Skip to main content
Updated Nov 16, 2026

Wait for the contact to finish

Use this action in an agent's tool when a contact's message reads unfinished, for example a sentence cut off halfway or a list still being typed, and answering now would mean answering half of it. The agent sends nothing now, and the conversation is looked at again a few seconds later, with everything that arrived in the meantime, as one message.

It only works in a tool the agent calls during a conversation. Anywhere else (a workflow, a pre-action, a step after the reply) the step fails.

In the editor​

In the action picker it is under Agent and conversation, as "In an agent's tool only: send nothing now, look again soon". In the flow, a step reads like: Send nothing now and look again in 20 seconds.

Under What it does there is one sentence to complete:

  1. Send nothing now, and look again in … seconds: type the Seconds. It is required.
  2. Use an expression instead turns the box into an expression, for example a tool's argument. Use a number instead turns it back.

More than 0 and 60 seconds at most: a longer wait counts as 60. The editor also says what the step does to the run:

  • This step does not stop the steps after it. When the agent's tool call has finished, the run ends without a reply from the agent, and the conversation is looked at again after the wait.
  • It only works in a tool the agent calls during a conversation, once for each message from the contact.

Later steps can read:

  • Wake-up time (deferredUntil): The date and time the conversation is looked at again, with whatever arrived during the wait.
  • Wait (seconds) (seconds): The wait actually used, in seconds: 60 at most.
  • Succeeded: Yes when this step did not fail.

Every step also has This step runs, If this step fails, the run, Edit as fields, View as JSON and Done: see The step editor.

Best for​

  • Contacts who send one thought as several short messages, when the agent notices it mid-conversation
  • Letting the agent decide, message by message, that it is too early to answer

Main fields​

The step's own fields, behind Edit as fields and in the API and MCP:

FieldWhat it does
Seconds to wait before waking the conversation (0 to 60) (seconds)Required. A CEL expression that evaluates to the wait, in seconds. It must be more than 0; values above 60 are clamped to 60

What later steps can use​

OutputMeaning
deferredUntilThe ISO-8601 date and time at which the conversation will be woken with whatever arrived during the wait
secondsThe wait actually used, in seconds, after clamping to 60

Read them with maps.get(workflow.step(N), 'deferredUntil') and maps.get(workflow.step(N), 'seconds'). This action publishes at the step root — there is no data object.

What happens during the wait​

  • The agent sends nothing from this run: no message, no error.
  • When the wait ends, the conversation is woken. If the contact sent more, the agent answers everything at once. If nothing arrived, it is woken anyway and answers what it already has.
  • One wait per message from the contact: the run started by the wake cannot wait again, and a second wait for the same message is refused. The agent then has to answer with what it has.

Tips​

  • 8 to 15 seconds is the useful range; pick how long the person takes to type, not the maximum.
  • It is not for "I need more time" or "I am still working": the contact sees nothing from the run that waits.
  • To make the agent wait after every message of a conversation, not once, use Set the reply delay instead.
  • The step fails with a code in workflow.step(N).executionContext.safeError.code: agent_wait_for_user_no_conversation (no conversation in the run), agent_wait_for_user_no_run (not inside an agent's run, for example a workflow step), agent_wait_for_user_invalid (not a number of seconds above 0), agent_wait_for_user_chained (a second wait for the same message).