MimirMemory is a path
Every memory lives at a path like deploy/rollback, with a one-line title, a summary and the full text. Agents list, read and search it like a filesystem.
Long-term memory for coding agents
Your agents forget everything. Fix that in five minutes.
Mimir keeps what an agent learns in your own Postgres — paths, versions, a digest to start from, and search by meaning — and hands it back at the start of the next session.
- One container
- Your Postgres
- Your keys
|OC | Mimir memory platform setup
❯ what do we know about token rotation?
+ Thought · 0.4s
✓ memory_digest
> repo: mimir
12 memories · auth, deploys, conventions
✓ memory_search
> query: "token rotation"
auth/token-rotation · auth/session-expiry
Tokens rotate from Settings → API keys. The old key keeps working for 24 hours, so deploy first and revoke after.
read 2 memories · 1.1s
>
~/Development/mimir:main1.9K (2%) · $0.002 ctrl+p commands
What it holds
One endpoint, every agent
MCP over HTTP for OpenCode, Claude Code, Codex, Cursor and the rest, plus a plain REST mirror for anything that speaks curl.
Yours alone
Runs in your homelab on one container and one database. User accounts, per-project grants, two-factor sign-in, encrypted backups.
Give your agents a memory.
Run it on your own machine, point one agent at it, and the next session starts where the last one ended.
Start in five minutes
docker compose up -d