Response Modes

Mode	Stored value	Intended routing
Triggered	triggered	Mentions and replies addressed to Fork
All messages	all	Evaluate every eligible message; increased usage is possible
Smart	smart	Use conversation context to decide whether to answer; fall back to Triggered when a decision is unavailable

Configure and verify

Open AI Channels for the selected server, choose a channel and save the response settings. Test with one clearly addressed request. Check the persisted settings, not only the state of an unsaved control.

Channel cooldown accepts an integer from 0 to 3,600 seconds. Zero disables that channel cooldown, not provider limits, command limits or authorization.

Tool output modes are different

The ai and direct options in tools set control how tool results are delivered, not whether every channel message receives a response. AI mode gives the result back to the model; direct mode posts command output. Neither grants missing permissions.

Use AI problems when a saved mode still appears quiet. Avoid enabling All messages across an entire server merely to diagnose one failed response.