theme="dark"] .nb-input::placeholder,
+  [data-theme="dark"] .input::placeholder,
+  [data-theme="dark"] .textarea::placeholder,
+  [data-theme="dark"] textarea::placeholder,
+  [data-theme="dark"] input::placeholder {
+    color: rgba(255, 255, 255, 0.78) !important;
+    opacity: 0.85;
+  }
 }

 /* satoshi.css */
@@ -58,4 +712,28 @@
  font-weight: 300 900;
  font-style: italic;
  font-display: swap;
-}
+}
+
+/* Minimal override: prevent DaisyUI .menu hover bg on our nb buttons */
+@layer utilities {
+
+  /* Let plain nb-btns remain transparent on hover within menus */
+  .menu li>.nb-btn:hover {
+    background-color: transparent;
+  }
+
+  /* Keep CTA background on hover within menus */
+  .menu li>.nb-cta:hover {
+    background-color: var(--color-accent);
+    color: var(--color-accent-content);
+  }
+
+  .toast-alert {
+    @apply mt-2 flex flex-col text-left gap-1;
+    box-shadow: var(--nb-shadow);
+  }
+
+  .toast-alert-title {
+    @apply text-lg font-bold;
+  }
+}
--- a/html-router/assets/d3.min.js
+++ b/html-router/assets/d3.min.js
--- a/html-router/assets/icon/android-chrome-192x192.png
+++ b/html-router/assets/icon/android-chrome-192x192.png
--- a/html-router/assets/icon/android-chrome-512x512.png
+++ b/html-router/assets/icon/android-chrome-512x512.png
--- a/html-router/assets/icon/apple-touch-icon.png
+++ b/html-router/assets/icon/apple-touch-icon.png
--- a/html-router/assets/icon/favicon-16x16.png
+++ b/html-router/assets/icon/favicon-16x16.png
--- a/html-router/assets/icon/favicon-32x32.png
+++ b/html-router/assets/icon/favicon-32x32.png
--- a/html-router/assets/icon/favicon.ico
+++ b/html-router/assets/icon/favicon.ico
--- a/html-router/assets/knowledge-graph.js
+++ b/html-router/assets/knowledge-graph.js
@@ -0,0 +1,425 @@
+// Knowledge graph renderer: interactive 2D force graph with
+// zoom/pan, search, neighbor highlighting, curved links with arrows,
+// responsive resize, and type/relationship legends.
+(function () {
+  const D3_SRC = '/assets/d3.min.js';
+
+  let d3Loading = null;
+
+  function ensureD3() {
+    if (window.d3) return Promise.resolve();
+    if (d3Loading) return d3Loading;
+    d3Loading = new Promise((resolve, reject) => {
+      const s = document.createElement('script');
+      s.src = D3_SRC;
+      s.async = true;
+      s.onload = () => resolve();
+      s.onerror = () => reject(new Error('Failed to load D3'));
+      document.head.appendChild(s);
+    });
+    return d3Loading;
+  }
+
+  // Simple palettes (kept deterministic across renders)
+  const PALETTE_A = ['#60A5FA', '#34D399', '#F59E0B', '#A78BFA', '#F472B6', '#F87171', '#22D3EE', '#84CC16', '#FB7185'];
+  const PALETTE_B = ['#94A3B8', '#A3A3A3', '#9CA3AF', '#C084FC', '#FDA4AF', '#FCA5A5', '#67E8F9', '#A3E635', '#FDBA74'];
+
+  function buildMap(values) {
+    const unique = Array.from(new Set(values.filter(Boolean)));
+    const map = new Map();
+    unique.forEach((v, i) => map.set(v, PALETTE_A[i % PALETTE_A.length]));
+    return map;
+  }
+
+  function linkColorMap(values) {
+    const unique = Array.from(new Set(values.filter(Boolean)));
+    const map = new Map();
+    unique.forEach((v, i) => map.set(v, PALETTE_B[i % PALETTE_B.length]));
+    return map;
+  }
+
+  function radiusForDegree(deg) {
+    const d = Math.max(0, +deg || 0);
+    const r = 6 + Math.sqrt(d) * 3; // gentle growth
+    return Math.max(6, Math.min(r, 24));
+  }
+
+  function curvedPath(d) {
+    const sx = d.source.x, sy = d.source.y, tx = d.target.x, ty = d.target.y;
+    const dx = tx - sx, dy = ty - sy;
+    const dr = Math.hypot(dx, dy) * 0.7; // curve radius
+    const mx = (sx + tx) / 2;
+    const my = (sy + ty) / 2;
+    // Offset normal to create a consistent arc
+    const nx = -dy / (Math.hypot(dx, dy) || 1);
+    const ny = dx / (Math.hypot(dx, dy) || 1);
+    const cx = mx + nx * 20;
+    const cy = my + ny * 20;
+    return `M ${sx},${sy} Q ${cx},${cy} ${tx},${ty}`;
+  }
+
+  function buildAdjacency(nodes, links) {
+    const idToNode = new Map(nodes.map(n => [n.id, n]));
+    const neighbors = new Map();
+    nodes.forEach(n => neighbors.set(n.id, new Set()));
+    links.forEach(l => {
+      const s = typeof l.source === 'object' ? l.source.id : l.source;
+      const t = typeof l.target === 'object' ? l.target.id : l.target;
+      if (neighbors.has(s)) neighbors.get(s).add(t);
+      if (neighbors.has(t)) neighbors.get(t).add(s);
+    });
+    return { idToNode, neighbors };
+  }
+
+  function attachOverlay(container, { onSearch, onToggleNames, onToggleEdgeLabels, onCenter }) {
+    const overlay = document.createElement('div');
+    overlay.className = 'kg-overlay';
+
+    const primaryRow = document.createElement('div');
+    primaryRow.className = 'kg-control-row kg-control-row-primary';
+
+    const secondaryRow = document.createElement('div');
+    secondaryRow.className = 'kg-control-row kg-control-row-secondary';
+
+    // search box
+    const input = document.createElement('input');
+    input.type = 'text';
+    input.placeholder = 'Search nodes…';
+    input.className = 'nb-input kg-search-input';
+    input.addEventListener('keydown', (e) => {
+      if (e.key === 'Enter') onSearch && onSearch(input.value.trim());
+    });
+
+    const searchBtn = document.createElement('button');
+    searchBtn.className = 'nb-btn btn-xs nb-cta kg-search-btn';
+    searchBtn.textContent = 'Go';
+    searchBtn.addEventListener('click', () => onSearch && onSearch(input.value.trim()));
+
+    const namesToggle = document.createElement('button');
+    namesToggle.className = 'nb-btn btn-xs kg-toggle';
+    namesToggle.type = 'button';
+    namesToggle.textContent = 'Names';
+    namesToggle.addEventListener('click', () => onToggleNames && onToggleNames());
+
+    const labelToggle = document.createElement('button');
+    labelToggle.className = 'nb-btn btn-xs kg-toggle';
+    labelToggle.type = 'button';
+    labelToggle.textContent = 'Labels';
+    labelToggle.addEventListener('click', () => onToggleEdgeLabels && onToggleEdgeLabels());
+
+    const centerBtn = document.createElement('button');
+    centerBtn.className = 'nb-btn btn-xs';
+    centerBtn.textContent = 'Center';
+    centerBtn.addEventListener('click', () => onCenter && onCenter());
+
+    primaryRow.appendChild(input);
+    primaryRow.appendChild(searchBtn);
+
+    secondaryRow.appendChild(namesToggle);
+    secondaryRow.appendChild(labelToggle);
+    secondaryRow.appendChild(centerBtn);
+
+    overlay.appendChild(primaryRow);
+    overlay.appendChild(secondaryRow);
+
+    container.style.position = 'relative';
+    container.appendChild(overlay);
+
+    return { input, overlay, namesToggle, labelToggle };
+  }
+
+  function attachLegends(container, typeColor, relColor) {
+    const wrap = document.createElement('div');
+    wrap.className = 'kg-legend';
+
+    function section(title, items) {
+      const sec = document.createElement('div');
+      sec.className = 'nb-card kg-legend-card';
+      const h = document.createElement('div'); h.className = 'kg-legend-heading'; h.textContent = title; sec.appendChild(h);
+      items.forEach(([label, color]) => {
+        const row = document.createElement('div'); row.className = 'kg-legend-row';
+        const sw = document.createElement('span'); sw.style.background = color; sw.style.width = '12px'; sw.style.height = '12px'; sw.style.border = '2px solid #000';
+        const t = document.createElement('span'); t.textContent = label || '—';
+        row.appendChild(sw); row.appendChild(t); sec.appendChild(row);
+      });
+      return sec;
+    }
+
+    const typeItems = Array.from(typeColor.entries());
+    if (typeItems.length) wrap.appendChild(section('Entity Type', typeItems));
+    const relItems = Array.from(relColor.entries());
+    if (relItems.length) wrap.appendChild(section('Relationship', relItems));
+
+    container.appendChild(wrap);
+    return wrap;
+  }
+
+  async function renderKnowledgeGraph(root) {
+    const container = (root || document).querySelector('#knowledge-graph');
+    if (!container) return;
+
+    await ensureD3().catch(() => {
+      const err = document.createElement('div');
+      err.className = 'alert alert-error';
+      err.textContent = 'Unable to load graph library (D3).';
+      container.appendChild(err);
+    });
+    if (!window.d3) return;
+
+    // Clear previous render
+    container.innerHTML = '';
+
+    const width = container.clientWidth || 800;
+    const height = container.clientHeight || 600;
+
+    const et = container.dataset.entityType || '';
+    const cc = container.dataset.contentCategory || '';
+    const qs = new URLSearchParams();
+    if (et) qs.set('entity_type', et);
+    if (cc) qs.set('content_category', cc);
+
+    const url = '/knowledge/graph.json' + (qs.toString() ? ('?' + qs.toString()) : '');
+    let data;
+    try {
+      const res = await fetch(url, { headers: { 'Accept': 'application/json' } });
+      if (!res.ok) throw new Error('Failed to load graph data');
+      data = await res.json();
+    } catch (_e) {
+      const err = document.createElement('div');
+      err.className = 'alert alert-error';
+      err.textContent = 'Unable to load graph data.';
