Function Calling (experimental)

Generate function calls based on user queries

Please note that this is a research preview of our function calling model

Overview

Fastino’s Function Calling model enables natural language execution of structured API calls based on user-defined schemas. This model interprets intent from freeform input and returns arguments for one or more functions in a machine-executable format. It supports multiple endpoints, nested parameters, and is ideal for building task-oriented virtual agents, workflow automation, or natural language interfaces over structured APIs.

Example Use Cases

• Conversational agents for customer support, service scheduling, or e-commerce

• Executing booking or reservation flows via natural language

• Smart form filling from unstructured queries

• Multistep workflow triggering in enterprise systems

• Extracting function-ready arguments for backend automation

Usage

Example Body

{
  "model_id": "fastino-function-calling-••••••••••",
  "input": [
    {
      "text": "Schedule a car maintenance service at the Downtown Auto Center on 15/09 at 10 AM for a sedan",
      "parameters": [
        {
          "name": "roadside_assistance_request",
          "parameters": {
            "type": "dict",
            "properties": {
              "location": {
                "type": "str",
                "description": "The location of the car"
              },
              "issue_description": {
                "type": "str",
                "description": "The description of the issue"
              },
              "estimate_arrival_time": {
                "type": "str",
                "description": "The estimated arrival time of the roadside assistance"
              }
            }
          },
          "description": "Request roadside assistance for a car"
        },
        {
          "name": "car_maintenance_scheduling",
          "parameters": {
            "type": "dict",
            "properties": {
              "car_type": {
                "type": "str",
                "description": "The type of car"
              },
              "appointment_date": {
                "type": "str",
                "description": "The date of the appointment"
              },
              "appointment_time": {
                "type": "str",
                "description": "The time of the appointment"
              },
              "service_location": {
                "type": "str",
                "description": "The location of the car"
              }
            }
          },
          "description": "Schedule a car maintenance service"
        },
        {
          "name": "test_drive_booking",
          "parameters": {
            "type": "dict",
            "properties": {
              "time": {
                "type": "str",
                "description": "The time of the test drive"
              },
              "dealership": {
                "type": "str",
                "description": "The dealership of the car"
              },
              "desired_date": {
                "type": "str",
                "description": "The date of the test drive"
              },
              "model_of_interest": {
                "type": "str",
                "description": "The model of the car"
              }
            }
          },
          "description": "Book a test drive for a car"
        }
      ]
    }
  ]
}

Example Response

[
  {
    "car_maintenance_scheduling": {
      "appointment_date": "15 / 09",
      "appointment_time": "10 am",
      "car_type": "sedan",
      "service_location": "downtown auto center"
    }
  }
]

Notes

• Each entry in the parameters array corresponds to a callable function with a name, description, and typed schema.

• The model returns a list of invocations with populated parameters when a matching intent is detected.

• Function names must be unique.

• Ideal for integrating with JSON-RPC, OpenAPI, or LangChain tool call workflows.

Let me know if you’d like a matching Markdown version or landing page variant.