Why some people choose to self-host AI
Handing private conversations to the cloud means you can't personally verify how they're stored and used: even if the terms say "won't be used for training," all you can do is trust. And once your account is locked, the platform shuts down, or compliance requirements change, those records are no longer in your hands. Self-hosting returns data sovereignty to the user — especially important for those sensitive about privacy or who want long-term control over their own data.
How easy self-hosting Enclave is
If you've used docker compose, the difficulty is no greater than running an ordinary web service. The "three-minute deployment" at the top of the README is real: clone the repo, copy .env, docker compose up — and the API, frontend, database, and vector index all run locally. The entire stack is MIT open-source, with no binary black boxes whatsoever.
Models can be cloud or local — even fully offline
Enclave's model layer is fully swappable: you can mix cloud APIs like OpenAI and Anthropic with local Ollama / vLLM, assigning them by conversation type. Switch all models to local and turn off real-world sync, and the entire system stops sending any outbound requests — the strongest privacy promise, because it depends on no promise at all, only on your own machine.
Ready to try it?
Open it in your browser — no credit card, no install.