+      container.appendChild(err);
+      return;
+    }
+
+    // Color maps
+    const typeColor = buildMap(data.nodes.map(n => n.entity_type));
+    const relColor = linkColorMap(data.links.map(l => l.relationship_type));
+    const { neighbors } = buildAdjacency(data.nodes, data.links);
+
+    // Build overlay controls
+    let namesVisible = true;
+    let edgeLabelsVisible = true;
+
+    const togglePressedState = (button, state) => {
+      if (!button) return;
+      button.setAttribute('aria-pressed', state ? 'true' : 'false');
+      button.classList.toggle('kg-toggle-active', !!state);
+    };
+
+    const { input, namesToggle, labelToggle } = attachOverlay(container, {
+      onSearch: (q) => focusSearch(q),
+      onToggleNames: () => {
+        namesVisible = !namesVisible;
+        label.style('display', namesVisible ? null : 'none');
+        togglePressedState(namesToggle, namesVisible);
+      },
+      onToggleEdgeLabels: () => {
+        edgeLabelsVisible = !edgeLabelsVisible;
+        linkLabel.style('display', edgeLabelsVisible ? null : 'none');
+        togglePressedState(labelToggle, edgeLabelsVisible);
+      },
+      onCenter: () => zoomTo(1, [width / 2, height / 2])
+    });
+
+    togglePressedState(namesToggle, namesVisible);
+    togglePressedState(labelToggle, edgeLabelsVisible);
+
+    // SVG + zoom
+    const svg = d3.select(container)
+      .append('svg')
+      .attr('width', '100%')
+      .attr('height', height)
+      .attr('viewBox', [0, 0, width, height])
+      .attr('style', 'cursor: grab; touch-action: none; background: transparent;')
+      .call(d3.zoom().scaleExtent([0.25, 5]).on('zoom', (event) => {
+        g.attr('transform', event.transform);
+      }));
+
+    const g = svg.append('g');
+
+    // Defs for arrows
+    const defs = svg.append('defs');
+    const markerFor = (key, color) => {
+      const id = `arrow-${key.replace(/[^a-z0-9_-]/gi, '_')}`;
+      if (!document.getElementById(id)) {
+        defs.append('marker')
+          .attr('id', id)
+          .attr('viewBox', '0 -5 10 10')
+          .attr('refX', 16)
+          .attr('refY', 0)
+          .attr('markerWidth', 6)
+          .attr('markerHeight', 6)
+          .attr('orient', 'auto')
+          .append('path')
+          .attr('d', 'M0,-5L10,0L0,5')
+          .attr('fill', color);
+      }
+      return `url(#${id})`;
+    };
+
+    // Forces
+    const linkForce = d3.forceLink(data.links)
+      .id(d => d.id)
+      .distance(l => 70)
+      .strength(0.5);
+
+    const simulation = d3.forceSimulation(data.nodes)
+      .force('link', linkForce)
+      .force('charge', d3.forceManyBody().strength(-220))
+      .force('center', d3.forceCenter(width / 2, height / 2))
+      .force('collision', d3.forceCollide().radius(d => radiusForDegree(d.degree) + 6))
+      .force('y', d3.forceY(height / 2).strength(0.02))
+      .force('x', d3.forceX(width / 2).strength(0.02));
+
+    // Links as paths so we can curve + arrow
+    const link = g.append('g')
+      .attr('fill', 'none')
+      .attr('stroke-opacity', 0.7)
+      .selectAll('path')
+      .data(data.links)
+      .join('path')
+      .attr('stroke', d => relColor.get(d.relationship_type) || '#CBD5E1')
+      .attr('stroke-width', 1.5)
+      .attr('marker-end', d => markerFor(d.relationship_type || 'rel', relColor.get(d.relationship_type) || '#CBD5E1'));
+
+    // Optional edge labels (midpoint)
+    const linkLabel = g.append('g')
+      .selectAll('text')
+      .data(data.links)
+      .join('text')
+      .attr('font-size', 9)
+      .attr('fill', '#475569')
+      .attr('text-anchor', 'middle')
+      .attr('opacity', 0.7)
+      .text(d => d.relationship_type || '');
+
+    // Nodes
+    const node = g.append('g')
+      .attr('stroke', '#fff')
+      .attr('stroke-width', 1.5)
+      .selectAll('circle')
+      .data(data.nodes)
+      .join('circle')
+      .attr('r', d => radiusForDegree(d.degree))
+      .attr('fill', d => typeColor.get(d.entity_type) || '#94A3B8')
+      .attr('cursor', 'pointer')
+      .on('mouseenter', function (_evt, d) { setHighlight(d); })
+      .on('mouseleave', function () { clearHighlight(); })
+      .on('click', function (_evt, d) {
+        // pin/unpin on click
+        if (d.fx == null) { d.fx = d.x; d.fy = d.y; this.setAttribute('data-pinned', 'true'); }
+        else { d.fx = null; d.fy = null; this.removeAttribute('data-pinned'); }
+      })
+      .call(d3.drag()
+        .on('start', (event, d) => {
+          if (!event.active) simulation.alphaTarget(0.3).restart();
+          d.fx = d.x; d.fy = d.y;
+        })
+        .on('drag', (event, d) => { d.fx = event.x; d.fy = event.y; })
+        .on('end', (event, d) => { if (!event.active) simulation.alphaTarget(0); }));
+
+    node.append('title').text(d => `${d.name} • ${d.entity_type} • deg ${d.degree}`);
+
+    // Labels
+    const label = g.append('g')
+      .selectAll('text')
+      .data(data.nodes)
+      .join('text')
+      .text(d => d.name)
+      .attr('font-size', 11)
+      .attr('fill', '#111827')
+      .attr('stroke', 'white')
+      .attr('paint-order', 'stroke')
+      .attr('stroke-width', 3)
+      .attr('dx', d => radiusForDegree(d.degree) + 6)
+      .attr('dy', 4);
+
+    // Legends
+    attachLegends(container, typeColor, relColor);
+
+    // Highlight logic
+    function setHighlight(n) {
+      const ns = neighbors.get(n.id) || new Set();
+      node.attr('opacity', d => (d.id === n.id || ns.has(d.id)) ? 1 : 0.15);
+      label.attr('opacity', d => (d.id === n.id || ns.has(d.id)) ? 1 : 0.15);
+      link
+        .attr('stroke-opacity', d => {
+          const s = (typeof d.source === 'object') ? d.source.id : d.source;
+          const t = (typeof d.target === 'object') ? d.target.id : d.target;
+          return (s === n.id || t === n.id || (ns.has(s) && ns.has(t))) ? 0.9 : 0.05;
+        })
+        .attr('marker-end', d => {
+          const c = relColor.get(d.relationship_type) || '#CBD5E1';
+          return markerFor(d.relationship_type || 'rel', c);
+        });
+      linkLabel.attr('opacity', d => {
+        const s = (typeof d.source === 'object') ? d.source.id : d.source;
+        const t = (typeof d.target === 'object') ? d.target.id : d.target;
+        return (s === n.id || t === n.id) ? 0.9 : 0.05;
+      });
+    }
+    function clearHighlight() {
+      node.attr('opacity', 1);
+      label.attr('opacity', 1);
+      link.attr('stroke-opacity', 0.7);
+      linkLabel.attr('opacity', 0.7);
+    }
+
+    // Search + center helpers
+    function centerOnNode(n) {
+      const k = 1.5; // zoom factor
+      const x = n.x, y = n.y;
+      const transform = d3.zoomIdentity.translate(width / 2 - k * x, height / 2 - k * y).scale(k);
+      svg.transition().duration(350).call(zoom.transform, transform);
+    }
+    function focusSearch(query) {
+      if (!query) return;
+      const q = query.toLowerCase();
+      const found = data.nodes.find(n => (n.name || '').toLowerCase().includes(q));
+      if (found) { setHighlight(found); centerOnNode(found); }
+    }
+
+    // Expose zoom instance
+    const zoom = d3.zoom().scaleExtent([0.25, 5]).on('zoom', (event) => g.attr('transform', event.transform));
+    svg.call(zoom);
+    function zoomTo(k, center) {
+      const transform = d3.zoomIdentity.translate(width / 2 - k * center[0], height / 2 - k * center[1]).scale(k);
+      svg.transition().duration(250).call(zoom.transform, transform);
+    }
+
+    // Tick update
+    simulation.on('tick', () => {
+      link.attr('d', curvedPath);
+      node.attr('cx', d => d.x).attr('cy', d => d.y);
+      label.attr('x', d => d.x).attr('y', d => d.y);
+      linkLabel.attr('x', d => (d.source.x + d.target.x) / 2).attr('y', d => (d.source.y + d.target.y) / 2);
+    });
+
+    // Resize handling
+    const ro = new ResizeObserver(() => {
+      const w = container.clientWidth || width;
+      const h = container.clientHeight || height;
+      svg.attr('viewBox', [0, 0, w, h]).attr('height', h);
+      simulation.force('center', d3.forceCenter(w / 2, h / 2));
+      simulation.alpha(0.3).restart();
+    });
+    ro.observe(container);
+  }
+
+  function tryRender(root) {
+    const container = (root || document).querySelector('#knowledge-graph');
+    if (container) renderKnowledgeGraph(root);
+  }
+
+  // Expose for debugging/manual re-render
+  window.renderKnowledgeGraph = () => renderKnowledgeGraph(document);
+
+  // Full page load
+  document.addEventListener('DOMContentLoaded', () => tryRender(document));
+
+  // HTMX partial swaps
+  document.body.addEventListener('htmx:afterSettle', (evt) => {
+    tryRender(evt && evt.target ? evt.target : document);
+  });
+})();
--- a/html-router/assets/style.css
+++ b/html-router/assets/style.css
--- a/html-router/assets/toast.js
+++ b/html-router/assets/toast.js
@@ -6,33 +6,31 @@
        return;
      }
      const alert = document.createElement('div');
-      // Base classes for the alert
-      alert.className = `alert alert-${type} mt-2 shadow-md flex flex-col text-start`;
+      alert.className = `alert toast-alert alert-${type}`;
+      alert.style.opacity = '1';
+      alert.style.transition = 'opacity 0.5s ease-out';

