Custom models
Add OpenAI-compatible endpoints for a workspace or the whole instance.
On /models, add a custom model:
- Display name
- Base URL (OpenAI-compatible)
- Provider model id
- Optional API key and context window
Presets (Ollama, LM Studio, …) and Fetch models discovery help fill fields.
Scope
| Scope | Who can add | Badge |
|---|---|---|
| This workspace | Workspace admin | This workspace |
| Entire instance | Instance admin | All workspaces |
Custom models merge with enabled Command Code entries in the composer catalog. Workspace owners/admins can turn individual custom models off, or use Turn all off / Turn all on, without deleting them. Chat works without COMMAND_CODE_API_KEY when a usable enabled custom model exists for the workspace.
Azure AI Foundry and other private hosts
Foundry (and similar private model platforms) is the model plane, not a second chat app. Register the OpenAI-compatible chat-completions URL, the provider model id, and an optional API key. Brain does not train or fine-tune inside the app — that stays in Foundry.
The same path works for a company proxy, Ollama, or LM Studio. See Architecture.