nambok / mentedb

What it does

• Automatic Memory Extraction LLM powered pipeline extracts structured memories from raw conversations
• Entity-Centric Memory Extracts typed entities (person, pet, place, event, item, organization) with structured attributes. Entity resolution merges attributes across mentions. Derived and labeled Related edges link entities to the memories t
• Hybrid Retrieval Vector similarity (HNSW) fused with BM25 keyword search via reciprocal rank fusion, plus tag, temporal, and validity filtering. Optional second-pass reranking (pluggable, e.g. a cross-encoder or LLM judge) and optional MMR
• Contextual Retrieval An optional per-memory context blurb is indexed and embedded alongside the content (never stored in it), so a memory is findable by situating terms it never literally contains. The caller generates the context; the engi
• Project Scope Weighting Recall can weight down memories tagged for other projects so the project you are working in ranks first, without hiding a strongly relevant cross-project memory