A free GPT-3.5 API built on OpenAI’s own web UI

What it does FreeGPT35 is a thin compatibility layer that parks itself in front of OpenAI’s free, login-free ChatGPT web interface and serves up a standard /v1/chat/completions endpoint. Any client that speaks the OpenAI API format—Bob, NextChat, LobeChat, or a raw curl—can point at it and authenticate with any random string as a Bearer token. It turns a browser-based chat session into something your scripts can talk to.

The interesting bit This is glue code in the truest sense: it bolts an API shape onto a web UI that was never meant to have one. The author is admirably upfront about the brittleness, noting that OpenAI’s frequent updates keep breaking the original approach and steering users toward a DuckDuckGo-based successor for newer GPT-3.5-Turbo-0125 access.

Key highlights

• Drop-in replacement for the OpenAI chat-completions endpoint.
• No real API key required; any placeholder string works as authorization.
• Packaged as a lightweight Node service with Docker and Docker Compose support.
• Includes ready-made Nginx reverse-proxy and load-balancer configurations.
• Works with third-party clients like OpenCat, Bob, ChatGPT-Next-Web, and Lobe-Chat.

Caveats

• The README explicitly warns that OpenAI’s updates have already disrupted the project, and the author now recommends a separate DuckDuckGo-based repo for continued access.
• The sources do not clarify how the underlying web scraping works, what rate limits apply, or how long a session stays valid.

Verdict A decent weekend project for developers who want a zero-cost API shim for personal tooling, but anyone shipping production code should treat it as a temporary hack rather than infrastructure.