Per Stark
5c2d2e24d3
Collapse the multi-strategy entity engine into one benchmarked chunk retrieval path, derive entities from retrieved chunks, and update consumers, docs, and clippy fixes across the workspace.
2.4 KiB
Features
Search vs Chat
Search — Use when you know what you're looking for. Full-text search matches query terms across your content.
Chat — Use when exploring concepts or reasoning about your knowledge. The AI analyzes your query and retrieves relevant context from your entire knowledge base.
Content Processing
Minne automatically processes saved content:
- Web scraping extracts readable text from URLs (via headless Chrome)
- Text analysis identifies key concepts and relationships
- Graph creation builds connections between related content
- Embedding generation enables semantic search
Knowledge Graph
Explore your knowledge as an interactive network:
- Manual curation — Create entities and relationships yourself
- AI automation — Let AI extract entities and discover relationships
- Hybrid approach — AI suggests connections for your approval
The D3-based graph visualization shows entities as nodes and relationships as edges.
Hybrid Retrieval
Minne uses chunk-first hybrid retrieval over the knowledge base:
- Vector similarity — Semantic matching via embeddings over text chunks
- Full-text search — Keyword matching with BM25 over the same chunk index
The two ranked candidate lists are merged with Reciprocal Rank Fusion (RRF). When a caller needs knowledge entities (search, ingestion linking, relationship suggestion), entities are derived from the top retrieved chunks grouped by source_id.
Optional reranking can rescore the fused chunk list with a cross-encoder model; see below.
Reranking (Optional)
When enabled, retrieval results are rescored with a cross-encoder model for improved relevance. Powered by fastembed-rs.
Trade-offs:
- Downloads ~1.1 GB of model data
- Adds latency per query
- Potentially improves answer quality, see blog post
Enable via RERANKING_ENABLED=true. See Configuration.
Multi-Format Ingestion
Supported content types:
- Plain text and notes
- URLs (web pages)
- PDF documents
- Audio files
- Images
Scratchpad
Quickly capture content without committing to permanent storage. Convert to full content when ready.
iOS Shortcut
Use the Minne iOS Shortcut for quick content capture from your phone.