A free GPT-3.5 proxy that runs on your own metal

What it does

Aurora is a Go server that sits between you and ChatGPT’s web interface, exposing a standard /v1/chat/completions endpoint. You self-host it, point your OpenAI client at your own IP, and get GPT-3.5 responses without an API key. It also bundles a minimal web UI at /web for browser use.

The interesting bit

The project leans on geographic luck: it only works from IP regions where ChatGPT allows free, no-login access. The README bluntly notes this limitation and suggests swapping BASE_URL to bypass it — a cat-and-mouse admission you don’t often see in “free GPT” tools.

Key highlights

• OpenAI API-compatible endpoint (gpt-3.5-turbo, streaming supported)
• Bundled web interface for non-technical users
• Self-hostable via single binary or Docker; TLS and proxy pool support via env vars
• No API key or OpenAI billing account needed (where free web access is available)
• Derived from ChatGPT-to-API; MIT licensed

Caveats

• Geographic restriction is real and unexplained in detail; “custom Baseurl” workaround is vague
• No mention of rate limits, token costs, or how session rotation is handled
• README is minimal; English version exists but content depth is unclear

Verdict

Worth a look if you need a self-hosted GPT-3.5 proxy for a region with free ChatGPT access and want to avoid API billing. Skip it if you need reliability guarantees, GPT-4, or operate from a restricted region without a working bypass.