Voice AI Custom Actions: Live Webhook Integrations
Voice AI Custom Actions fire POST webhooks mid-call so the agent can hit your CRM, scheduling, or order systems in real time. Authentication, headers, and dynamic parameters are configurable, with a built-in test tool.
Voice AI Custom Actions let your AI agent hit external APIs during a live call. The caller asks for an order status; the agent calls your order system; the answer comes back inside the same conversation. No hold time, no follow-up tasks.
What a Custom Action does
Each Custom Action is a POST webhook the AI fires when a configured trigger fires during conversation. The request can include:
- Webhook endpoint URL.
- Headers (API keys, tokens).
- Authentication — Bearer tokens, Basic Auth, or custom header keys.
- Request body with dynamic parameters extracted from the call.
Conversation triggers
Triggers control when an action fires. Common patterns:
- A specific phrase ("I want to check my appointment").
- The presence of an email address in the caller's speech.
- A spoken digit string (such as an order number).
Triggers can layer with conditions like "only run if parameter X is present" so actions only fire when there is enough context.
Dynamic parameter collection
Voice AI extracts and labels values during the call and slots them into the webhook payload. Supported types:
- Text (String)
- Number
- Phone Number
- Date
Example: the AI captures both an order number and an email before triggering a shipping lookup, so the request hits your API with both values populated.
How to set up a Custom Action
- Go to Voice AI > Agent Goals. Switch to Advanced Mode if needed, then open Custom Actions.
- Click + New Action.
- Enter a name and the conversation trigger conditions.
- Add the webhook URL and select POST.
- Add headers and any required authentication.
- Define the dynamic parameters the AI should extract from the call.
- Use Test Webhook to validate the request and response before saving.
Built-in test tool
Test Webhook simulates a call scenario, passes test parameters, and shows you:
- The full request — headers and body.
- The raw response — 200 OK, 404, etc.
- Configuration issues to fix before saving.
Example: a webhook returns an estimated delivery date, and the AI delivers the answer inline — "Your package is expected to arrive by Thursday."
Frequently Asked Questions
Can I use GET or other request types?
Where do I set up Custom Actions?
Is authentication supported?
What happens if the webhook fails mid-call?
Can multiple webhooks fire in one call?
Do I need a developer to set this up?
Was this article helpful?
Thanks for the feedback.