Skip to main content
Actions connect an agent to your systems. They can pull context before a conversation starts, fetch or write data while the agent is talking, and follow up once the outcome is known. Each action is either an App Center app action, a custom HTTP request, or custom code. This page explains the three entry points so you can put each piece of work in the right place.

​
Action entry points

Agent workflow with pre- and post-conversation action nodes

Pre- and post-conversation actions in the agent workflow

In chat agents the same stages are labeled Pre-Chat Actions and post-chat outcomes. The behavior is the same.

​
Pre-conversation actions

Use these for anything the agent must know before it says hello. When a call is skipped because a pre-conversation action failed, the attempt is marked as not placed because it did not pass the pre-call checks. Pre-conversation action requests and responses appear in the conversation transcript for debugging.

​
In-call actions

In-call actions run inside the script, while the conversation is live.
Script canvas with the If interested block open: step 2 holds an action phrase and an HTTP (Send HTTP request) action named add_to_crm

An action step inside a script block

Key settings (full reference in Actions): To pause the script until one or more actions return, add a Wait step after them.

​
Post-conversation actions

Post-conversation actions run after the outcome is known. Set each action to run only for specific outcomes, such as goal reached, Call me later, or a custom success criterion like “Follow-Up Scheduled”, so your systems only receive relevant results. You can use:
  • Contact variables
  • Answers from post-call analysis (key questions)
  • The success criteria result
  • Variables Extracted with Prompt: values the AI extracts from the transcript after the call
Combine an outcome filter with key-question answers. For example, run “Create deal” only when the goal was reached, and pass the budget answer into the deal amount.
For a single request that carries the full call result (transcript, analysis, recording URL), use a webhook instead of a post-conversation action.

​
Ways to connect an action

Callers sends actions and webhooks from these IP addresses. If your systems are behind a firewall, allow both:
Requests from HTTP actions are signed. See Validating webhook signatures.

​
Plan your actions

1

Map the journey

List what must happen before, during, and after the conversation, and the data each step needs.
2

Connect your apps

Authorize every tool in the App Center so its actions appear in the editor.
3

Add actions to the right stage

Put enrichment in pre-conversation actions, live lookups in script steps, and follow-ups in post-conversation actions with outcome filters.
4

Test end to end

Use Send Request on each action, then run test calls and check the action logs in the transcript.

​
Troubleshooting

If Prevent call on unsuccessful response is on, a request that returns a 4xx or 5xx status skips the call. Open a recent attempt’s transcript to see the request and response, fix the endpoint or increase the timeout, or turn the setting off.
Long-running requests hold the conversation. Shorten the request, turn on Run in background if the agent doesn’t need the result to continue, and add a fallback phrase so the caller isn’t left in silence.
Check its outcome filter. It only runs for the outcomes you selected. Post-conversation actions also wait for post-call analysis to finish.
Allow the two IP addresses above in your firewall, and make sure the URL is reachable over the public internet.

Actions

Full configuration reference for every action type.

Variables

Contact and transcript variables you can pass to actions.

App Center

Connect the apps your actions use.

Webhooks

Receive the full conversation result on your server.