Migrate from OpenRouter

You can keep the OpenAI Python or TypeScript SDK for Chat Completions. Point it at AnonRouter, use an AnonRouter inference key, and confirm that your model ID and request options are supported. Migrate one call before switching production traffic.

1. Set up your key and model

Create an inference API key with OpenAI-compatible access enabled and store it in the server-side environment variable ANONROUTER_API_KEY.
Pick a callable chat model from the model catalog and set ANONROUTER_MODEL to its exact API ID. Model names can overlap across services, but availability and capabilities vary by provider route.
Set the SDK base URL to https://api.anonrouter.ai/v1. Remove any old provider-specific default headers and routing options for the first test.

Create an API key

You can look up callable text model IDs without sending a key:

curl -fsS https://control.anonrouter.ai/v1/catalog/models |
  jq -r '.models[] | select(.callable == true and .primaryModality == "text") | .id'

Compatibility mode accepts a static bearer key. The compatibility broker handles key identity and request content together in memory. Use the private ticket flow when you need those kept separate. The ticket flow requires a separate authorization request.

2. Send a chat completion

Run the examples on a trusted server with the OpenAI SDK installed and both environment variables set. The examples can use your balance. Automatic retries are disabled because a disconnect can occur after a paid generation.

import os

from openai import OpenAI

client = OpenAI(
    base_url="https://api.anonrouter.ai/v1",
    api_key=os.environ["ANONROUTER_API_KEY"],
    max_retries=0,
)

completion = client.chat.completions.create(
    model=os.environ["ANONROUTER_MODEL"],
    max_completion_tokens=256,
    messages=[{"role": "user", "content": "Hello"}],
)

print(completion.choices[0].message.content)

With the same authenticated client, use client.models.list() for model discovery. Read each entry’s provider_routes when choosing a specific provider. Top-level capabilities describe the primary route.

3. Review request and routing differences

If your app uses	What to check on AnonRouter
Automatic model selection	Use `/auto` and review your routing policy. Use an exact model ID when you need predictable model selection.
Provider filters, sorting, or fallbacks	Rebuild the `provider` object using AnonRouter’s supported keys, values, and provider slugs. Unknown keys in that object are rejected. See provider routing.
Cross-model fallbacks	Keep the primary ID in `model`. Compatibility mode accepts up to four exact fallback IDs in `models`. See model fallbacks.
`provider.data_collection`	Review each route’s privacy label and choose a `provider.minimum_privacy` floor if needed. A privacy tier does not translate directly from the old flag.
Model suffixes	Review the supported suffixes. Free routes require account opt-in and explicit consent to Anonymous routing. See free models.
`max_tokens`	Use `max_completion_tokens` or keep `max_tokens`. If you send both, their values must match.
Tools, structured output, or image inputs	Check the selected route’s capabilities. Use `tools` instead of legacy `functions`. JSON Schema enforcement depends on the provider. See compatibility restrictions.
Reasoning options	Use supported effort values for the chosen model. See reasoning controls for valid fields and disable behavior.
Plugins or web search	Use the AnonRouter web search tool where supported. Unsupported top-level chat fields can be silently dropped.

Model fallbacks

Compatibility mode tries the primary model first. It can try the next ID in models after an eligible failure before any output, such as no available route or a rate limit. It does not switch models after output starts or retry invalid credentials with a different model. Set a privacy floor if every candidate must meet the same minimum. See provider routing for fallback between providers of one model.

4. Test the rest of your app

If you stream, test stream completion and interruption. Treat a stream that ends without [DONE] as incomplete. Review ambiguous failures before retrying to avoid a second charge.
If you use embeddings, choose a callable embedding model and test client.embeddings.create(...) separately.
Image generation and speech require the Studio flow. The static-key Chat Completions setup does not enable those endpoints.
The Responses API and single-model retrieval are unavailable on this endpoint. Use Chat Completions and client.models.list().
Keep SDK error handling, but check the HTTP status and AnonRouter’s error.code before branching on a provider-specific error.

See the API reference for endpoint details and compatibility mode for authentication and request restrictions.