Rewriting openai-api-server-via-codex in Go

I have been making good use of openai-api-server-via-codex, a translation proxy I built that exposes an OpenAI API-compatible interface through a Codex subscription. It received almost no attention when I first released it, but its GitHub star count has recently started to grow. It is still only around 30 stars as I write this, but I have also received several pull requests.

I originally implemented it in Python without giving the language much thought. There was no particular reason for this tool to be written in Python, however, and both its startup time and memory use had room for improvement. I therefore had an AI coding agent help rewrite it in Go.

The new version starts almost instantly and uses less than 10 MB of memory. That makes it light enough to leave running all the time without thinking about it. To preserve compatibility with the existing workflow, it still starts through uvx, which now launches the Go binary behind the scenes.

---

Perhaps I have simply been lucky, but every pull request I have received has been carefully prepared. They did not feel like careless AI-generated patches. The authors clearly took the time to make the changes easy to review and merge, from keeping the scope focused to providing useful explanations. I was pleasantly surprised.

Coding agents themselves have improved, but people also seem to be getting better at using them to produce changes that fit naturally into a collaborative development workflow. I appreciate having fewer small details that require human attention.