API

Streaming

Consume normalized server-sent events without buffering the response.

Stream with Python

import os

from openai import OpenAI

client = OpenAI(
    api_key=os.environ["SKILLROUTER_API_KEY"],
    base_url="https://api.useskillrouter.com/v1",
)

stream = client.chat.completions.create(
    model="skillrouter/seo-audit",
    messages=[{"role": "user", "content": "Audit example.com"}],
    stream=True,
    stream_options={"include_usage": True},
)

for chunk in stream:
    payload = chunk.to_dict()
    if payload.get("error"):
        error = payload["error"]
        raise RuntimeError(f"{error.get('code')}: {error.get('message')}")
    if chunk.choices and chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="", flush=True)

The wire endpoint uses data-only SSE records and terminates with data: [DONE]. Request stream_options.include_usage when you need the final usage chunk.

Cancellation and usage

Close the response body when the client no longer needs output.
SkillRouter aborts the upstream request when the disconnect can be detected.
A final usage chunk is emitted when the upstream provider supplies streaming usage.
Partial streams can create partial usage and a reconciled ledger charge.
An error before the first model event uses its normal HTTP status. A later error is a chat.completion.chunk with finish_reason error and a top-level SkillRouter error, followed by [DONE].
Do not parse SSE by splitting arbitrary network chunks on newlines without buffering incomplete frames.