Migration

Migrating from OpenAI

Move OpenAI API traffic to TokenRouter with a base URL swap.

Migration is two lines: change the base URL and swap your OpenAI key for a tr_ key. Everything else — request shapes, streaming, tools, SDK types — stays the same.

The change

python

from openai import OpenAI

client = OpenAI(
-   api_key=os.environ["OPENAI_API_KEY"],
+   api_key=os.environ["TOKENROUTER_API_KEY"],  # tr_...
+   base_url="https://api.tokenrouter.io/v1",
)

Model names keep working: gpt-5-mini resolves as a bare alias, or use the explicit openai/gpt-5-mini. Add a provider key for OpenAI in the console first — TokenRouter forwards requests on your key, with zero markup.

What carries over

/v1/chat/completions with streaming, stream_options.include_usage, tool calling, JSON mode, and vision inputs.
/v1/responses — stateless subset.
/v1/embeddings — passthrough to OpenAI embedding models.
/v1/models — now lists every model across your providers.

Not supported

Routes outside the gateway’s scope are not proxied: fine-tuning, Assistants, Files, Batches, and Uploads. Keep calling api.openai.com directly for those workflows; point your inference traffic at TokenRouter.

The upside

Once traffic flows through the gateway, you can put hard-cap budgets on every team, restrict models per environment, and swap in any of the other 12 providers — or auto — without touching application code again.