Self-host with Ollama
The app can send chat to an Ollama server you run. Account data stays in the local database. Inference stays on that server.
docker compose up -ddocker compose exec ollama ollama pull llama3.2- Set
OLLAMA_BASE_URL=http://127.0.0.1:11434andMOCK_AI=false. - Restart and pick Llama 3.2 (Ollama). Keep the Built with Llama notice.
The compose file starts Ollama only. The app process is still the Worker runtime, because that is what serves the site and the database.