-      // Build inner HTML based on whether title is provided
-      let innerHTML = '';
      if (title) {
-        innerHTML += `<div class="font-bold text-lg">${title}</div>`; // Title element
-        innerHTML += `<div>${description}</div>`; // Description element
-      } else {
-        // Structure without title
-        innerHTML += `<span>${description}</span>`;
+        const titleEl = document.createElement('div');
+        titleEl.className = 'toast-alert-title';
+        titleEl.textContent = title;
+        alert.appendChild(titleEl);
      }

-      alert.innerHTML = innerHTML;
+      const bodyEl = document.createElement(title ? 'div' : 'span');
+      bodyEl.textContent = description;
+      alert.appendChild(bodyEl);
+
      container.appendChild(alert);

      // Auto-remove after a delay
      setTimeout(() => {
-        // Optional: Add fade-out effect
        alert.style.opacity = '0';
-        alert.style.transition = 'opacity 0.5s ease-out';
-        setTimeout(() => alert.remove(), 500); // Remove after fade
-      }, 3000); // Start fade-out after 3 seconds
+        setTimeout(() => alert.remove(), 500);
+      }, 3000);
    };

    document.body.addEventListener('toast', function (event) {
-      console.log(event);
      // Extract data from the event detail, matching the Rust payload
      const detail = event.detail;
      if (detail && detail.description) {
@@ -54,4 +52,3 @@
      if (container) container.innerHTML = '';
    });
  })
-
--- a/html-router/package.json
+++ b/html-router/package.json
@@ -1,7 +1,6 @@
 {
  "name": "html-router",
  "version": "1.0.0",
-  "main": "tailwind.config.js",
  "scripts": {
    "tailwind": "npx @tailwindcss/cli -i app.css -o assets/style.css -w -m"
  },
@@ -14,4 +13,4 @@
    "daisyui": "^5.0.12",
    "tailwindcss": "^4.1.2"
  }
-}
+}
--- a/html-router/src/lib.rs
+++ b/html-router/src/lib.rs
@@ -2,6 +2,7 @@ pub mod html_state;
 pub mod middlewares;
 pub mod router_factory;
 pub mod routes;
+pub mod utils;

 use axum::{extract::FromRef, Router};
 use axum_session::{Session, SessionStore};
@@ -35,5 +36,7 @@ where
        .add_protected_routes(routes::content::router())
        .add_protected_routes(routes::knowledge::router())
        .add_protected_routes(routes::ingestion::router())
+        .add_protected_routes(routes::scratchpad::router())
+        .with_compression()
        .build()
 }
--- a/html-router/src/middlewares/analytics_middleware.rs
+++ b/html-router/src/middlewares/analytics_middleware.rs
@@ -1,5 +1,6 @@
 use axum::{
    extract::{Request, State},
+    http::Method,
    middleware::Next,
    response::Response,
 };
@@ -19,7 +20,8 @@ where
    S: ProvidesDb + Clone + Send + Sync + 'static,
 {
    let path = request.uri().path();
-    if !path.starts_with("/assets") && !path.contains('.') {
+    // Only count visits/page loads for GET requests to non-asset, non-static paths
+    if request.method() == Method::GET && !path.starts_with("/assets") && !path.contains('.') {
        if !session.get::<bool>("counted_visitor").unwrap_or(false) {
            let _ = Analytics::increment_visitors(state.db()).await;
            session.set("counted_visitor", true);
--- a/html-router/src/middlewares/compression.rs
+++ b/html-router/src/middlewares/compression.rs
@@ -0,0 +1,7 @@
+use tower_http::compression::CompressionLayer;
+
+/// Provides a default compression layer that negotiates encoding based on the
+/// `Accept-Encoding` header of the incoming request.
+pub fn compression_layer() -> CompressionLayer {
+    CompressionLayer::new()
+}
--- a/html-router/src/middlewares/mod.rs
+++ b/html-router/src/middlewares/mod.rs
@@ -1,3 +1,4 @@
 pub mod analytics_middleware;
 pub mod auth_middleware;
+pub mod compression;
 pub mod response_middleware;
--- a/html-router/src/middlewares/response_middleware.rs
+++ b/html-router/src/middlewares/response_middleware.rs
@@ -188,7 +188,7 @@ where
                if is_htmx {
                    (StatusCode::OK, [(axum_htmx::HX_REDIRECT, path)], "").into_response()
                } else {
-                    Redirect::to(&path).into_response()
+                    Redirect::to(path).into_response()
                }
            }
        }
@@ -205,26 +205,26 @@ pub enum HtmlError {

 impl From<AppError> for HtmlError {
    fn from(err: AppError) -> Self {
-        HtmlError::AppError(err)
+        Self::AppError(err)
    }
 }

 impl From<surrealdb::Error> for HtmlError {
    fn from(err: surrealdb::Error) -> Self {
-        HtmlError::AppError(AppError::from(err))
+        Self::AppError(AppError::from(err))
    }
 }

 impl From<minijinja::Error> for HtmlError {
    fn from(err: minijinja::Error) -> Self {
-        HtmlError::TemplateError(err.to_string())
+        Self::TemplateError(err.to_string())
    }
 }

 impl IntoResponse for HtmlError {
    fn into_response(self) -> Response {
        match self {
-            HtmlError::AppError(err) => match err {
+            Self::AppError(err) => match err {
                AppError::NotFound(_) => TemplateResponse::not_found().into_response(),
                AppError::Auth(_) => TemplateResponse::unauthorized().into_response(),
                AppError::Validation(msg) => TemplateResponse::bad_request(&msg).into_response(),
@@ -233,7 +233,7 @@ impl IntoResponse for HtmlError {
                    TemplateResponse::server_error().into_response()
                }
            },
-            HtmlError::TemplateError(err) => {
+            Self::TemplateError(err) => {
                error!("Template error: {}", err);
                TemplateResponse::server_error().into_response()
            }
--- a/html-router/src/router_factory.rs
+++ b/html-router/src/router_factory.rs
@@ -13,7 +13,7 @@ use crate::{
    html_state::HtmlState,
    middlewares::{
        analytics_middleware::analytics_middleware, auth_middleware::require_auth,
-        response_middleware::with_template_response,
+        compression::compression_layer, response_middleware::with_template_response,
    },
 };

@@ -48,6 +48,7 @@ pub struct RouterFactory<S> {
    nested_protected_routes: Vec<(String, Router<S>)>,
    custom_middleware: MiddleWareVecType<S>,
    public_assets_config: Option<AssetsConfig>,
+    compression_enabled: bool,
 }

 struct AssetsConfig {
@@ -69,6 +70,7 @@ where
            nested_protected_routes: Vec::new(),
            custom_middleware: Vec::new(),
            public_assets_config: None,
+            compression_enabled: false,
        }
    }

@@ -115,6 +117,12 @@ where
        self
    }

+    /// Enables response compression when building the router.
+    pub const fn with_compression(mut self) -> Self {
+        self.compression_enabled = true;
+        self
+    }
+
    pub fn build(self) -> Router<S> {
        // Start with an empty router
        let mut public_router = Router::new();
@@ -169,21 +177,26 @@ where
        }

