YantrikDB
An embeddable Rust memory engine for AI agents that manages what is remembered instead of only storing it, applying temporal decay, contradiction detection, consolidation and a knowledge graph.
YantrikDB targets a common failure of bolting a vector store onto an agent: nothing is ever forgotten, near-duplicate memories accumulate, and contradictory facts come back ranked side by side with no signal that they disagree. The engine instead manages memory actively, with temporal decay, autonomous consolidation, contradiction detection and a knowledge graph, exposed as a Rust library with Python bindings and approximate nearest-neighbor recall through HNSW.
An MCP integration lets an agent in Claude, Cursor or Windsurf auto-recall context, auto-remember decisions and auto-detect contradictions with no extra prompting, and a companion server wraps the same engine with an HTTP API, a live demo and multi-node high availability. A file-backed store loads a small self-hosted embedding model on first use, falling back to a bundled model if the download fails, and the library bundles its own SQLite, which must not be opened a second time from the same process.