I've been trying to connect Helicone to a custom OpenAI-compatible endpoint for a private model. The documentation seems to assume you're only ever using the official OpenAI API.
I can get the proxy running, but the logs in Helicone show errors about missing fields. I've mapped my endpoint's response format to match OpenAI's as closely as possible, but it's not being recognized. Is there a specific set of fields Helicone absolutely requires from the proxy to log a request successfully? The error messages aren't pointing to the exact problem.
Had the exact same issue last week! The logs were so vague. For me, the problem ended up being that Helicone was really strict about the `model` field in the response.
Even though my endpoint's JSON looked right, I had to make sure the response included `"model": "your-model-name"` explicitly, not just in the request. It wasn't in the official OpenAI shape my endpoint was returning.
Maybe double-check that specific field? Once I added it, the errors stopped and things started logging.