How to Create a Custom Tool to Extract Data and Fill Custom Fields

Custom tools are what let your voice or chat assistant do anything on the internet — fetch data, extract information from the conversation, or trigger a workflow in another system, with full awareness of what the response means.

How a custom tool works

A custom tool is a function call with an API call wrapper: the AI decides when to use it based on your description, sends a POST request to your endpoint with the parameters it extracted, and reads the response back into the conversation for context. This covers use cases from simple contact-info updates to multi-stage flows like generating a Stripe customer ID and sending a unique checkout link by SMS.

What each input means

• Name (raw JSON format): a unique, plain-text name matching what the tool does — update_contact, get_delivery_date. This is what the AI calls when it decides the tool applies.
• Description (conversational text is fine): the most important field. It tells the AI what the tool does and exactly when to call it. Example: “Get the delivery date for a customer’s order. Call this whenever you need to know the delivery date, for example when a customer asks ‘Where is my package.’”
• Endpoint URL: the specific API or webhook endpoint the tool sends its POST request to.
• Authorization & key: a toggle plus a Bearer Token field, for endpoints that require an API key or authorization header.
• Parameters: key/value pairs where the value is a description of what to extract and in what format. Parameter names should be lowercase JSON-style (underscores, no spaces or special characters); the description should be explicit about the data and format wanted — for example, revenue: The annualized revenue of the user in raw number format. The AI fills these values in from the conversation itself.

Example use cases from real setups

• Update contact information in a CRM.
• Trigger an inbound webhook.
• Find a contact by contact ID.
• Update contact fields using extracted conversation data.
• Search the web mid-conversation.
• Submit a support ticket on the user’s behalf.
• Screenshot or scrape a website (with authorization) and understand the response.
• Get available times from — and book appointments to — a CRM calendar, including multi-calendar setups.
• Create a Stripe checkout session using a customer ID generated by an earlier tool call in the same conversation.
• Make a contextually aware call request from an SMS conversation.
• Calculate distance using the user’s address.
• Create a new Stripe customer and return the resulting customer information.

Further reading

• OpenAI docs on function calling
• What is function calling? (plain explanation)

Next step

Once you’re comfortable with custom tools, apply the same pattern to tags and live transfer: how to create a tool to add and remove tags and how to set up live call transfer (cold and warm). If fields aren’t populating after setup, see fix: AI assistant not filling custom fields.