valite
Sign inJoin waitlist
Docs navigation

Docs/Integrations/OpenAI SDK

OpenAI SDK

Change the base URL, add one header — chat completions, responses, embeddings, and streaming all pass through unchanged.

Setup

Requests to /openai/… are forwarded to api.openai.com byte-for-byte. Your OpenAI key stays in Authorization exactly as the SDK sends it; your valite key rides x-valite-api-key.

from openai import OpenAI

client = OpenAI(
    api_key="sk-...",  # unchanged
    base_url="https://gateway.valite.ai/openai/v1",
    default_headers={"x-valite-api-key": "vk_..."},
)

resp = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Hello"}],
)

Streaming and tools

Streaming, tool calls, JSON mode, and the Responses API work unchanged — the gateway relays whatever the OpenAI API supports. For streamed requests, valite injects stream_options.include_usage into the forwarded copy so token usage is always metered; the response you receive is not modified.

Note

Callsites using JSON mode or tool calls are treated as schema-strict: replay candidates must produce 100% parseable, valid output to be routable.

Workflow attribution

Tag the workflow in the base URL (/openai/wf/<workflow>/v1) or per request:

Python
client.chat.completions.create(
    model="gpt-4o",
    messages=[...],
    extra_headers={"x-valite-workflow": "support-triage"},
)

See Workflows & Runs for run grouping via x-valite-run-id or traceparent.

← Previous

Cost Optimization

Next →

Anthropic SDK