Tool calling - ClearMaas

OpenAI-style tools / tool_choice works against every chat-capable provider. The gateway translates to the upstream’s native tool-call shape.

Minimal example

from openai import OpenAI
client = OpenAI(base_url="https://api.clearmaas.com/v1", api_key="sk-...")

tools = [{
    "type": "function",
    "function": {
        "name": "get_weather",
        "parameters": {
            "type": "object",
            "properties": {"city": {"type": "string"}},
            "required": ["city"],
        },
    },
}]

resp = client.chat.completions.create(
    model="openai/gpt-4o-mini",
    messages=[{"role": "user", "content": "Weather in Paris?"}],
    tools=tools,
)
print(resp.choices[0].message.tool_calls)

Cross-provider behavior

Same client code works across every chat-capable provider �?ClearMaas adapts your OpenAI-style tools to each upstream’s native shape:

OpenAI / Grok / DeepSeek: native target �?tools keeps the OpenAI shape on the wire
Anthropic: OpenAI tools is mapped to Anthropic tools with input_schema, preserving JSON Schema properties and required
Gemini: OpenAI tools is mapped to Gemini tools[].functionDeclarations, with name / description / parameters carried through

Gemini reserved function names

On Gemini targets the gateway recognizes three reserved function.name values and turns them into Gemini’s native built-in tools instead of custom function declarations:

Reserved name	Maps to
`googleSearch`	Gemini Google Search grounding (see Web search)
`codeExecution`	Gemini built-in code execution
`urlContext`	Gemini built-in URL-context tool

Functions with one of these names don’t need parameters �?pass {type: "function", function: {name: "googleSearch"}} and Gemini takes it from there. Pick a different name for your own custom tools so you don’t collide with these built-ins.

​Minimal example

​Cross-provider behavior

​Gemini reserved function names

Minimal example

Cross-provider behavior

Gemini reserved function names