Function Calling

What is it?

Function Calling is a feature that allows you to call external APIs from your Voker deployments.

How to use

Start by creating a new function.
- Specify the API Endpoint
  - Enter the full URL of the API
  - Choose the correct HTTP method (GET, POST)
- Define the inputs
  - Define all required inputs for the API, including
    - Query parameters
    - Headers
    - Body
  - For each input, specify:
    - Name
    - Input type (static, input, function parameter)
      - For any input marked as a function parameter, be sure to add a clear description — this is what the LLM sees when deciding how to fill in values.
      - If it’s a body field, also specify the type (e.g., text, whole number, etc.)

Run a Test Execute to validate the function.
- Make sure the LLM is collecting the correct information from the conversation (e.g., query parameters, headers, or body fields) needed to make the API call.
- Note: This works alongside other functions such as Internet Search.

Verify that the function is called successfully.
- Once the function is triggered, a collapsed section will appear in the chat showing the function call details.
- In the input section, you should see the function name and all provided inputs.
- In the output section, you should see the API response.

Supported HTTP Request Structures

Our UI allows you to configure HTTP requests for LLM powered function calling with the following structures:

HTTP Method: Select from GET, POST, PUT, or PATCH.
URL and Path Parameters: Enter the base URL (e.g., https://api.example.com). Add path parameters using the “Add Path Param” button, which inserts placeholders like {param_name} into the URL. These can be static values, bound to inputs, or function parameters filled by the LLM.
Query Parameters: Add key-value pairs that will be appended to the URL as query strings (e.g., ?key=value). Each can be:
- Static: A fixed key and value. (e.g. API token)
- Input Binding: Bound to a specified Voker agent input.
- Function Parameter: A key with a value provided by the LLM, including a description for the LLM to understand its purpose.
Headers: Similar to query parameters, add key-value pairs for headers.
Body Parameters: Configure the request body as either an object {} or an array [] of objects (toggle via the “Array” button).
- Each body item is a flat key-value pair, with the same options as headers and query params (static, input binding, or function parameter).
- For function parameters in the body, specify the data type: Text (string), Whole Number (integer), Decimal Number (number), or True/False (boolean).
- Note: Nested objects or arrays within the body are not currently supported. The body is limited to a flat structure or an array of flat objects.

Example HTTP Requests for Function Calling

✅ Supported: Flat Object in Body

This example shows a valid POST request with a flat object in the body, including static headers/ path parameters.

curl -X POST \
  'https://api.example.com?filter=active' \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer abc123' \
  -d '{
    "name": "John Doe",
    "age": 30,
    "is_active": true
  }'

✅ Supported: Array in Body

This example shows a valid POST request with an array of flat objects in the body, including static headers and a query parameter.

curl -X POST \
  'https://api.example.com/users?status=active' \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer xyz789' \
  -d '[
    {
      "id": 1,
      "name": "Jane Smith",
      "role": "developer"
    }
  ]'

❌ Not Supported: Nested Object in Body

This example shows a POST request with a nested object in the body, which is not supported on out platform.

curl -X POST \
  'https://api.example.com/users/12345' \
  -H 'Content-Type: application/json' \
  -d '{
    "name": "John Doe",
    "details": {
      "age": 30,
      "address": {
        "street": "123 Main St",
        "city": "Anytown"
      }
    },
    "is_active": true
  }'