Actions

Pro & Max

Connect your chatbot to your own API so it can look things up or send data in the middle of a conversation.

An Action is a request your chatbot can make to an HTTPS API you run. When a visitor asks "where's my order?", the chatbot can call your order-status endpoint and answer with the real status. Actions work in two directions:

  • Look up information — order status, stock levels, account details.
  • Submit data — leads, support tickets, service requests.

Actions run on website chat, WhatsApp, Instagram, and Messenger. The request is made from Hyperleap's servers, not the visitor's browser, and your credentials are never shown to the AI model or the visitor.

Before You Start

  • Pro or Max plan — On other plans, saved Actions stay readable but can't be added or changed.
  • An API you control — A public HTTP or HTTPS endpoint. Requests to private or internal network addresses are blocked, and redirects aren't followed, so use the final URL.
  • GPT-6 Luna as the model — Claude models answer in plain chat and don't run Actions. See System Prompt & Models.

Open your chatbot, select the Integrations tab, and choose "Actions". Start from "Information lookup", "Data submission", or "Custom API". A chatbot can have up to 10 Actions.

Actions tab with Information lookup, Data submission, and Custom API starting points
Integrations → Actions

Configure an Action

The editor walks you through five steps. Click "Next" to move between them.

Configure an Action window on the Purpose step with Action name, Purpose and when to use it, and Visitor progress text
Step 1, Purpose: name the Action and tell the AI when to use it
1

Purpose

  • Action name — Letters, numbers, underscores, or hyphens, unique within the chatbot.
  • Purpose and when to use it — The AI reads this to decide when to call the Action, so be specific: "Use when a customer asks where their order is and gives an order number."
  • Visitor progress text — What the visitor sees while the request runs, such as "Checking your order…".
2

Request

Choose the HTTP method (GET, POST, PUT, PATCH, or DELETE) and enter the Endpoint URL. Put path inputs in the URL as placeholders, such as https://api.example.com/orders/{order_number}. Add up to 10 Non-secret headers if your API needs them; credentials go in the next step.
Configure an Action window on the Request step with HTTP method, Endpoint URL, and Non-secret headers
Step 2, Request: the method and the endpoint your API exposes
3

Authentication

Pick No authentication, Bearer token, API key header, or Connected account. The Credential field is write-only: it's stored encrypted and never displayed again. Leave it blank later to keep the saved one.
4

Inputs

Describe what the chatbot should collect from the visitor, up to 20 inputs. Each has a name, a type (string, number, or true/false), a location (URL path, query string, or a flat JSON body), and a Required checkbox. Every placeholder in the URL needs a matching required path input, and GET Actions can't use body inputs.
5

Review

Check the destination and permissions, leave Enabled ticked, and click "Add to draft".
Note:
With Connected account, the account's token is sent to your endpoint as a Bearer token. Only point it at an endpoint you trust with that account.

Example: Order Status Lookup

Action nameorder_status
PurposeUse when a customer asks about an order and gives an order number.
RequestGET https://api.example.com/orders/{order_number}
AuthenticationAPI key header
Inputsorder_number · string · path · required

Test and Publish

Each Action card shows where it stands:

  • Authentication ready or Setup incomplete — An Action can't be switched on until its authentication is complete.
  • Draft changes or Published — Edits reach your live chatbot when you publish.
  • Edit, Duplicate, and Remove, plus an on/off switch for each Action.

To try an Action before publishing, click "Save draft and test", then "Save draft" and "Open fresh draft preview", and ask a question that should trigger it.

Note:
Testing and live conversations make real requests to your API. Use test accounts and test data while you try a data-submission Action.

Good to Know

  • Your endpoint has 20 seconds to respond.
  • Actions connect the chatbot to your API. They aren't a built-in integration with the system behind it, so your API decides what data the chatbot can see or change.
  • Restoring an older version from Version History doesn't restore credentials or connected accounts. An Action missing its credential comes back switched off so you can repair it.
Tip:
Only need to notify your systems after something happens, without the chatbot reading a reply? Webhooks are simpler.

Explore Hyperleap AI

An AI front desk that answers customer questions, collects contact details, and books appointments on your website and messaging channels.