Most GenAI prototypes call an LLM API straight from a script or notebook. A real product needs something in between the client and the model provider. That layer should handle failures, protect the provider from too much traffic, and let you control cost.

This project is that layer, built with FastAPI. It sits in front of the AI endpoints and adds what you need before going live: clear error responses, rate limits, webhooks and consistent API behaviour.

Errors from the provider, timeouts and validation problems all come back in the same format, so clients always know what went wrong.

Rate limiting per key or per route stops extra requests before they reach the model, which saves money and protects the provider.

Webhooks let long jobs tell other systems when they finish, so the HTTP request does not have to stay open.

Auth hooks, request logs and routing rules let you run several models or environments behind one entry point.