Skip to main content

​
Function Calling

Function calling lets agents take actions during a conversation — like checking order status, looking up account info, or booking appointments — by calling external HTTP endpoints.

​
How It Works

  1. You define functions in the agent config with a name, description, endpoint, and parameters
  2. The LLM decides when to call a function based on the conversation
  3. The agent worker makes the HTTP request to your endpoint
  4. The response is fed back to the LLM to incorporate into its reply

​
Configuring Functions

Add functions to your agent configuration: 
{
  "functions": [
    {
      "name": "get_order_status",
      "description": "Look up the status of a customer order by order ID",
      "endpoint": "https://api.example.com/orders/{order_id}/status",
      "method": "GET",
      "parameters": {
        "type": "object",
        "properties": {
          "order_id": {
            "type": "string",
            "description": "The order ID to look up"
          }
        },
        "required": ["order_id"]
      }
    },
    {
      "name": "create_ticket",
      "description": "Create a support ticket",
      "endpoint": "https://api.example.com/tickets",
      "method": "POST",
      "parameters": {
        "type": "object",
        "properties": {
          "subject": {
            "type": "string",
            "description": "Ticket subject"
          },
          "description": {
            "type": "string",
            "description": "Ticket description"
          }
        },
        "required": ["subject"]
      }
    }
  ]
}

​
Parameter Substitution

URL parameters wrapped in {braces} are automatically substituted from the function arguments. For example, {order_id} in the endpoint URL is replaced with the actual order ID from the LLM’s function call.

​
Dashboard UI

The agent settings page provides a visual editor for adding and configuring functions — no JSON editing required. Each function has fields for:
  • Function name
  • Description (used by the LLM to decide when to call it)
  • HTTP method and endpoint URL
  • Parameter schema with types and descriptions