Set up the assistant
Connect a provider and choose a model for new chats.
Manual work does not require an assistant provider. Connect one when you want help reading context, preparing records, or investigating results.
Add a connection
Open Account menu → Settings → Assistant. Choose a provider preset or a custom endpoint. Enter the connection and authentication details. A local server can use Authentication → None when that server expects no credential.
Add and test models
Fetch the provider’s model list or add exact model IDs manually. Test response checks text output; Test agent tools checks a complete tool call and result exchange. Set model-specific image, reasoning, context, and output limits when needed.
Choose a default
Select the default for new chats. Each chat can choose a saved model and thinking setting. Existing chats retain their choices when you change the default; changing a model waits for active work to settle.
Start with a concrete request
Describe the question, intended use, recipient, and evidence available. Add saved context, then review the proposed work before execution or a substantive decision.
Assistant provider settings are separate from the API profiles used to obtain model predictions for tests.