Typed memories (episodic / semantic / procedural / working) linked by a real relation graph — causes, contradicts, supersedes, same_as and more.
OpenAI-embedding vector search over Qdrant with a transparent text fallback, optional recency weighting, and namespace isolation.
Export/import the full graph as PMF (JSON, binary MessagePack, or AES-256-GCM encrypted) with Merkle + content-hash integrity. No lock-in.
REST API, an MCP server (12 tools), TypeScript & Python SDKs, and a CLI — wire memory into Claude, your IDE, or any agent runtime.
One docker compose up brings up the API, web UI, SurrealDB and Qdrant. Open source under Apache-2.0. Upgrade to Pro for federation and higher limits when you need them. See the benchmark →