        // Apply common middleware
+        router = router.layer(from_fn_with_state(
+            self.app_state.clone(),
+            analytics_middleware::<HtmlState>,
+        ));
+        router = router.layer(map_response_with_state(
+            self.app_state.clone(),
+            with_template_response::<HtmlState>,
+        ));
+        router = router.layer(
+            AuthSessionLayer::<User, String, SessionSurrealPool<Any>, Surreal<Any>>::new(Some(
+                self.app_state.db.client.clone(),
+            ))
+            .with_config(AuthConfig::<String>::default()),
+        );
+        router = router.layer(SessionLayer::new((*self.app_state.session_store).clone()));
+
+        if self.compression_enabled {
+            router = router.layer(compression_layer());
+        }
+
        router
-            .layer(from_fn_with_state(
-                self.app_state.clone(),
-                analytics_middleware::<HtmlState>,
-            ))
-            .layer(map_response_with_state(
-                self.app_state.clone(),
-                with_template_response::<HtmlState>,
-            ))
-            .layer(
-                AuthSessionLayer::<User, String, SessionSurrealPool<Any>, Surreal<Any>>::new(Some(
-                    self.app_state.db.client.clone(),
-                ))
-                .with_config(AuthConfig::<String>::default()),
-            )
-            .layer(SessionLayer::new((*self.app_state.session_store).clone()))
    }
 }
--- a/html-router/src/routes/account/handlers.rs
+++ b/html-router/src/routes/account/handlers.rs
@@ -24,7 +24,10 @@ pub async fn show_account_page(
    RequireUser(user): RequireUser,
    State(state): State<HtmlState>,
 ) -> Result<impl IntoResponse, HtmlError> {
-    let timezones = TZ_VARIANTS.iter().map(|tz| tz.to_string()).collect();
+    let timezones = TZ_VARIANTS
+        .iter()
+        .map(std::string::ToString::to_string)
+        .collect();
    let conversation_archive = User::get_user_conversations(&user.id, &state.db).await?;

    Ok(TemplateResponse::new_template(
@@ -102,7 +105,10 @@ pub async fn update_timezone(
        ..user.clone()
    };

-    let timezones = TZ_VARIANTS.iter().map(|tz| tz.to_string()).collect();
+    let timezones = TZ_VARIANTS
+        .iter()
+        .map(std::string::ToString::to_string)
+        .collect();

    // Render the API key section block
    Ok(TemplateResponse::new_partial(
--- a/html-router/src/routes/admin/handlers.rs
+++ b/html-router/src/routes/admin/handlers.rs
@@ -1,13 +1,23 @@
+use async_openai::types::ListModelResponse;
 use axum::{extract::State, response::IntoResponse, Form};
 use serde::{Deserialize, Serialize};

-use common::storage::types::{
-    analytics::Analytics,
-    conversation::Conversation,
-    system_prompts::{DEFAULT_INGRESS_ANALYSIS_SYSTEM_PROMPT, DEFAULT_QUERY_SYSTEM_PROMPT},
-    system_settings::SystemSettings,
-    user::User,
+use common::{
+    error::AppError,
+    storage::types::{
+        analytics::Analytics,
+        conversation::Conversation,
+        knowledge_entity::KnowledgeEntity,
+        system_prompts::{
+            DEFAULT_IMAGE_PROCESSING_PROMPT, DEFAULT_INGRESS_ANALYSIS_SYSTEM_PROMPT,
+            DEFAULT_QUERY_SYSTEM_PROMPT,
+        },
+        system_settings::SystemSettings,
+        text_chunk::TextChunk,
+        user::User,
+    },
 };
+use tracing::{error, info};

 use crate::{
    html_state::HtmlState,
@@ -24,27 +34,41 @@ pub struct AdminPanelData {
    analytics: Analytics,
    users: i64,
    default_query_prompt: String,
+    default_image_prompt: String,
    conversation_archive: Vec<Conversation>,
+    available_models: ListModelResponse,
 }

 pub async fn show_admin_panel(
    State(state): State<HtmlState>,
    RequireUser(user): RequireUser,
 ) -> Result<impl IntoResponse, HtmlError> {
-    let settings = SystemSettings::get_current(&state.db).await?;
-    let analytics = Analytics::get_current(&state.db).await?;
-    let users_count = Analytics::get_users_amount(&state.db).await?;
-    let conversation_archive = User::get_user_conversations(&user.id, &state.db).await?;
+    let (
+        settings_res,
+        analytics_res,
+        user_count_res,
+        conversation_archive_res,
+        available_models_res,
+    ) = tokio::join!(
+        SystemSettings::get_current(&state.db),
+        Analytics::get_current(&state.db),
+        Analytics::get_users_amount(&state.db),
+        User::get_user_conversations(&user.id, &state.db),
+        async { state.openai_client.models().list().await }
+    );

    Ok(TemplateResponse::new_template(
        "admin/base.html",
        AdminPanelData {
            user,
-            settings,
-            analytics,
-            users: users_count,
+            settings: settings_res?,
+            analytics: analytics_res?,
+            available_models: available_models_res
+                .map_err(|e| AppError::InternalError(e.to_string()))?,
+            users: user_count_res?,
            default_query_prompt: DEFAULT_QUERY_SYSTEM_PROMPT.to_string(),
-            conversation_archive,
+            default_image_prompt: DEFAULT_IMAGE_PROCESSING_PROMPT.to_string(),
+            conversation_archive: conversation_archive_res?,
        },
    ))
 }
@@ -79,7 +103,7 @@ pub async fn toggle_registration_status(
    // Early return if the user is not admin
    if !user.admin {
        return Ok(TemplateResponse::redirect("/"));
-    };
+    }

    let current_settings = SystemSettings::get_current(&state.db).await?;

@@ -103,11 +127,16 @@ pub async fn toggle_registration_status(
 pub struct ModelSettingsInput {
    query_model: String,
    processing_model: String,
+    image_processing_model: String,
+    voice_processing_model: String,
+    embedding_model: String,
+    embedding_dimensions: Option<u32>,
 }

 #[derive(Serialize)]
 pub struct ModelSettingsData {
    settings: SystemSettings,
+    available_models: ListModelResponse,
 }

 pub async fn update_model_settings(
@@ -118,23 +147,81 @@ pub async fn update_model_settings(
    // Early return if the user is not admin
    if !user.admin {
        return Ok(TemplateResponse::redirect("/"));
-    };
+    }

    let current_settings = SystemSettings::get_current(&state.db).await?;

+    // Determine if re-embedding is required
+    let reembedding_needed = input
+        .embedding_dimensions
+        .is_some_and(|new_dims| new_dims != current_settings.embedding_dimensions);
+
    let new_settings = SystemSettings {
        query_model: input.query_model,
        processing_model: input.processing_model,
-        ..current_settings
+        image_processing_model: input.image_processing_model,
+        voice_processing_model: input.voice_processing_model,
+        embedding_model: input.embedding_model,
+        // Use new dimensions if provided, otherwise retain the current ones.
+        embedding_dimensions: input
+            .embedding_dimensions
+            .unwrap_or(current_settings.embedding_dimensions),
+        ..current_settings.clone()
    };

    SystemSettings::update(&state.db, new_settings.clone()).await?;

+    if reembedding_needed {
+        info!("Embedding dimensions changed. Spawning background re-embedding task...");
+
+        let db_for_task = state.db.clone();
+        let openai_for_task = state.openai_client.clone();
+        let new_model_for_task = new_settings.embedding_model.clone();
+        let new_dims_for_task = new_settings.embedding_dimensions;
+
+        tokio::spawn(async move {
+            // First, update all text chunks
+            if let Err(e) = TextChunk::update_all_embeddings(
+                &db_for_task,
+                &openai_for_task,
+                &new_model_for_task,
+                new_dims_for_task,
+            )
+            .await
+            {
+                error!("Background re-embedding task failed for TextChunks: {}", e);
+            }
+
+            // Second, update all knowledge entities
+            if let Err(e) = KnowledgeEntity::update_all_embeddings(
+                &db_for_task,
+                &openai_for_task,
+                &new_model_for_task,
+                new_dims_for_task,
+            )
+            .await
+            {
+                error!(
+                    "Background re-embedding task failed for KnowledgeEntities: {}",
+                    e
+                );
+            }
+        });
+    }
+
+    let available_models = state
+        .openai_client
+        .models()
+        .list()
+        .await
+        .map_err(|_e| AppError::InternalError("Failed to get models".to_string()))?;
+
    Ok(TemplateResponse::new_partial(
        "admin/base.html",
        "model_settings_form",
        ModelSettingsData {
            settings: new_settings,
+            available_models,
        },
    ))
 }
@@ -152,7 +239,7 @@ pub async fn show_edit_system_prompt(
    // Early return if the user is not admin
    if !user.admin {
        return Ok(TemplateResponse::redirect("/"));
-    };
+    }

    let settings = SystemSettings::get_current(&state.db).await?;

@@ -183,7 +270,7 @@ pub async fn patch_query_prompt(
    // Early return if the user is not admin
    if !user.admin {
        return Ok(TemplateResponse::redirect("/"));
-    };
+    }

    let current_settings = SystemSettings::get_current(&state.db).await?;

