LLM Wiki Memory
Agentic memory for AI apps — knowledge graph retrieval, facts, events, and semantic recall
Platform-agnostic TypeScript engine for hybrid LLM memory. Knowledge-graph retrieval (graph RAG) over an ontology of entities, episodic fact extraction, hybrid cosine + keyword search, and OKF v0.1 portability. Bring your own SQLite adapter.
npm Packages
Platform-agnostic
SQLiteAdapter interface, zero runtime deps — bring your own adapter.
Hybrid retrieval
Embeddings + cosine similarity fused with MiniSearch keyword scoring, using a two-phase pre-filtered SELECT so recall stays fast at scale.
Immutable vs mutable facts
Document-sourced vs inferred vs user-stated memory types.
Maintenance jobs
Librarian, heal, reembed, and prune background jobs.
Per-entity seeded ontology
Strict, Emergent, or Off modes for structured knowledge.
OKF v0.1 import/export
Open Knowledge Format interoperability under the llm-wiki/1 profile — memory is portable, not trapped.
Production-proven in Clanker AI
Powers persistent memory in the Clanker AI personal assistant across iOS, Android, and web.
Tiered AI memory architecture
Inspired by Andrej Karpathy's LLM Wiki memory spec.
Powers persistent memory in Clanker AI— our personal AI assistant with real-time voice.