Function calling

Equip Meridian with tools it can invoke. Define a JSON schema, set tool_choice, and receive structured arguments ready for your Python runtime.

Defining a tool

A tool is a JSON object with a name, description, and a JSON Schema parameters block. Meridian uses the schema to validate arguments before calling your function.

tools = [
  {
    "type": "function",
    "function": {
      "name": "get_weather",
      "description": "Return current conditions for a city.",
      "parameters": {
        "type": "object",
        "properties": {
          "city": {
            "type": "string",
            "description": "City name, e.g. Stockholm"
          },
          "units": {
            "type": "string",
            "enum": ["celsius", "fahrenheit"],
            "description": "Temperature unit"
          }
        },
        "required": ["city"]
      }
    }
  }
]

Controlling tool_choice

By default Meridian decides whether to call a tool. Force a specific tool — or require any tool — with the tool_choice parameter.

Python example

Send tools alongside your prompt, inspect the response for a tool_calls array, execute the matching function, and return the result to Meridian.

import meridian

client = meridian.Client()

tools = [
  {
    "type": "function",
    "function": {
      "name": "get_weather",
      "description": "Return current conditions for a city.",
      "parameters": {
        "type": "object",
        "properties": {
          "city": {"type": "string"}
        },
        "required": ["city"]
      }
    }
  }
]

response = client.chat.completions.create(
  model="meridian-1",
  messages=[{"role": "user", "content": "What's the weather in Oslo?"}],
  tools=tools,
  tool_choice="auto",
)

# Check if the model wants to call a tool
if response.choices[0].message.tool_calls:
  tool_call = response.choices[0].message.tool_calls[0]
  if tool_call.function.name == "get_weather":
    import json
    args = json.loads(tool_call.function.arguments)
    city = args["city"]
    # Call your actual weather API here
    result = f"It is 12 °C and cloudy in {city}."

    # Return the tool result to Meridian
    response = client.chat.completions.create(
      model="meridian-1",
      messages=[
        {"role": "user", "content": "What's the weather in Oslo?"},
        response.choices[0].message,
        {"role": "tool", "tool_call_id": tool_call.id, "content": result},
      ],
    )

print(response.choices[0].message.content)