Connect a provider and offer models
An administrator can connect an API-key provider and make its models available in the instance catalog. The connection alone does not grant every account permission to use those models.
Add an API-key provider
- Open Settings →
<assistant name>AI → AI providers and choose Add provider. - Select OpenAI-compatible or Anthropic API from the menu.
- Enter the API address; an OpenAI-compatible endpoint requires its version path, for example
https://ai.example.com/v1. Add an API key if the endpoint requires one. - For OpenAI-compatible endpoints, choose models returned by discovery or enter model IDs, one per line, if discovery returns none.
- Save the provider. Do not place keys in chat or documentation.
Offer and verify models
The assistant name in that Settings path is dynamic: it displays the brand configured for this instance. OAuth and linked key-account sign-in are separate connection flows. Check Settings → Models → Model catalog for offered models. Set Default chat model in Model roles only after the intended provider/model pair is available. This default applies when an account has no personal choice; account permissions still apply. In Users, grant the model to accounts that need it. A grant cannot make an unconfigured model run.
Use Test connection beside the provider name. It sends a real request and may consume the provider's paid budget. Success names the responding model; failure shows the provider's reason. A stored sign-in alone does not prove current connectivity.
Common failures
If discovery shows no models, check the API address and whether the endpoint supports /models; enter model IDs manually when necessary. Confirm the provider is active and its account or key is connected. For connection types and troubleshooting, see AI Providers & Models; for user access, see Add a user and grant the right access.