AI Actions (Tools): Automate Tasks with Your Agent

AI Actions (also called Tools) transform your AI agent from a knowledge lookup bot into a true automation engine. Rather than only answering questions from static content, an agent equipped with AI Actions can reach into your backend systems in real time — looking up a specific order, checking whether a subscription is active, triggering a password reset, or issuing a refund — all without human intervention. The agent decides on its own when a tool call is necessary based on what the user is asking.

What Are AI Actions?

Tools are how your AI support agent can interact with your systems and backend in a more autonomous way. For each ticket or conversation, the agent evaluates whether it needs to use one of the tools you’ve configured, and only invokes a tool when it determines one is required.

AI Actions differ from the User Data API. The User Data API retrieves information about a specific, verified user (e.g. from an email ticket or authenticated chat). AI Actions can be triggered by any user and are better suited to lookups that don’t require identity verification, or to write-actions like refunds and updates.

What can AI Actions do?

You can build a tool for practically anything you’d use an API for today:

Data lookups — property details, product availability, order status by order ID
Write actions — process refunds, cancel subscriptions, update address records
Automation triggers — kick off a background check, create a support ticket, send a confirmation email
Knowledge enrichment — query an internal knowledge base or CRM to augment the agent’s response

How to Configure an AI Action

Open Tasks & Tools in your Dashboard

Define the API endpoint

Enter the POST API endpoint your tool will call. For example:

https://api.yourcompany.com/orders/lookup

All tools use POST requests. My AskAI will send a JSON body to this URL each time the agent invokes the action.

Add authentication headers

Add any required Authorization headers so My AskAI can authenticate with your API. For example:

Authorization: Bearer sk_live_abc123xyz

See Authorizations below for the full list of supported credential types.

Define the JSON request body

Specify the fields the tool needs to send. Include the parameters your API expects — the agent will populate these from the conversation context. For example:

{
  "email": "{{user_email}}",
  "order_id": "{{order_id}}"
}

Define the JSON response body

Tell the agent which fields to look for in the API response, so it knows how to interpret and surface the result. For example:

{
  "order": {
    "id": "ORD-9182",
    "status": "In Transit",
    "estimated_delivery": "2024-11-08",
    "carrier": "FedEx",
    "tracking_number": "794644792798"
  }
}

Save and test

Save the tool, then open the chat widget in your Dashboard and ask a question that should trigger it. Confirm the agent calls the API and returns an accurate answer. Once satisfied, the tool is active.

Authorizations

Many APIs require credentials. My AskAI stores these securely and attaches them automatically whenever the tool is invoked — your users never see them. Supported authorization methods:

Method	How it works
Bearer token	Static token passed as `Authorization: Bearer <token>`
API key header	Custom header such as `X-API-Key: <key>`
Basic auth	Base64-encoded username:password in the Authorization header
OAuth 2.0	OAuth credential flow; token refresh handled automatically

You can manage all credentials under Connections > Authorizations in your Dashboard.

Use API credentials with the minimum required scopes. If a tool only needs to read order data, use a read-only key. Avoid giving your AI agent write access to systems unless the action explicitly requires it.

Example Action Definition

Below is a representative example of a complete tool definition for an order-status lookup:

{
  "name": "get_order_status",
  "description": "Retrieves the current status and tracking information for a given order ID.",
  "endpoint": "https://api.yourcompany.com/orders/status",
  "method": "POST",
  "headers": {
    "Authorization": "Bearer sk_live_abc123xyz",
    "Content-Type": "application/json"
  },
  "request_body": {
    "order_id": "{{order_id}}",
    "email": "{{user_email}}"
  },
  "response_body": {
    "order": {
      "id": "string",
      "status": "string",
      "estimated_delivery": "string",
      "carrier": "string",
      "tracking_number": "string"
    }
  }
}

The agent extracts order_id and email from the conversation, calls the endpoint, and uses the response to compose a natural-language answer for the user.

Use Cases

Order status and shipment tracking

A user asks “Where is my order #4521?” — the agent extracts the order ID, calls your order-management API, and responds with the current status, carrier, and estimated delivery date. No human involvement required.

Subscription and account checks

A user asks “What plan am I on?” or “Is my trial still active?” — the agent looks up their subscription record and responds with their current plan, renewal date, and any active features.

Self-service refunds

A user requests a refund. The agent calls your payments API to check eligibility (e.g. within the refund window, correct item condition), confirms the refund with the user, then triggers the refund via API — all within the same conversation thread.

Password and access resets

A user is locked out. The agent calls your auth API to trigger a password reset email or generate a temporary access link, and confirms the action to the user instantly.

Internal knowledge base search

The agent can be given access to an internal search API that isn’t part of your public help center — querying internal runbooks, product specs, or database records to deliver more precise answers.

Address and contact updates

A user wants to change their delivery address. The agent collects the new address, confirms it with the user, then calls your CRM or order API to update the record — no human agent needed.

How the Agent Decides to Use a Tool

The agent does not call a tool on every message. It evaluates the user’s intent and determines whether:

The question can be answered from its existing knowledge base alone, or
Live, dynamic data is required (e.g. a specific order, a specific account) that it cannot know from static content.

If live data is needed, the agent selects the most relevant configured tool, extracts the required parameters from the conversation, and makes the API call — all transparently and in real time.

Write clear, specific descriptions for each tool so the agent can reliably decide when to use it. A description like “Retrieves live order status for a given order ID” is much better than “Order lookup.”

Pricing

Tool calls are charged at $0.02 per reply in which a tool was used. You are only charged for responses where an actual API call was made — if the agent answers from its knowledge base, there is no tool charge. For complex automations involving multiple tool calls within a structured flow, see Tasks.

Core Features

Integrations & Channels

Advanced Features

Content & Knowledge

AI Actions (Tools): Automate Tasks with Your Agent

What Are AI Actions?

What can AI Actions do?

How to Configure an AI Action

Authorizations

Example Action Definition

Use Cases

How the Agent Decides to Use a Tool

Pricing

Build docs developers (and LLMs) love

Core Features

Integrations & Channels

Advanced Features

Content & Knowledge

Documentation Index

​What Are AI Actions?

​What can AI Actions do?

​How to Configure an AI Action

​Authorizations

​Example Action Definition

​Use Cases

​How the Agent Decides to Use a Tool

​Pricing

Build docs developers (and LLMs) love

What Are AI Actions?

What can AI Actions do?

How to Configure an AI Action

Authorizations

Example Action Definition

Use Cases

How the Agent Decides to Use a Tool

Pricing