Models & connectors

Two different connections

Model configuration selects the AI service used by the workbench. Connectors authorize access to external data and operations. Configuring one does not configure the other, and connector OAuth is not a login method.

Configure a model

Open model management (/agent/llm) from the user menu. Supply the service endpoint, model identifier and credentials required by your provider and the current instance UI.

Try a small workbench request before starting a long task. On failure, check the returned error, endpoint, credential validity, model access and provider quota.

Configure connector providers

Administrators manage providers under System settings → Connectors. The page shows the instance callback URL and provider configuration status.

1. Create an application at the external service.
2. Register the exact callback URL shown by your instance.
3. Save the provider configuration and enable it.
4. Each user then authorizes their own connection on the connectors page.

A listed provider does not imply support for every one of its APIs. Use the capabilities actually exposed by your instance.

Troubleshoot authorization

Compare callback protocol, host and path when redirects fail. Reconnect expired authorizations. Diagnose model and connector errors separately.

Use dedicated credentials and grant only the access required for the workflow. Keep secrets out of public workspace instructions and screenshots.