c0: Bi-temporal Knowledge Graph Memory for LLMs
What it is
c0 is a Rust-based external memory system for LLMs that stores knowledge as a bi-temporal knowledge graph in Neo4j rather than flat vector chunks. It uses a hybrid retrieval cascade (exact match, BM25, and vector search fused via RRF) and includes a self-improving reflection loop that automatically commits new concepts when lookups fail.
The problem it solves
Standard vector RAG retrieves text blobs without understanding how knowledge changes over time, fails to distinguish corrected facts from stale ones, and cannot perform point-in-time queries or multi-hop reasoning effectively.
What you could use it for
- Correcting stale LLM training data by retrieving current, time-versioned facts
- Performing point-in-time ('as-of') queries on evolving knowledge bases
- Building persistent, correctable long-term memory for autonomous agents
- Benchmarking and comparing structured graph memory against flat vector RAG
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.