Skip to main content
Version: 0.8.4

Configure a model connection

Adding, editing and testing model connections requires administrator system-settings privileges. Advanced Users select existing models for Agents or their own Uno; Users operate prepared Agents. If the entry is missing, check your role rather than treating access restrictions as an API-key failure.

Add a provider​

In first-run setup, or System Settings → Model Configuration → Add Provider, choose the connection plan that matches your provider account. Open its API-key and setup links, obtain a key with access to the listed models, then select Test and connect. A chat subscription, standard API and coding plan can use different credentials and endpoints; they are not interchangeable.

The test runs from the connected Engine, not merely your browser. It can incur a small provider charge. If it fails, check the key, account balance, model permission, endpoint and the Engine's network access before retrying.

Subscription and custom connections​

Choose ChatGPT subscription for the supported sign-in flow instead of pasting a subscription credential into a standard API preset. Choose the custom configuration entry when no preset matches your service. In either case, verify the connection before assigning it to an Agent.

Advanced settings exposes the API format, base URL and available models. Each model uses the identifier sent in API calls. Configure context/output limits and supported capabilities to match the provider, and choose the default model from the available models. Do not copy capability values from an unrelated service.

Presets can list multiple models and contain localized setup information. Provider catalogs change; treat the current preset as a starting point, not a guarantee of access or perpetual model availability. Updating a catalog is not permission to silently replace an existing connection's settings.

The context window is shared by input and the requested maximum output. Leaving maximum output empty uses the displayed default; a custom limit also reserves space before the next request. Long conversations can be summarized automatically. For a clear context-overflow response, Clawuno can make up to two history-compaction recovery attempts in the same turn without replaying tools or changing your saved limits. If the current request or tool definitions still cannot fit, summarization fails, or you stop the run, it stops rather than retrying indefinitely. Keep the original error when asking for help; reduce the current input or review the model's actual limits instead of repeatedly resending the same request.

Check an existing connection​

Use the connection card's test action and wait for each configured model's result. Keep failed models out of critical automated work until resolved. An Agent can use a configured model; changing a provider does not rewrite previous conversations or their usage records.

Ask Uno:

Explain which model connection this Agent uses and which settings I need to check if its connection test fails.