API-only mode
For people who want their models behind one key and nothing else: the server becomes purely an API front, with the product switched off.
Switching it on
Admin → General → Instance mode: Full (the default) or API only.
What changes
| Keeps working | Switched off |
|---|---|
The /v1 API (chat, models, embeddings, audio), the /api/v1 endpoints, discovery, device sign-in and the device link | Chats and agent chats — a linked device’s too — and the terminal panel |
| Signing in; Settings: account, tokens and devices, usage, preferences | The library and /mcp |
| All of Admin | Reports, scheduled tasks, folders |
| Branding, the installable app, push notifications | Uploads and audio in the composer |
- Pages that are switched off send you to Settings with a notice; their APIs answer 404 with a sentence saying why.
- Schedules wait rather than fire, and nothing starts a reply in the background.
- Administrators are not exempt: the mode is what the server is, not what one person may do.
- Nothing is deleted. Switching back to Full restores everything as it was.
Typical setup
- Add your connections and enable the models.
- Create a group for API users and give it Use the API and access to the models.
- Optionally turn on credits and give the group a plan.
- Each person makes a token under Settings → Security and points any OpenAI client at
https://your-server/v1. See API.