Anthropic

The Messages API over raw httpx2. Registered as anthropic, with the alias claude.

streaming structured output (emulated) tool calls health discovery

Setup

client = ai.Client([
    ai.ProviderSettings.of("anthropic", api_key="env://ANTHROPIC_API_KEY"),
])
result = client.generate(prompt, target="anthropic:claude-sonnet-4-5")

No extra required.

Supported

Behavior	Support
Streaming	Native, typed SSE events
Structured output	Emulated as a forced tool call
Tools	Native
Reasoning	Extended thinking, budgeted in tokens
Usage	Input, output, plus cache read/write
Cost	Catalogued pricing

Thinking

Reasoning effort maps to a thinking budget, because Anthropic budgets in tokens rather than naming levels:

Effort	Wire form
minimal	{"type": "disabled"}
low	1024 tokens
medium	4096 tokens
high	16384 tokens

Thinking arrives as ReasoningDelta events. It starts the first-token clock — the model is working and the user sees activity — but is excluded from the answer text.

Structured output

Anthropic has no response_format field, so a schema becomes a single forced tool call, which the API does constrain. You still get a normal validated result.structured; the emulation is invisible except in structured_mechanism.

Notes

• System messages become the top-level system field.
• max_tokens is required by the API; AnyInfer sends 4096 when you set none rather than letting the request fail with a 400.
• Tool results ride on a user turn in this dialect, not a tool role.
• Model listing is cursor-paginated, and pagination is followed automatically.

Wire contract

For the exact request/response fields this adapter depends on, see contracts/anthropic.md.