Tool Calling and Function Calling with Meliai Models

Tool calling — sometimes called function calling — lets a model signal that it wants to invoke a specific function in your application rather than generating a plain text reply. Your code executes the function, returns the result to the model in a follow-up message, and the model incorporates that result into its final response. This two-step pattern is the foundation for building agents, enriching responses with real-time data, and automating structured workflows. Meliai’s implementation is fully compatible with the OpenAI tools interface, so existing tool-calling code points at Meliai with no changes beyond the base_url and API key.

Define Tools

Pass a tools array to the chat completions request. Each element has type: "function" and a function object that describes the name, purpose, and expected parameters using a JSON Schema subset.

Python

tools = [
    {
        "type": "function",
        "function": {
            "name": "get_weather",
            "description": "Get the current weather for a city",
            "parameters": {
                "type": "object",
                "properties": {
                    "city": {"type": "string", "description": "City name"}
                },
                "required": ["city"],
            },
        },
    }
]

Write clear, specific description strings. The model uses these descriptions — not your code — to decide whether and when to call a tool, so precision here directly affects reliability.

Full Example

The example below registers two tools, sends a user message, and inspects the model’s response to determine whether a tool call was requested.

Python

import os
import json
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["MELIAI_API_KEY"],
    base_url="https://api.meliai.ai/v1",
)

tools = [
    {
        "type": "function",
        "function": {
            "name": "get_weather",
            "description": "Get the current weather for a city",
            "parameters": {
                "type": "object",
                "properties": {
                    "city": {"type": "string", "description": "City name"}
                },
                "required": ["city"],
            },
        },
    },
    {
        "type": "function",
        "function": {
            "name": "search_web",
            "description": "Search the web and return a list of relevant results",
            "parameters": {
                "type": "object",
                "properties": {
                    "query": {"type": "string", "description": "Search query string"}
                },
                "required": ["query"],
            },
        },
    },
]

response = client.chat.completions.create(
    model="<MODEL_ID>",
    messages=[{"role": "user", "content": "What's the weather in Hamburg?"}],
    tools=tools,
    tool_choice="auto",
)

message = response.choices[0].message
if message.tool_calls:
    tool_call = message.tool_calls[0]
    args = json.loads(tool_call.function.arguments)
    print(f"Calling {tool_call.function.name} with {args}")
    # Execute your function and send the result back (see next section)

Sending Tool Results Back

After you run the function, append the original assistant message and a new tool role message to the conversation, then make a second request. The model will use the tool output to generate its final reply.

Python

# 1. Capture the assistant's tool-call message
assistant_message = response.choices[0].message

# 2. Execute your function
weather_result = {"temperature": "18°C", "condition": "Partly cloudy"}

# 3. Build the follow-up messages list
messages = [
    {"role": "user", "content": "What's the weather in Hamburg?"},
    assistant_message,                          # preserves tool_calls field
    {
        "role": "tool",
        "tool_call_id": assistant_message.tool_calls[0].id,
        "content": json.dumps(weather_result),
    },
]

# 4. Send the second request — no tools array required for the final turn
final_response = client.chat.completions.create(
    model="<MODEL_ID>",
    messages=messages,
)

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

Always pass the assistant message object (not a reconstructed dict) so that the tool_calls field is preserved exactly as returned by the API.

tool_choice Options

The tool_choice parameter controls how the model decides whether to call a tool.

tool_choice values

Value	Behaviour
`"auto"`	The model decides whether to call a tool or reply with text (default)
`"none"`	The model will never call a tool — respond with text only
`"required"`	The model must call at least one tool before replying
`{"type": "function", "function": {"name": "get_weather"}}`	Force the model to call a specific function

Use "required" when you need a guaranteed structured output via a tool and want to avoid a plain-text fallback.

Checking Tool Calling Support

Not every model supports tool calling. To check whether a specific model accepts the tools parameter, retrieve its metadata from the models endpoint and inspect supported_parameters.

curl

curl https://api.meliai.ai/v1/models/<MODEL_ID> \
  -H "Authorization: Bearer $MELIAI_API_KEY"

Look for "tools" in the supported_parameters array of the response before building tool-calling workflows against a new model.

Use GET /v1/models to list all available models and filter by supported_parameters to find every model that supports tool calling. Models without "tools" in that list will return an error if you include a tools array in the request.

Overview

Models & Routing

Features

Privacy

Tool Calling and Function Calling with Meliai Models

Define Tools

Full Example

Sending Tool Results Back

tool_choice Options

Checking Tool Calling Support

​Define Tools

​Full Example

​Sending Tool Results Back

​tool_choice Options

​Checking Tool Calling Support

Define Tools

Full Example

Sending Tool Results Back

tool_choice Options

Checking Tool Calling Support