Tool Calling (Function Calling)

Enable your AI models to call external functions and APIs using OpenAI-compatible tool calling.

Overview

Tool calling allows models to generate structured JSON arguments for functions you define. Instead of answering from their training data, models can request that your application executes a function and returns the result - enabling real-time data access, calculations, and external API integrations.

Oxlo's tool calling is fully compatible with the OpenAI tools API format, so you can switch providers without changing your code.

Supported Models

The following models support tool calling with both single and parallel tool invocation:

Model	Single Tool	Parallel Tools	Notes
deepseek-v3-0324
deepseek-v3.2
deepseek-v4-flash			1M context
deepseek-r1-0528			Reasoning model - intermittent multi-tool
llama-4-maverick-17b
llama-3.3-70b			Azure limits to 1 tool definition
qwen-3-32b
qwen-3-coder-30b
gpt-oss-120b
gpt-oss-20b
kimi-k2.5
kimi-k2.6			Also supports reasoning
kimi-k2-thinking
oxlo-claw			GPT-4o class
mistral-7b
glm-5
minimax-m2.5

Basic Usage

Define your tools in the tools array and set tool_choice to "auto". The model will decide when to call a tool based on the user's message.

python

from openai import OpenAI

client = OpenAI(
    base_url="https://api.oxlo.ai/v1",
    api_key="YOUR_API_KEY"
)

tools = [{
    "type": "function",
    "function": {
        "name": "get_weather",
        "description": "Get the current weather in a city",
        "parameters": {
            "type": "object",
            "properties": {
                "location": {"type": "string", "description": "City name"},
                "unit": {"type": "string", "enum": ["celsius", "fahrenheit"]}
            },
            "required": ["location"]
        }
    }
}]

response = client.chat.completions.create(
    model="deepseek-v3.2",
    messages=[{"role": "user", "content": "What's the weather in Tokyo?"}],
    tools=tools,
    tool_choice="auto"
)

# Check if the model wants to call a tool
choice = response.choices[0]
if choice.finish_reason == "tool_calls":
    for tool_call in choice.message.tool_calls:
        print(f"Function: {tool_call.function.name}")
        print(f"Arguments: {tool_call.function.arguments}")

Multi-Step Tool Calling

After executing a tool, send the result back to the model to continue the conversation. This enables multi-step workflows where the model reasons about tool results.

python

import json

# Step 1: Model requests a tool call
response = client.chat.completions.create(
    model="kimi-k2.6",
    messages=[{"role": "user", "content": "What's the weather in Tokyo?"}],
    tools=tools,
    tool_choice="auto"
)

tool_call = response.choices[0].message.tool_calls[0]

# Step 2: Execute the function (your code)
weather_result = {"temperature": 22, "condition": "Sunny", "unit": "celsius"}

# Step 3: Send result back to the model
messages = [
    {"role": "user", "content": "What's the weather in Tokyo?"},
    response.choices[0].message,  # assistant message with tool_calls
    {
        "role": "tool",
        "tool_call_id": tool_call.id,
        "content": json.dumps(weather_result)
    }
]

# Model generates final answer using the tool result
final = client.chat.completions.create(
    model="kimi-k2.6",
    messages=messages,
    tools=tools
)
print(final.choices[0].message.content)
# "The weather in Tokyo is 22°C and sunny!"

Limitations

Llama 3.3 70B: Only supports 1 tool definition per request. If you send multiple tools, Oxlo automatically truncates to the first tool and retries.

DeepSeek R1 models: Reasoning models (R1-70B, R1-8B) do not support tool calling. If tools are sent, Oxlo gracefully falls back to a text response.

Gemma 3 models: Gemma 3 (27B, 4B) outputs tool calls as code blocks rather than structured JSON. Structured tool_calls support is planned.