@@ -216,7 +303,7 @@ pub async fn show_edit_ingestion_prompt(
    // Early return if the user is not admin
    if !user.admin {
        return Ok(TemplateResponse::redirect("/"));
-    };
+    }

    let settings = SystemSettings::get_current(&state.db).await?;

@@ -242,7 +329,7 @@ pub async fn patch_ingestion_prompt(
    // Early return if the user is not admin
    if !user.admin {
        return Ok(TemplateResponse::redirect("/"));
-    };
+    }

    let current_settings = SystemSettings::get_current(&state.db).await?;

@@ -261,3 +348,62 @@ pub async fn patch_ingestion_prompt(
        },
    ))
 }
+
+#[derive(Serialize)]
+pub struct ImagePromptEditData {
+    settings: SystemSettings,
+    default_image_prompt: String,
+}
+
+pub async fn show_edit_image_prompt(
+    State(state): State<HtmlState>,
+    RequireUser(user): RequireUser,
+) -> Result<impl IntoResponse, HtmlError> {
+    // Early return if the user is not admin
+    if !user.admin {
+        return Ok(TemplateResponse::redirect("/"));
+    }
+
+    let settings = SystemSettings::get_current(&state.db).await?;
+
+    Ok(TemplateResponse::new_template(
+        "admin/edit_image_prompt_modal.html",
+        ImagePromptEditData {
+            settings,
+            default_image_prompt: DEFAULT_IMAGE_PROCESSING_PROMPT.to_string(),
+        },
+    ))
+}
+
+#[derive(Deserialize)]
+pub struct ImagePromptUpdateInput {
+    image_processing_prompt: String,
+}
+
+pub async fn patch_image_prompt(
+    State(state): State<HtmlState>,
+    RequireUser(user): RequireUser,
+    Form(input): Form<ImagePromptUpdateInput>,
+) -> Result<impl IntoResponse, HtmlError> {
+    // Early return if the user is not admin
+    if !user.admin {
+        return Ok(TemplateResponse::redirect("/"));
+    }
+
+    let current_settings = SystemSettings::get_current(&state.db).await?;
+
+    let new_settings = SystemSettings {
+        image_processing_prompt: input.image_processing_prompt,
+        ..current_settings.clone()
+    };
+
+    SystemSettings::update(&state.db, new_settings.clone()).await?;
+
+    Ok(TemplateResponse::new_partial(
+        "admin/base.html",
+        "system_prompt_section",
+        SystemPromptSectionData {
+            settings: new_settings,
+        },
+    ))
+}
--- a/html-router/src/routes/admin/mod.rs
+++ b/html-router/src/routes/admin/mod.rs
@@ -5,8 +5,9 @@ use axum::{
    Router,
 };
 use handlers::{
-    patch_ingestion_prompt, patch_query_prompt, show_admin_panel, show_edit_ingestion_prompt,
-    show_edit_system_prompt, toggle_registration_status, update_model_settings,
+    patch_image_prompt, patch_ingestion_prompt, patch_query_prompt, show_admin_panel,
+    show_edit_image_prompt, show_edit_ingestion_prompt, show_edit_system_prompt,
+    toggle_registration_status, update_model_settings,
 };

 use crate::html_state::HtmlState;
@@ -24,4 +25,6 @@ where
        .route("/update-query-prompt", patch(patch_query_prompt))
        .route("/edit-ingestion-prompt", get(show_edit_ingestion_prompt))
        .route("/update-ingestion-prompt", patch(patch_ingestion_prompt))
+        .route("/edit-image-prompt", get(show_edit_image_prompt))
+        .route("/update-image-prompt", patch(patch_image_prompt))
 }
--- a/html-router/src/routes/auth/signin.rs
+++ b/html-router/src/routes/auth/signin.rs
@@ -27,13 +27,14 @@ pub async fn show_signin_form(
    if auth.is_authenticated() {
        return Ok(TemplateResponse::redirect("/"));
    }
-    match boosted {
-        true => Ok(TemplateResponse::new_partial(
+    if boosted {
+        Ok(TemplateResponse::new_partial(
            "auth/signin_base.html",
            "body",
            (),
-        )),
-        false => Ok(TemplateResponse::new_template("auth/signin_base.html", ())),
+        ))
+    } else {
+        Ok(TemplateResponse::new_template("auth/signin_base.html", ()))
    }
 }

--- a/html-router/src/routes/auth/signup.rs
+++ b/html-router/src/routes/auth/signup.rs
@@ -29,13 +29,14 @@ pub async fn show_signup_form(
        return Ok(TemplateResponse::redirect("/"));
    }

-    match boosted {
-        true => Ok(TemplateResponse::new_partial(
+    if boosted {
+        Ok(TemplateResponse::new_partial(
            "auth/signup_form.html",
            "body",
            (),
-        )),
-        false => Ok(TemplateResponse::new_template("auth/signup_form.html", ())),
+        ))
+    } else {
+        Ok(TemplateResponse::new_template("auth/signup_form.html", ()))
    }
 }

@@ -48,7 +49,7 @@ pub async fn process_signup_and_show_verification(
        Ok(user) => user,
        Err(e) => {
            tracing::error!("{:?}", e);
-            return Ok(Html(format!("<p>{}</p>", e)).into_response());
+            return Ok(Html(format!("<p>{e}</p>")).into_response());
        }
    };

--- a/html-router/src/routes/chat/chat_handlers.rs
+++ b/html-router/src/routes/chat/chat_handlers.rs
@@ -137,7 +137,7 @@ pub async fn show_existing_chat(
        ChatPageData {
            history: messages,
            user,
-            conversation: Some(conversation.clone()),
+            conversation: Some(conversation),
            conversation_archive,
        },
    ))
@@ -157,7 +157,7 @@ pub async fn new_user_message(

    if conversation.user_id != user.id {
        return Ok(TemplateResponse::unauthorized().into_response());
-    };
+    }

    let user_message = Message::new(conversation_id, MessageRole::User, form.content, None);

--- a/html-router/src/routes/chat/message_response_stream.rs
+++ b/html-router/src/routes/chat/message_response_stream.rs
@@ -9,11 +9,8 @@ use axum::{
    },
 };
 use composite_retrieval::{
-    answer_retrieval::{
-        create_chat_request, create_user_message_with_history, format_entities_json,
-        LLMResponseFormat,
-    },
-    retrieve_entities,
+    answer_retrieval::{create_chat_request, create_user_message_with_history, LLMResponseFormat},
+    retrieve_entities, retrieved_entities_to_json,
 };
 use futures::{
    stream::{self, once},
@@ -136,7 +133,7 @@ pub async fn get_response_stream(
    };

    // 3. Create the OpenAI request
-    let entities_json = format_entities_json(&entities);
+    let entities_json = retrieved_entities_to_json(&entities);
    let formatted_user_message =
        create_user_message_with_history(&entities_json, &history, &user_message.content);
    let settings = match SystemSettings::get_current(&state.db).await {
@@ -251,7 +248,7 @@ pub async fn get_response_stream(
                    Err(e) => {
                        yield Ok(Event::default()
                            .event("error")
-                            .data(format!("Stream error: {}", e)));
+                            .data(format!("Stream error: {e}")));
                    }
                }
            }
@@ -260,7 +257,11 @@ pub async fn get_response_stream(
        .chain(stream::once(async move {
            if let Some(message) = rx_final.recv().await {
                // Don't send any event if references is empty
-                if message.references.as_ref().is_some_and(|x| x.is_empty()) {
+                if message
+                    .references
+                    .as_ref()
+                    .is_some_and(std::vec::Vec::is_empty)
+                {
                    return Ok(Event::default().event("empty")); // This event won't be sent
                }

--- a/html-router/src/routes/content/handlers.rs
+++ b/html-router/src/routes/content/handlers.rs
@@ -3,11 +3,12 @@ use axum::{
    response::IntoResponse,
    Form,
 };
-use axum_htmx::{HxBoosted, HxRequest};
+use axum_htmx::{HxBoosted, HxRequest, HxTarget};
 use serde::{Deserialize, Serialize};

 use common::storage::types::{
-    conversation::Conversation, file_info::FileInfo, text_content::TextContent, user::User,
+    conversation::Conversation, file_info::FileInfo, knowledge_entity::KnowledgeEntity,
+    text_chunk::TextChunk, text_content::TextContent, user::User,
 };

 use crate::{
@@ -16,7 +17,12 @@ use crate::{
        auth_middleware::RequireUser,
        response_middleware::{HtmlError, TemplateResponse},
    },
+    utils::pagination::{paginate_items, Pagination},
+    utils::text_content_preview::truncate_text_contents,
 };
+use url::form_urlencoded;
+
+const CONTENTS_PER_PAGE: usize = 12;

 #[derive(Serialize)]
 pub struct ContentPageData {
@@ -25,11 +31,20 @@ pub struct ContentPageData {
    categories: Vec<String>,
    selected_category: Option<String>,
    conversation_archive: Vec<Conversation>,
+    pagination: Pagination,
+    page_query: String,
+}
+
+#[derive(Serialize)]
+pub struct RecentTextContentData {
+    pub user: User,
+    pub text_contents: Vec<TextContent>,
 }

