Expand description
recall-graph — Knowledge graph with semantic search for AI memory systems.
Provides a structured graph layer (Layer 0) underneath flat-file memory systems. Used by recall-echo (pulse-null entities) and recall-claude (Claude Code users).
Modules§
- confidence
- Bayesian confidence model for relationship edges.
- crud
- Entity and relationship CRUD operations.
- dedup
- LLM-powered entity deduplication — skip, create, or merge decisions.
- embed
- Text embedding via fastembed (BGE-Small-EN-v1.5, 384 dimensions).
- error
- Typed error handling for recall-graph.
- extract
- Conversation chunking and LLM-powered entity/relationship extraction.
- ingest
- Ingestion orchestrator — chunk → episode → extract → dedup → relationships.
- llm
- Minimal LLM provider trait for knowledge graph operations.
- pipeline
- Pipeline document parser — converts praxis pipeline markdown documents into graph entities.
- pipeline_
sync - Pipeline sync engine — reconcile flat-file pipeline state with the graph.
- query
- Hybrid query — combines semantic search, graph expansion, and episode search.
- search
- Semantic search across entities and episodes using HNSW KNN + hotness scoring.
- store
- SurrealDB embedded store — open, schema init.
- traverse
- Graph traversal — recursive depth-first with cycle detection.
- types
- vigil_
sync - Vigil-pulse sync engine — ingest metacognitive signals and outcomes into the graph.
Structs§
- Graph
Memory - The main entry point for graph memory operations.