The current implementation of the ext_proc filter assumes traffic goes to the /v1/chat/completions endpoint, and the translators return an error if the request goes to a different path.

It will be a common case for existing applications to be already consuming other OpenAI API endpoints (for example, the /v1/models to get the list of available models), and having requests to those endpoints fail when going through the AI gateway makes adoption harder.

It would be great to have a mechanism to configure the ext_proc to not fail on unknown paths and allow the request instead, so that existing applications can be proxied by the ai-gateway more easily