Persistent agent memory.
Memories survive across sessions and conversations, stored in PostgreSQL with pgvector.
Open RLM Memory
A local-first memory server giving AI agents a persistent, searchable long-term brain over the Model Context Protocol, in your own PostgreSQL.
What it is
Open RLM Memory is a local-first memory server for AI agents. Out of the box, an AI agent forgets everything the moment a conversation ends. Open RLM Memory gives it a persistent, searchable long-term brain: memories are stored, embedded, and retrievable across sessions, exposed to any agent that speaks the Model Context Protocol.
Local-first is the defining choice. The whole system runs on your own machine or server: memories live in your PostgreSQL database, and language-model and embedding calls go to any OpenAI-compatible endpoint, with a locally running LM Studio as the default. Nothing about an agent's accumulated knowledge has to leave your infrastructure, and there is no hosted service in the loop.
It is deliberately simple to operate. One FastAPI application serves both the backend and the frontend on a single port, and data isolation is handled with lightweight namespaces rather than a heavyweight identity stack.
Who it is for
Agent builders and developers who want durable context for their AI systems without shipping that context to someone else's cloud.
What it does
Semantic search.
Vector search over stored memories, so agents recall by meaning rather than exact wording.
Search caching.
An app-owned PostgreSQL cache table speeds up repeated queries.
Namespace isolation.
Data separation via a simple namespace header, with no OAuth, JWT, or external identity provider required.
Bring your own model.
Any OpenAI-compatible endpoint works for LLM and embedding calls; LM Studio is the local default.
MCP integration.
Agents connect over the Model Context Protocol, with an MCP Inspector guide for testing the surface.
Under the hood
A single FastAPI app serving backend and frontend on port 8000, PostgreSQL with pgvector for storage, and Docker support for containerized runs. The docs cover installation, configuration, local development, monitoring, production deployment, the security model, and full API and data-model references. An earlier-stage project than some of its siblings, and honest about it.
Interested?
If your agents keep forgetting what they learned yesterday, we can give them a memory that stays on your hardware.