Willow

Memory system for AI agents built on a structured, evolving knowledge graph with git-like versioning. The graph store is Rust over SurrealDB, compiled to a native Node binary via napi-rs, with six hybrid search strategies (lexical, vector and graph walks), episodic memory, temporal metadata, community detection and time-decay scoring. A companion MCP server exposes eight graph tools. The hosted beta deployment has since been taken down; the demo build is public and the code still runs locally.

Why it matters

You can’t bolt a "remember this" tool onto a chat LLM and expect good memory. Willow runs separate background agents for indexing, graph evolution, maintenance and retrieval, because the chat-facing model is too lazy about upkeep.