How-To: Actions

An Action represents a specific task or operation that your Agent performs during an interaction, such as gathering information, making decisions, or triggering external processes.

​

Available Actions

Actions are triggered when certain conditions are met. For both “End Conversation” and “Transfer Call,” there are two primary trigger types:

1. Phrase-based
2. Function-based

​

Phrase-Based Triggers

Phrase-based triggers offer a more consistent method but are somewhat restrictive. The action is triggered when the Agent says a specific phrase.

For instance, if you want to transfer a call to a human, you would configure the action like this:

You can prepare your prompt for this scenario and specify that, if the customer wants to talk to the support team, the Agent will say a specific phrase. For example:
“If the customer wants to talk to or be transferred to the support team, say the phrase ‘Let me transfer you to the support team.’”
Once the phrase is detected, the action is triggered.

​

Function-Based Triggers

Function-based triggers give the Agent more flexibility. The Agent decides when the conditions are met and when to perform the action. This approach is less restrictive but may be less predictable.

Feel free to explore and experiment with both methods!

​

Available Actions:

• End Conversation: Triggered by a function call or phrase-based command.
• DTMF: No specific trigger required.
• Transfer Call: Triggered by a function call or phrase-based command, along with the phone number for the transfer.
• External: (Check below).

​

External Actions

External actions provide the most flexibility, allowing your Agent to interact with external endpoints during a call to create or retrieve data.

Let’s break it down:

• Action Name
  The name of your external action.

• Description
  A brief description of your external action.

• URL
  The endpoint where the data will be sent.

• Speak on Send
  The message the Agent will speak when the external action is triggered. Example: “Let me retrieve your contact information for you.”

• Speak on Receive
  The message the Agent will speak when the external action returns the data. Example: “Ok, I got your contact information.”

• Processing Model
  Currently, only one mode is supported: ‘Muted.’

• Signature Secret
  Currently not supported.

• Input Schema
  The heart and soul of your external action. The Agent will fill out the input schema, and once the requirements are met, the external action will be triggered.
  The input schema is in JSON format.
  We use OpenAI’s function-call functionality; more information can be found here.
  The input JSON follows this schema:

{
  "type": "object",
  "required": ["user_id", "user_email", "issue", "date", "user_name"],
  "properties": {
    "date": {
      "type": "string",
      "format": "date",
      "description": "The date the issue was reported, in YYYY-MM-DD format."
    },
    "issue": {
      "type": "string",
      "description": "A description of the issue reported by the user."
    },
    "user_id": {
      "type": "number",
      "description": "The unique identifier for the user."
    },
    "user_name": {
      "type": "string",
      "description": "The user name."
    },
    "user_email": {
      "type": "string",
      "description": "The email address of the user."
    }
  }
}

You can explore our API’s Actions Creation feature here.

Ready to continue? Check out our Webhooks Guide for more details.