 #[derive(Deserialize)]
 pub struct FilterParams {
    category: Option<String>,
+    page: Option<usize>,
 }

 pub async fn show_content_page(
@@ -40,17 +55,32 @@ pub async fn show_content_page(
    HxBoosted(is_boosted): HxBoosted,
 ) -> Result<impl IntoResponse, HtmlError> {
    // Normalize empty strings to None
-    let has_category_param = params.category.is_some();
-    let category_filter = params.category.as_deref().unwrap_or("").trim();
+    let category_filter = params
+        .category
+        .as_ref()
+        .map(|c| c.trim())
+        .filter(|c| !c.is_empty());

    // load categories and filtered/all contents
    let categories = User::get_user_categories(&user.id, &state.db).await?;
-    let text_contents = if !category_filter.is_empty() {
-        User::get_text_contents_by_category(&user.id, category_filter, &state.db).await?
-    } else {
-        User::get_text_contents(&user.id, &state.db).await?
+    let full_contents = match category_filter {
+        Some(category) => {
+            User::get_text_contents_by_category(&user.id, category, &state.db).await?
+        }
+        None => User::get_text_contents(&user.id, &state.db).await?,
    };

+    let (page_contents, pagination) = paginate_items(full_contents, params.page, CONTENTS_PER_PAGE);
+    let text_contents = truncate_text_contents(page_contents);
+
+    let page_query = category_filter
+        .map(|category| {
+            let mut serializer = form_urlencoded::Serializer::new(String::new());
+            serializer.append_pair("category", category);
+            format!("&{}", serializer.finish())
+        })
+        .unwrap_or_default();
+
    let conversation_archive = User::get_user_conversations(&user.id, &state.db).await?;
    let data = ContentPageData {
        user,
@@ -58,19 +88,19 @@ pub async fn show_content_page(
        categories,
        selected_category: params.category.clone(),
        conversation_archive,
+        pagination,
+        page_query,
    };

-    if is_htmx && !is_boosted && has_category_param {
-        // If HTMX partial request with filter applied, return partial content list update
-        return Ok(TemplateResponse::new_partial(
+    if is_htmx && !is_boosted {
+        Ok(TemplateResponse::new_partial(
            "content/base.html",
            "main",
            data,
-        ));
+        ))
+    } else {
+        Ok(TemplateResponse::new_template("content/base.html", data))
    }
-
-    // Otherwise full page response including layout
-    Ok(TemplateResponse::new_template("content/base.html", data))
 }

 pub async fn show_text_content_edit_form(
@@ -102,13 +132,32 @@ pub async fn patch_text_content(
    State(state): State<HtmlState>,
    RequireUser(user): RequireUser,
    Path(id): Path<String>,
+    HxTarget(target): HxTarget,
    Form(form): Form<PatchTextContentParams>,
 ) -> Result<impl IntoResponse, HtmlError> {
    User::get_and_validate_text_content(&id, &user.id, &state.db).await?;

    TextContent::patch(&id, &form.context, &form.category, &form.text, &state.db).await?;

-    let text_contents = User::get_text_contents(&user.id, &state.db).await?;
+    if target.as_deref() == Some("latest_content_section") {
+        let text_contents =
+            truncate_text_contents(User::get_latest_text_contents(&user.id, &state.db).await?);
+
+        return Ok(TemplateResponse::new_template(
+            "dashboard/recent_content.html",
+            RecentTextContentData {
+                user,
+                text_contents,
+            },
+        ));
+    }
+
+    let (page_contents, pagination) = paginate_items(
+        User::get_text_contents(&user.id, &state.db).await?,
+        Some(1),
+        CONTENTS_PER_PAGE,
+    );
+    let text_contents = truncate_text_contents(page_contents);
    let categories = User::get_user_categories(&user.id, &state.db).await?;
    let conversation_archive = User::get_user_conversations(&user.id, &state.db).await?;

@@ -121,6 +170,8 @@ pub async fn patch_text_content(
            categories,
            selected_category: None,
            conversation_archive,
+            pagination,
+            page_query: String::new(),
        },
    ))
 }
@@ -134,15 +185,29 @@ pub async fn delete_text_content(
    let text_content = User::get_and_validate_text_content(&id, &user.id, &state.db).await?;

    // If it has file info, delete that too
-    if let Some(file_info) = &text_content.file_info {
-        FileInfo::delete_by_id(&file_info.id, &state.db).await?;
+    if let Some(file_info) = text_content.file_info.as_ref() {
+        let file_in_use =
+            TextContent::has_other_with_file(&file_info.id, &text_content.id, &state.db).await?;
+
+        if !file_in_use {
+            FileInfo::delete_by_id(&file_info.id, &state.db, &state.config).await?;
+        }
    }

+    // Delete related knowledge entities and text chunks
+    KnowledgeEntity::delete_by_source_id(&id, &state.db).await?;
+    TextChunk::delete_by_source_id(&id, &state.db).await?;
+
    // Delete the text content
    state.db.delete_item::<TextContent>(&id).await?;

    // Get updated content, categories and return the refreshed list
-    let text_contents = User::get_text_contents(&user.id, &state.db).await?;
+    let (page_contents, pagination) = paginate_items(
+        User::get_text_contents(&user.id, &state.db).await?,
+        Some(1),
+        CONTENTS_PER_PAGE,
+    );
+    let text_contents = truncate_text_contents(page_contents);
    let categories = User::get_user_categories(&user.id, &state.db).await?;
    let conversation_archive = User::get_user_conversations(&user.id, &state.db).await?;

@@ -154,6 +219,8 @@ pub async fn delete_text_content(
            categories,
            selected_category: None,
            conversation_archive,
+            pagination,
+            page_query: String::new(),
        },
    ))
 }
@@ -181,16 +248,11 @@ pub async fn show_recent_content(
    State(state): State<HtmlState>,
    RequireUser(user): RequireUser,
 ) -> Result<impl IntoResponse, HtmlError> {
-    let text_contents = User::get_latest_text_contents(&user.id, &state.db).await?;
-
-    #[derive(Serialize)]
-    pub struct RecentTextContentData {
-        pub user: User,
-        pub text_contents: Vec<TextContent>,
-    }
+    let text_contents =
+        truncate_text_contents(User::get_latest_text_contents(&user.id, &state.db).await?);

    Ok(TemplateResponse::new_template(
-        "/dashboard/recent_content.html",
+        "dashboard/recent_content.html",
        RecentTextContentData {
            user,
            text_contents,
--- a/html-router/src/routes/index/handlers.rs
+++ b/html-router/src/routes/index/handlers.rs
@@ -4,17 +4,21 @@ use axum::{
    http::{header, HeaderMap, HeaderValue, StatusCode},
    response::IntoResponse,
 };
+use chrono::{DateTime, Utc};
+use futures::try_join;
 use serde::Serialize;
-use tokio::{fs::File, join};
-use tokio_util::io::ReaderStream;

 use crate::{
+    html_state::HtmlState,
    middlewares::{
        auth_middleware::RequireUser,
        response_middleware::{HtmlError, TemplateResponse},
    },
+    utils::text_content_preview::truncate_text_contents,
    AuthSessionType,
 };
+use common::storage::store;
+use common::storage::types::user::DashboardStats;
 use common::{
    error::AppError,
    storage::types::{
@@ -24,12 +28,11 @@ use common::{
    },
 };

-use crate::html_state::HtmlState;
-
 #[derive(Serialize)]
 pub struct IndexPageData {
    user: Option<User>,
    text_contents: Vec<TextContent>,
+    stats: DashboardStats,
    active_jobs: Vec<IngestionTask>,
    conversation_archive: Vec<Conversation>,
 }
@@ -42,26 +45,30 @@ pub async fn index_handler(
        return Ok(TemplateResponse::redirect("/signin"));
    };

-    let active_jobs = User::get_unfinished_ingestion_tasks(&user.id, &state.db).await?;
+    let (text_contents, conversation_archive, stats, active_jobs) = try_join!(
+        User::get_latest_text_contents(&user.id, &state.db),
+        User::get_user_conversations(&user.id, &state.db),
+        User::get_dashboard_stats(&user.id, &state.db),
+        User::get_unfinished_ingestion_tasks(&user.id, &state.db)
+    )?;

-    let text_contents = User::get_latest_text_contents(&user.id, &state.db).await?;
-
-    let conversation_archive = User::get_user_conversations(&user.id, &state.db).await?;
+    let text_contents = truncate_text_contents(text_contents);

    Ok(TemplateResponse::new_template(
        "dashboard/base.html",
        IndexPageData {
            user: Some(user),
            text_contents,
-            active_jobs,
+            stats,
            conversation_archive,
+            active_jobs,
        },
    ))
 }

 #[derive(Serialize)]
 pub struct LatestTextContentData {
-    latest_text_contents: Vec<TextContent>,
+    text_contents: Vec<TextContent>,
    user: User,
 }

@@ -73,30 +80,35 @@ pub async fn delete_text_content(
    // Get and validate TextContent
    let text_content = get_and_validate_text_content(&state, &id, &user).await?;

