What it does

This is a full-stack ChatGPT web client: Vue3 frontend, Express backend, Docker-ready. You type, it streams responses. The project is squarely aimed at developers in China who need workarounds for network blocks, with extensive proxy support (SOCKS, HTTPS, reverse proxies) and one-click deploys to Railway and Sealos.

The interesting bit

The README is unusually honest about its own trade-offs. It runs two unofficial API methods side by side and labels the free one “Relatively unreliable” and the paid official API “Relatively stupid.” That candor extends to security warnings: it explicitly tells users not to use public proxies, warns that reverse proxies expose access tokens to third parties, and even suggests changing the HTML title so scrapers don’t find your instance.

Key highlights

• Dual backend: official OpenAI API key or web access token through a reverse proxy (default: ai.fakeopen.com)
• Multi-session storage with context logic, data import/export, and message-to-image export
• Rate limiting (MAX_REQUEST_PER_HOUR), access control (AUTH_SECRET_KEY), and anti-crawler nginx rules included
• Docker Compose, Railway, and Sealos deployment templates with full environment variable tables
• Multilingual UI with theme switching

Caveats

• The access-token method depends on third-party reverse proxies that the README itself calls risky; the default proxy is someone else’s server
• Node version support is narrow (^16 || ^18 || ^19, with polyfill hacks for 14+)
• The project appears to be in maintenance mode: roadmap shows “More…” unchecked, and the model list still prominently features deprecated GPT-3.5 variants alongside GPT-4

Verdict

Worth a look if you need a self-hosted ChatGPT UI with flexible proxy options and don’t mind wiring up your own infrastructure. Skip it if you want official SDK guarantees, multi-user billing, or modern model management — this is a deployment scaffold, not a platform.