Migrating to Inferray
Moving to Inferray is mostly a configuration change. Your SDK, prompts and request shapes stay the same; the base URL, the key and the model IDs change.
- From OpenAI or Anthropic — keep the official SDKs, and gain every other lab's models on the same key
- From OpenRouter — same OpenAI-compatible format, model IDs without the provider prefix
- From the Vercel AI Gateway — repoint the AI SDK's gateway provider or its OpenAI-compatible endpoint
A safe rollout
- Create a key in a separate project, such as
migration-test, so test traffic is easy to find. - Change the base URL and key in one environment, and map your model IDs to the catalogue.
- Run your test suite and compare outputs, latency and cost in the logs.
- Move production over, one service at a time.
Turn on auto top-up before production traffic arrives, so a low balance never stops requests.
Keep the old configuration behind an environment variable until you're confident: switching back is the same two settings.