Connect a model provider
Set up external inference independently from a Cloud workspace.
Your space → Models & voice → Providers
Before you begin
A compatible Mellow build with the relevant capability configured.
Prepare account access
Check whether the service requires a subscription, API key or browser sign-in. These methods are not interchangeable.
Add the provider
Choose a preset or custom configuration and review its endpoint. Enter credentials using the dedicated field, not a chat prompt.
Verify a conversation
After connection succeeds, select one of its models and send a small request. Confirm that provider handled the request before adding it to routines.
Controls & reference
| Control or concept | How to use it |
|---|---|
| Preset / Custom | Choose a supported service preset or the protocol and endpoint of your own server. |
| Authentication | Enter credentials through the app’s secure credential controls. |
| Enabled / Auto-connect | Whether the connection is available and reconnects automatically. |
| Connection state | Read errors and readiness on the provider card. |
| Model picker | Select a model discovered through the intended provider. |
Try it with a small task
Connect one provider and ask for a short greeting. Inspect the selected model name and the completed response.
If it does not work
- An enabled provider can still fail authentication.
- Prompts and relevant context are sent to the provider you choose.
- Missing ChatGPT sign-in tokens require the provider’s sign-in flow; workspace authorization does not create those tokens.
Good to know
- An enabled provider can still fail authentication.
- Prompts and relevant context are sent to the provider you choose.
- Missing ChatGPT sign-in tokens require the provider’s sign-in flow; workspace authorization does not create those tokens.