Open Memory Protocol (OMP)

One API for any AI memory provider.

OMP is an open standard plus reference Python SDK that lets agentic applications talk to any memory backend — Postgres, Mem0, Supermemory, Letta, your own — through one stable set of verbs.

30-second quickstart

docker run --rm -d -p 5432:5432 -e POSTGRES_PASSWORD=postgres pgvector/pgvector:pg16
pip install -e ./sdk-python
$env:PG_URL = "postgresql://postgres:postgres@localhost:5432/postgres"
python examples/01_quickstart.py

from openmem import Memory

mem = Memory(provider="postgres", url="postgresql://localhost/omp")
mem.add(content="user prefers pnpm over npm", user_id="kek",
        scope="coding/preferences", tags=["tooling", "nodejs"])

ctx = mem.context("set up a new node project", user_id="kek", token_budget=500)
print(ctx.text)

Provider matrix

Provider	Mode	Status
Postgres + pgvector	Native (ref)	🟢 Stable
Native OMP endpoints	Passthrough	🟢 Stable
Mem0	Translation	🟢 Stable
Supermemory	Translation	🟢 Stable
Letta	Translation	🟢 Stable

Live capability advertisement is authoritative — every adapter exposes its real verb set and feature flags via mem.capabilities() and GET /capabilities on omp-server. See docs/providers.mdx for the per-feature matrix.

Documents

• SPEC — narrative spec
• OpenAPI — canonical schema (Constitution Principle I)
• Constitution — non-negotiable engineering principles
• SDK README
• CHANGELOG

Tooling

• openmem-eval — manual benchmark harness comparing recall, MRR, and latency across configured providers. Never runs in CI. Default invocation is a dry-run that makes zero network calls. See specs/004-eval-kit/quickstart.md for usage and docs/eval/ for a sample report and trace from a real postgres run.

• openmem.AsyncMemory — async/await-native facade mirroring openmem.Memory. Postgres + passthrough adapters use native async clients (asyncpg, httpx); mem0/supermemory/letta are wrapped with a per-instance thread pool. Cancellation propagates within 50 ms on the native tier. Install with pip install 'openmem[async]' and see specs/005-async-fastapi/quickstart.md §1–§4.

• omp-server — production-shaped FastAPI HTTP server that exposes the full OMP verb set over HTTP using AsyncMemory under the hood. Install with pip install 'openmem[server]' then boot, e.g.:

  omp-server --provider postgres --url postgresql://user:pass@host:5432/db --port 8080

  Trusted-network deployment only — auth is deferred to a future release. See specs/005-async-fastapi/contracts/http-server.md for the route table, error-envelope mapping, and CORS / size-limit / health-check contracts.

HTTP server

pip install 'openmem[server]'
omp-server --provider postgres --url postgresql://user:pass@host:5432/db
# omp-server: serving postgres at http://127.0.0.1:8080

Routes mirror spec/omp-0.1.openapi.yaml exactly: POST/GET/PATCH/DELETE /memories[/{id}], GET /memories/search, POST /context, GET /audit, GET /capabilities, GET /healthz. Every request emits one INFO access log line with no user_id and no secret-keyed values.

License

TBD.