Set up Ollama locally

Unimatrix can work alongside a local Ollama installation for fully local inference while still syncing memory to your account (or staying fully local if you self-host the whole stack).

Install Ollama and pull a model as usual, then point your Unimatrix client configuration at your local Ollama endpoint (default `http://localhost:11434`) in Settings → Integrations → Local Models.

Memories you save while using an Ollama-backed client are stored the same way as any other memory — encrypted client-side — and sync to the cloud the next time you are online, so you can pick up the same context from a cloud-connected client later.

If you are self-hosting the entire Unimatrix stack (see the Self-Hosting Guide), you can also run the Librarian background classification job against your own Ollama instance by setting the `OLLAMA_URL` and `LIBRARIAN_MODEL` environment variables.