About Remem

Remem is a persistent memory engine for AI agents. It ships as two Rust services —remem-server (the REST API and embedded storage engine) and remem-mcp(the MCP protocol adapter) — with no external database dependency: no PostgreSQL, no Neo4j, no Qdrant. Just docker compose up -d and both are running.

We built Remem because every agent memory solution we tried required stitching together multiple services, paid for graph access separately, and had no concept of memory lifecycle. Memories accumulated, search degraded, and agents kept starting from zero.

What makes Remem different

• MCP-native from the ground up — eight tools designed for how LLMs consume structured data, not retrofitted REST wrappers.
• Built-in relationship graph — auto-discovery writes typed edges between semantically related memories. No extra service, no extra cost.
• Memory lifecycle — TTL expiration, importance decay, promotion to long-term, soft deletion. The longer you use it, the better search stays.
• Embedded storage — HNSW vector index, CSR connection graph, BTree for time-windowed queries, an inverted tag index, and a WAL-backed LSM-tree key-value store all run in-process inside remem-server. No external database.

Open source

Remem is source-available under theFunctional Source License. You can use it freely — including in commercial products — with one restriction: you cannot offer Remem as a competing managed service. Every release converts to Apache 2.0 two years after its release date.

The source is on GitHub. Bug reports, questions, and contributions are welcome.