Custom actions
A custom action turns an HTTP endpoint into a tool your bot can use. When a question needs live data or an operation — checking an order, creating a ticket — the AI calls your action, collects the inputs it needs, and weaves the result into its answer.
Building an action#
Open Integrations → Custom and create an action. The builder walks through four parts:
Identity
An internal name and a description. The description tells the AI what the action is for, so make it clear.System connection
The HTTP method, URL and headers of your endpoint. You can import a cURL command to fill these in at once.Parameters
The inputs the action needs — see below.Response parsing
Pick which fields to pull out of the JSON response so the bot can use them in its reply.
Importing from cURL#
If you already have a working request, paste its cURL command and the builder fills in the method, URL and headers for you:
curl https://api.example.com/orders/123 \
-H "Authorization: Bearer YOUR_TOKEN" \
-H "Accept: application/json"Parameters#
Parameters are the values the action needs to run. For each one you define:
- Field name and data type.
- Bot question — what the bot asks the customer to get this value.
- Tech description — context for the AI on how to use the field.
- Required — whether the action can run without it.
- Body path — where the value goes in the request.
Test, then attach#
Use the Test panel to run the action with real values and confirm it returns what you expect; the summary view previews how the bot will execute it.
Once it works, attach the action to a bot from its Tools & integrations settings. The AI decides when to call it based on your description.
Mind what you expose
An action can read and write in your systems. Scope tokens to the minimum needed, and prefer read-only endpoints unless a write is truly required.