Conditional actions

By default, playbooks use LLMs to control the conversation, take actions, and respond to end-users. If you need more explicit control over the agent's behavior, you can use flows or conditional actions, which are described in this guide.

Conditional actions are an optional configuration for playbooks. You can set conversation triggers and conditions that invoke specific actions, such as responding with a specific message, calling an external tool, invoking another playbook, or performing any other specified action.

Triggers

You can define triggers as the first step to determine whether a conditional action is applicable.

The following list details trigger types and the specific triggers for each type:

  • Lifecycle stage:
    • Playbook start: Triggers the first time a playbook is started.
    • Before the LLM decides its next action: Triggers every time before the LLM predicts the next action.
    • Before the LLM executes its next action: Triggers every time before the LLM executes an action.
  • Event:
    • Custom event
    • No input from user

Conditions

Evaluate any optional conditions after a trigger occurs. Conditions use the same condition syntax used by flow routes.

If you don't supply conditions, the trigger alone initiates actions.

You can use the following syntax to reference parameters and last and next actions:

  • Session parameters:
    • $session.params.PARAMETER_NAME
  • Playbook input parameters:
    • $playbook.input.INPUT_FIELD_NAME
  • Last action (previously executed action):
    • $last-action.name
    • $last-action.input.INPUT_FIELD_NAME
    • $last-action.output.OUTPUT_FIELD_NAME
    • $last-action.status
  • Next action (LLM-predicted next action):
    • $next-action.name
    • $next-action.input.INPUT_FIELD_NAME

You can also use system functions.

Examples:

  • $session.params.current_user = null
  • $last-action.name = "query_order"
  • $last-action.name = "query_order" AND $last-action.status != SUCCESS
  • $last-action.name = "query_order" AND $last-action.output.orders = null

Actions

After a trigger and optional condition evaluation occur, you can configure the following actions:

  • Provide a response to the end-user: Text, conditional response, custom payload, and prerecorded audio. When supplying text, you can insert [pause], [pauseShort], or [pause long] to inject pauses in agent speech, for example, How can I help you today? [pause long] I can help with...
  • Override the next LLM action: Tool use, playbook invocation, playbook completion, flow invocation, playbook transition, flow transition.
  • Change speech settings: Model name, no speech timeout, endpointer sensitivity, adaptation.
  • Enable barge-in.
  • Allow cancellation of partial response playback.
  • Send Call Companion SMS.
  • Enable DTMF: See DTMF for telephony integrations.
Previous
Code blocks
Next
Playbook examples