Chat
Route: /chat
A chat client for OpenAI v1-compatible model providers. Point it at a hosted provider or a model running on your own machine.
Supported providers
The Chat tool speaks the OpenAI v1/chat/completions API shape. Local providers
are proxied through Acolyte so the browser isn't blocked by CORS:
| Provider | Default endpoint |
|---|---|
| llama.cpp | http://localhost:8080/v1/chat/completions |
| Ollama | http://localhost:11434/v1/chat/completions |
| Docker Model Runner | http://localhost:12434/engines/v1/chat/completions |
Any other OpenAI-compatible endpoint can be used as well.
Features
- Multi-turn conversations with streaming responses.
- Saved conversations persisted to IndexedDB.
- Configurable provider and model.
- Bring your own key — credentials are sent only as the request's
Authorizationheader.
How it works
Requests go to
POST /api/genai/chat/completions,
which forwards them to the selected provider. The proxy forwards only the
Authorization and Content-Type headers — no credentials are stored on the
server.
Using it
- Run a local model server (e.g.
ollama serve) or have a hosted endpoint ready. - Open Chat from the sidebar.
- Select the provider and model; supply an API key if the provider needs one.
- Start chatting. Conversations are saved locally so you can revisit them.
Notes
- Local providers must be running and listening on their default ports (above) for the proxy to reach them.
- Saved conversations live in your browser only.