Actions

When actions run

Action sources

In-call actions

Add an action to the script

1

Add an Action step

Open a block on the Script tab and click Action. Edit the Action Phrase: what the agent says while the action runs (default: “Let me pull that up for you real quick.”).

2

Choose the action

Click Choose action. In Add Action, pick an action from My Actions or Connected Apps, or click Create Custom Action (HTTP) or Create Custom Code.

3

Describe when it runs

Fill Action Name and When should the custom action take place? Be specific: “When the caller gives their order number and asks where the order is”. The agent uses this description to decide when to call the action.

4

Configure and test

Fill in the request (below), click Send Test Request, check the response, and click Add.

Custom HTTP action

What your endpoint receives

POST /orders/lookup HTTP/1.1
Host: api.example.com
Content-Type: application/json
X-Api-Key: <your header, as configured>
X-Signature: 9b2d0f3c6e1a…   (hex HMAC-SHA256 of the body)
X-Signature-Type: HMAC

{"orderNumber":"A-10293","phone":"14155552671","callId":"6650c4d0a1b2c3d4e5f60901"}

• Content-Type follows your Body Type (application/json for JSON).
• X-Signature is the HMAC-SHA256 of the raw body, hex-encoded, using your account’s webhook secret. With Sign with RSA-SHA256 on, it’s a base64 RSA signature and X-Signature-Type is RSA. Verify it as described in Validate signatures.
• Requests come from the IP addresses listed under Security.

What the agent receives back

{ "status": "shipped", "carrier": "UPS", "eta": "2026-10-29", "trackingUrl": "https://…" }

Custom code

export const code = async (inputs) => {
  const res = await fetch(`https://api.example.com/customers/${inputs.customerId}`, {
    headers: { Authorization: `Bearer ${inputs.apiToken}` },
  });
  if (!res.ok) {
    return { error: `Lookup failed with ${res.status}` };
  }
  const customer = await res.json();

  // Return only what the agent needs, as a JSON-serializable value.
  return { tier: customer.tier, balance: customer.balance, renewsOn: customer.renewal_date };
};

{
  "customerId": "{{crm_id}}",
  "orderNumber": "[[orderNumber]]",
  "callId": "<#callId#>",
  "apiToken": "sk_live_…"
}

{
  "dependencies": {
    "date-fns": "3.6.0",
    "zod": "3.23.8"
  }
}

• Entry point. The source must contain export const code = async (inputs) => { … }. Anything else fails validation.
• No placeholders in code. {{ }}, [[ ]], and << >> are not substituted inside the Code or Dependencies tabs; they’d ship as literal strings. Put the value in Inputs and read it from inputs.
• Return JSON. Return a plain object, array, string, number, or boolean. Functions, class instances, and circular structures can’t be passed to the agent.
• Build time. The build badge shows Preparing environment…, Ready, or Build failed. The first build installs your dependencies and can take a few minutes; later runs reuse it. An action called while the environment is still building is retried by the agent.
• Response Timeout: how long to wait for the code to finish. Default 30 seconds.
• Transcript Variables on a code action work like Variables Extracted with Prompt on HTTP actions: the AI fills them from the conversation, and you reference them in Inputs with [[name]].

Run options

Use the response

• Answer with real data: “Your order shipped on Tuesday and arrives Friday.”
• Branch on the result: give subflows titles that describe the result, such as Account is active and Account is suspended, and the agent picks the matching path.
• Insert a value directly: add an Action tag to a later step to speak a specific field from the response.

Pre-conversation actions

• The request times out after 5 seconds.
• Mapped variables don’t need to be in your CSV; they’re optional on upload. Names such as contactId and callId are reserved.
• Request data can use contact values like {{name}}, {{phoneNumber}}, and {{contact.data.accountId}}, plus metadata like {{callId}}.

{
  "preCallActions": [
    {
      "activePieceAction": {
        "id": "HTTP",
        "actionId": "send_request",
        "requestData": {
          "method": "GET",
          "url": "https://api.example.com/customers?phone={{contact.phoneNumber}}&account={{contact.data.accountId}}",
          "headers": { "Authorization": "Bearer sk_live_…" }
        }
      },
      "contactMapper": {
        "leadScore": "data.leadScore",
        "accountManager": "data.owner.name"
      },
      "timeout": 5000,
      "executionScope": "EVERY_ATTEMPT",
      "storeInfo": false,
      "suppressCallOnHttpError": true
    }
  ]
}

Post-call actions

1. On the Workflow tab, click Add on Post-Call Actions.
2. In Add Webhook, pick an App Center action or click Create Custom Webhook.
3. Under Perform action for (or Send the webhook for), select the events that trigger it, for example a specific success criterion, voicemail, no answer, or call finished. An action with no events selected shows Not Selected and doesn’t run.

Security

IP allowlist

35.245.204.86
35.221.61.75

Request signing

Test and debug

• Before saving: Send Test Request asks for mock values for each variable. Click Generate mock values to fill them with AI. The response appears on the right.
• During test calls: pre-conversation actions run by default. Turn off Run Pre-Conversation Actions in the test form to skip them.
• After a call: open the call transcript. Actions appear in sections (Pre-Conversation Actions, in-call actions inline, Post-Conversation Actions). Click any action to see its Headers, Request, and Response. See all actions lists them by stage.

Troubleshooting

Related