-    // Perform concurrent deletions
-    let (_res1, _res2, _res3, _res4, _res5) = join!(
-        async {
-            if let Some(file_info) = text_content.file_info {
-                FileInfo::delete_by_id(&file_info.id, &state.db).await
-            } else {
-                Ok(())
-            }
-        },
-        state.db.delete_item::<TextContent>(&text_content.id),
-        TextChunk::delete_by_source_id(&text_content.id, &state.db),
-        KnowledgeEntity::delete_by_source_id(&text_content.id, &state.db),
-        KnowledgeRelationship::delete_relationships_by_source_id(&text_content.id, &state.db)
-    );
+    // Remove stored assets before deleting the text content record
+    if let Some(file_info) = text_content.file_info.as_ref() {
+        let file_in_use =
+            TextContent::has_other_with_file(&file_info.id, &text_content.id, &state.db).await?;
+
+        if !file_in_use {
+            FileInfo::delete_by_id(&file_info.id, &state.db, &state.config).await?;
+        }
+    }
+
+    // Delete the text content and any related data
+    TextChunk::delete_by_source_id(&text_content.id, &state.db).await?;
+    KnowledgeEntity::delete_by_source_id(&text_content.id, &state.db).await?;
+    KnowledgeRelationship::delete_relationships_by_source_id(&text_content.id, &state.db).await?;
+    state
+        .db
+        .delete_item::<TextContent>(&text_content.id)
+        .await?;

    // Render updated content
-    let latest_text_contents = User::get_latest_text_contents(&user.id, &state.db).await?;
+    let text_contents =
+        truncate_text_contents(User::get_latest_text_contents(&user.id, &state.db).await?);

    Ok(TemplateResponse::new_partial(
-        "index/signed_in/recent_content.html",
+        "dashboard/recent_content.html",
        "latest_content_section",
        LatestTextContentData {
-            user: user.to_owned(),
-            latest_text_contents,
+            user: user.clone(),
+            text_contents,
        },
    ))
 }
@@ -128,6 +140,32 @@ pub struct ActiveJobsData {
    pub user: User,
 }

+#[derive(Serialize)]
+struct TaskArchiveEntry {
+    id: String,
+    state_label: String,
+    state_raw: String,
+    attempts: u32,
+    max_attempts: u32,
+    created_at: DateTime<Utc>,
+    updated_at: DateTime<Utc>,
+    scheduled_at: DateTime<Utc>,
+    locked_at: Option<DateTime<Utc>>,
+    last_error_at: Option<DateTime<Utc>>,
+    error_message: Option<String>,
+    worker_id: Option<String>,
+    priority: i32,
+    lease_duration_secs: i64,
+    content_kind: String,
+    content_summary: String,
+}
+
+#[derive(Serialize)]
+struct TaskArchiveData {
+    user: User,
+    tasks: Vec<TaskArchiveEntry>,
+}
+
 pub async fn delete_job(
    State(state): State<HtmlState>,
    RequireUser(user): RequireUser,
@@ -138,7 +176,7 @@ pub async fn delete_job(
    let active_jobs = User::get_unfinished_ingestion_tasks(&user.id, &state.db).await?;

    Ok(TemplateResponse::new_partial(
-        "index/signed_in/active_jobs.html",
+        "dashboard/active_jobs.html",
        "active_jobs_section",
        ActiveJobsData {
            user: user.clone(),
@@ -153,9 +191,8 @@ pub async fn show_active_jobs(
 ) -> Result<impl IntoResponse, HtmlError> {
    let active_jobs = User::get_unfinished_ingestion_tasks(&user.id, &state.db).await?;

-    Ok(TemplateResponse::new_partial(
+    Ok(TemplateResponse::new_template(
        "dashboard/active_jobs.html",
-        "active_jobs_section",
        ActiveJobsData {
            user: user.clone(),
            active_jobs,
@@ -163,6 +200,70 @@ pub async fn show_active_jobs(
    ))
 }

+pub async fn show_task_archive(
+    State(state): State<HtmlState>,
+    RequireUser(user): RequireUser,
+) -> Result<impl IntoResponse, HtmlError> {
+    let tasks = User::get_all_ingestion_tasks(&user.id, &state.db).await?;
+
+    let entries: Vec<TaskArchiveEntry> = tasks
+        .into_iter()
+        .map(|task| {
+            let (content_kind, content_summary) = summarize_task_content(&task);
+
+            TaskArchiveEntry {
+                id: task.id.clone(),
+                state_label: task.state.display_label().to_string(),
+                state_raw: task.state.as_str().to_string(),
+                attempts: task.attempts,
+                max_attempts: task.max_attempts,
+                created_at: task.created_at,
+                updated_at: task.updated_at,
+                scheduled_at: task.scheduled_at,
+                locked_at: task.locked_at,
+                last_error_at: task.last_error_at,
+                error_message: task.error_message.clone(),
+                worker_id: task.worker_id.clone(),
+                priority: task.priority,
+                lease_duration_secs: task.lease_duration_secs,
+                content_kind,
+                content_summary,
+            }
+        })
+        .collect();
+
+    Ok(TemplateResponse::new_template(
+        "dashboard/task_archive_modal.html",
+        TaskArchiveData {
+            user,
+            tasks: entries,
+        },
+    ))
+}
+
+fn summarize_task_content(task: &IngestionTask) -> (String, String) {
+    match &task.content {
+        common::storage::types::ingestion_payload::IngestionPayload::Text { text, .. } => {
+            ("Text".to_string(), truncate_summary(text, 80))
+        }
+        common::storage::types::ingestion_payload::IngestionPayload::Url { url, .. } => {
+            ("URL".to_string(), url.to_string())
+        }
+        common::storage::types::ingestion_payload::IngestionPayload::File { file_info, .. } => {
+            ("File".to_string(), file_info.file_name.clone())
+        }
+    }
+}
+
+fn truncate_summary(input: &str, max_chars: usize) -> String {
+    if input.chars().count() <= max_chars {
+        input.to_string()
+    } else {
+        let truncated: String = input.chars().take(max_chars).collect();
+        format!("{truncated}…")
+    }
+}
+
 pub async fn serve_file(
    State(state): State<HtmlState>,
    RequireUser(user): RequireUser,
@@ -177,14 +278,10 @@ pub async fn serve_file(
        return Ok(TemplateResponse::unauthorized().into_response());
    }

-    let path = std::path::Path::new(&file_info.path);
-
-    let file = match File::open(path).await {
-        Ok(f) => f,
-        Err(_e) => return Ok(TemplateResponse::server_error().into_response()),
+    let stream = match store::get_stream_at(&file_info.path, &state.config).await {
+        Ok(s) => s,
+        Err(_) => return Ok(TemplateResponse::server_error().into_response()),
    };
-
-    let stream = ReaderStream::new(file);
    let body = Body::from_stream(stream);

    let mut headers = HeaderMap::new();
--- a/html-router/src/routes/index/mod.rs
+++ b/html-router/src/routes/index/mod.rs
@@ -5,7 +5,9 @@ use axum::{
    routing::{delete, get},
    Router,
 };
-use handlers::{delete_job, delete_text_content, index_handler, serve_file, show_active_jobs};
+use handlers::{
+    delete_job, delete_text_content, index_handler, serve_file, show_active_jobs, show_task_archive,
+};

 use crate::html_state::HtmlState;

