Provider Options

Provider-specific request options belong in ModelRequest.options, namespaced by provider.

Why

This keeps the runtime core provider-neutral while still allowing provider adapters to expose vendor-specific controls.

Current namespaces

• openai
• openai_chat
• anthropic

OpenAI Responses example

request = ModelRequest(
    messages=[Message.user("Summarize this briefly.")],
    options={
        "openai": {
            "extra_body": {
                "reasoning": {"effort": "medium"},
            },
        },
    },
)

OpenAI Chat Completions example

request = ModelRequest(
    messages=[Message.user("Summarize this briefly.")],
    options={
        "openai_chat": {
            "extra_body": {
                "store": False,
            },
            "temperature": 0.2,
        },
    },
)

Anthropic Messages example

request = ModelRequest(
    messages=[Message.user("Summarize this briefly.")],
    options={
        "anthropic": {
            "max_tokens": 512,
            "thinking": {
                "type": "enabled",
                "budget_tokens": 1024,
            },
            "metadata": {
                "user_id": "u_123",
            },
        },
    },
)

Recommendation

Keep constructor arguments for connection-level setup such as:

• api_key
• base_url
• custom SDK client

Keep request-level behavior in ModelRequest.options.