Intents

Train what users can ask and build the response sub-flow.

An intent represents something the user wants to do — check balance, pay a bill, reset password. Each intent has training phrases, optional parameters, and its own visual sub-flow.

Training phrases

Add one phrase per line in the left panel. The matcher tries substring overlap first, then LLM-based matching if no phrase hits. Include varied wording users might say.

Parameters

Parameters bind session variables to entities. Set a prompt (first ask) and reprompt (invalid input). Mark required parameters so the flow waits until values are captured.

Sub-flow canvas

  • The intent entry node is seeded automatically (INTENT · ENTRY header).
  • Use the intent quick-switcher above the canvas to jump between intents.
  • Entity nodes on the canvas should align with parameters when slot-filling is needed.
  • Autosave runs across phrases, parameters, nodes, and edges together.

Fallback intent

Mark one intent as fallback to handle unmatched utterances. Configure the fallback event prompt under agent Settings for a global message before fallback routing.