@@ -24,6 +26,7 @@ where
 {
    Router::new()
        .route("/jobs/{job_id}", delete(delete_job))
+        .route("/jobs/archive", get(show_task_archive))
        .route("/active-jobs", get(show_active_jobs))
        .route("/text-content/{id}", delete(delete_text_content))
        .route("/file/{id}", get(serve_file))
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Per Stark	798b1468b6	release: 0.2.5	2025-10-24 22:13:08 +02:00
Per Stark	3b805778b4	feat: scratchpad additional improvements changelog fix: wording	2025-10-22 22:29:41 +02:00
Per Stark	07b3e1a0e8	refactor: implemented state machine for ingestion pipeline, improved performance changelog additional moving around moved files around a bit	2025-10-20 13:08:00 +02:00
Per Stark	83d39afad4	refactor: implemented state machines for retrieval pipeline, improved tracing	2025-10-19 09:18:17 +02:00
Per Stark	21e4ab1f42	chore: clippy composite retrieval	2025-10-16 20:37:51 +02:00
Per Stark	3c97d8ead5	chore: clippy ingestion-pipeline	2025-10-16 20:36:39 +02:00
Per Stark	ab68bccb80	chore: clippy api-router	2025-10-16 20:33:57 +02:00
Per Stark	99b88c3063	chore: clippy html-router	2025-10-16 20:30:17 +02:00
Per Stark	44e5d8a2fc	chore: clippy fixes	2025-10-16 20:29:15 +02:00
Per Stark	7332347f1a	feat: quick search knowledge entities	2025-10-16 20:08:01 +02:00
Per Stark	199186e5a3	fix: variable name	2025-10-16 11:24:07 +02:00
Per Stark	64728468cd	design: knowledge new entity button to the left	2025-10-16 10:24:33 +02:00
Per Stark	c3a7e8dc59	chore: clippy performance improvements	2025-10-15 22:24:59 +02:00
Per Stark	35ff4e1464	feat: manual entity creation chore: clippy	2025-10-15 21:50:52 +02:00
Per Stark	2964f1a5a5	release: 0.2.4	2025-10-15 09:09:35 +02:00
Per Stark	cb7f625b81	fix: score normalization for vector search	2025-10-14 21:13:58 +02:00
Per Stark	dc40cf7663	feat: hybrid search	2025-10-14 20:38:43 +02:00
Per Stark	aa0b1462a1	feat: task archive fix: simplified	2025-10-14 10:38:09 +02:00
Per Stark	41fc7bb99c	feat: state machine for tasks, multiple workers	2025-10-12 22:21:20 +02:00
Per Stark	61d8d7abe7	release: 0.2.3	2025-10-12 20:15:10 +02:00
Per Stark	b7344644dc	fix: embedding dimension change on fresh db	2025-10-12 20:13:23 +02:00
Per Stark	3742598a6d	chore: changed image in readme	2025-10-08 12:04:47 +02:00
Per Stark	c6a6080e1c	release: 0.2.2	2025-10-07 11:51:33 +02:00
Per Stark	1159712724	fix: convert to surrealdb datetime before conversion	2025-10-03 15:33:28 +02:00
Per Stark	e5e1414f54	chore: clippy magic	2025-10-01 15:39:45 +02:00
Per Stark	fcc49b1954	design: new icons to match new theme	2025-10-01 10:17:43 +02:00
Per Stark	022f4d8575	fix: compliant with gpt-5 models	2025-10-01 10:17:31 +02:00
Per Stark	945a2b7f37	fix: do not log config here	2025-09-30 15:22:14 +02:00
Per Stark	ff4ea55cd5	fix: user guard on knowledge relationship deletion	2025-09-30 11:15:53 +02:00
Per Stark	c4c76efe92	test: startup smoke test	2025-09-29 21:15:34 +02:00
Per Stark	c0fcad5952	fix: deletion of items, shared files etc	2025-09-29 20:28:06 +02:00
Per Stark	b0ed69330d	fix: improved concurrency	2025-09-28 22:08:08 +02:00
Per Stark	5cb15dab45	feat: pdf support	2025-09-28 20:53:51 +02:00
Per Stark	7403195df5	release: 0.2.1 chore: remove stale todo chore: version bump	2025-09-24 10:25:56 +02:00
Per Stark	9faef31387	fix: json response in api to work with ios shortcuts fix corrected lockfile	2025-09-23 22:01:58 +02:00
Per Stark	110f7b8a8f	no graph screenshot in readme, too much image	2025-09-23 09:08:17 +02:00
Per Stark	f343005af8	release: 0.2.0 Includes: - Design overhaul - New graph visualization - Statistics on dashboard - Supports audio files for ingestion - Pagination for entities and content - Improved performance - Improved support for running as a systemd-service version bump	2025-09-23 09:06:59 +02:00
Per Stark	e1d98b0c35	fix: filter improvement	2025-09-23 08:59:57 +02:00
Per Stark	c12d00edaa	feat: pagination for entities and knowledge pages	2025-09-22 20:54:30 +02:00
Per Stark	903585bfef	fix: improved performance by truncating not displayed text	2025-09-22 15:37:55 +02:00
Per Stark	f592eb7200	fix: improved db schema strictness fix	2025-09-22 15:37:22 +02:00
Per Stark	c2839f8db3	fix: somewhat improved concurrency limiting edge creation to sequential due to surrealdb	2025-09-21 11:57:39 +02:00
Per Stark	9a7c57cb19	fix: contrast on code blocks	2025-09-21 10:43:57 +02:00
Per Stark	fe5143cd7f	design: additional design improvements chore: remove stale comment	2025-09-20 22:09:37 +02:00
Per Stark	3f774302c7	fix: dark mode prose text color	2025-09-17 14:48:38 +02:00
Per Stark	6ea51095e8	design: neobrutalist_theme into main	2025-09-17 10:01:20 +02:00
Per Stark	62d909bb7e	refactor: merge new storage backend into main This is in preparation for s3 storage support	2025-09-14 12:22:03 +02:00
Per Stark	69954cf78e	chore: clippy helps out	2025-09-06 21:00:39 +02:00
Per Stark	153efd1a98	feat: d3js instead of plotly, improved graph visualisation	2025-09-06 20:44:41 +02:00
Per Stark	fdf29bb735	fix: only count get requests as page loads	2025-09-06 11:16:12 +02:00
Per Stark	e150b476c3	feat: observability endpoints	2025-09-06 11:14:24 +02:00
Per Stark	2e076c8236	docs: clarify test suite, ask for contributions	2025-08-23 22:31:19 +02:00
Per Stark	a0632c9768	fix: write to stderr for journalctl compatability	2025-08-22 13:41:30 +02:00
Per Stark	33300d3193	feat: handles submitted audio	2025-07-29 18:39:26 +02:00
Per Stark	b8272d519d	release: 0.1.4 now supports sending of images as content remember version bump fix: typo in readme updated windows build machine in gh actions	2025-07-01 15:18:44 +02:00
Per Stark	ec16f2100c	fix: remove remnant job table, normalize taskstatus enum	2025-06-27 23:18:16 +02:00
Per Stark	43263fa77e	feat: completed image ingestion with system settings styling	2025-06-27 21:48:29 +02:00
Per Stark	f1548d18db	chore: fixed schema typo, error naming, updated compose example	2025-06-26 16:12:43 +02:00
Per Stark	f2bafe0205	fix: concurrent calls in admin page to api and db	2025-06-19 15:43:31 +02:00
Per Stark	9a23c1ea1b	feat: image ingestion	2025-06-17 08:26:15 +02:00
Per Stark	f567b7198b	fix: max-vw to prevent overflow, and some cleaning of wip comments	2025-06-14 22:58:43 +02:00
Per Stark	32141fce6e	release: 0.1.3 Support for other providers than OpenAI Changing of HTTP port Bugfixes	2025-06-08 20:35:17 +02:00
Per Stark	477f26174c	fix: overflow on long content list tiles	2025-06-08 20:34:34 +02:00
Per Stark	37584ed9fd	Merge branch 'custom_llm_base' fix: updated readme and corrected server and worker to updates added migration fix: openai url typo & displaying models chore: tidying up	2025-06-08 08:28:14 +02:00
Per Stark	a363c6cc05	feat: support for other providers of ai models	2025-06-06 23:16:41 +02:00
Per Stark	811aaec554	fix: graphmapper gracefully failing	2025-06-06 23:15:09 +02:00
Per Stark	7aa7e71287	chore: updated readme	2025-06-05 23:38:13 +02:00
Per Stark	d2772bd09c	feat: port selection	2025-05-30 07:44:26 +02:00
Per Stark	81825e2525	chore: added screenshot to readme fix fix	2025-05-26 12:31:04 +02:00
				`@@ -0,0 +1 @@`
				{"schemas":"--- original\n+++ modified\n@@ -98,7 +98,7 @@\n DEFINE INDEX IF NOT EXISTS knowledge_entity_user_id_idx ON knowledge_entity FIELDS user_id;\n DEFINE INDEX IF NOT EXISTS knowledge_entity_source_id_idx ON knowledge_entity FIELDS source_id;\n DEFINE INDEX IF NOT EXISTS knowledge_entity_entity_type_idx ON knowledge_entity FIELDS entity_type;\n-DEFINE INDEX IF NOT EXISTS knowledge_entity_created_at_idx ON knowledge_entity FIELDS created_at; # For get_latest_knowledge_entities\n+DEFINE INDEX IF NOT EXISTS knowledge_entity_created_at_idx ON knowledge_entity FIELDS created_at;\n\n # Defines the schema for the 'message' table.\n\n@@ -157,6 +157,8 @@\n DEFINE FIELD IF NOT EXISTS require_email_verification ON system_settings TYPE bool;\n DEFINE FIELD IF NOT EXISTS query_model ON system_settings TYPE string;\n DEFINE FIELD IF NOT EXISTS processing_model ON system_settings TYPE string;\n+DEFINE FIELD IF NOT EXISTS embedding_model ON system_settings TYPE string;\n+DEFINE FIELD IF NOT EXISTS embedding_dimensions ON system_settings TYPE int;\n DEFINE FIELD IF NOT EXISTS query_system_prompt ON system_settings TYPE string;\n DEFINE FIELD IF NOT EXISTS ingestion_system_prompt ON system_settings TYPE string;\n\n","events":null}
				`@@ -0,0 +1 @@`
				`{"schemas":"--- original\n+++ modified\n@@ -160,6 +160,7 @@\n DEFINE FIELD IF NOT EXISTS query_system_prompt ON system_settings TYPE string;\n DEFINE FIELD IF NOT EXISTS ingestion_system_prompt ON system_settings TYPE string;\n DEFINE FIELD IF NOT EXISTS image_processing_prompt ON system_settings TYPE string;\n+DEFINE FIELD IF NOT EXISTS voice_processing_model ON system_settings TYPE string;\n\n # Defines the schema for the 'text_chunk' table.\n\n","events":null}`