Memory for your agent

Mind Layer stores facts your agent can retrieve in later conversations. Run it locally, keep the database, and inspect the context used for a reply. No Sonzai account is required.

Use it with your agent or try Moss, a character that remembers your interactions.

Meet Moss

Run make demo and open http://127.0.0.1:8090. Choose an interaction, talk to Moss, and look at the memories behind a reply. Story choices change its traits through explicit rules. Offline mode recalls stored facts; open-ended dialogue uses your configured model. Read the demo guide.

Run Mind Layer

With Go 1.26 installed, start the service from the repository root:

go run ./cmd/mind-layer

Then run the example in another terminal:

python3 examples/quickstart.py

No provider key is needed. Facts persist in data/memory.db across restarts. The first build downloads dependencies; keyword memory operations run locally without network calls.

Choose how to retrieve memories

Keyword search can find a stored fact about “vegetarian dinner” when your agent asks for those words. For differently worded questions, configure an embedding model and use semantic or hybrid search. You can also configure a chat model to extract facts and write replies. Configure a provider.

Mind Layer runs as one Go process with an embedded database. It is Apache-2.0, with no Sonzai subscription, license server, metering or telemetry dependency. Optional model usage is billed by your provider; you can also use a compatible local model server. You cover your own hardware and hosting.

This standalone edition reuses Sonzai’s BM25 and multilingual tokenization code; its persistence, API and provider integration are new. It includes a subset of the hosted platform. Read the tested capabilities, API reference, benchmarks, and data lifecycle for the exact scope.