Analysis updated 2026-05-18
Point developer tools like Codex CLI or Cursor at a self-hosted API endpoint backed by a ChatGPT Plus subscription.
Pool several ChatGPT Plus accounts behind one gateway to spread out rate limits for a small team.
Track API usage and billing across multiple proxied accounts through a built-in gateway.
View call logs and generate images through the included dashboard and image tool.
| liyueyuan123/llm-api-gateway | autolearnmem/automem | billy-ellis/exr-imageio-poc | |
|---|---|---|---|
| Stars | 32 | 32 | 32 |
| Language | Python | Python | Python |
| Setup difficulty | hard | hard | moderate |
| Complexity | 4/5 | 5/5 | 3/5 |
| Audience | developer | researcher | researcher |
Figures from each repo's GitHub metadata at analysis time.
Requires a VPS, a domain, and one or more paid ChatGPT Plus accounts, plus periodic manual token renewal.
ChatGPT Plus is a monthly subscription that gives you access to OpenAI's models through a web interface. Many developer tools, like Codex CLI and Cursor, want to connect to OpenAI's API instead, which charges separately by the amount of text processed. This project bridges that gap: you run it on a cheap VPS (a rented server), and it turns your Plus subscription into a standard API endpoint that those tools can point at. The setup runs five services together. Two are reverse proxies that sit in front of different OpenAI entry points: one handles Codex CLI with automatic token renewal and supports the Responses API with prompt caching, while the other handles ChatGPT Plus text and image generation with a token that needs refreshing roughly every ten days. A third service is an API gateway called new-api that pools those two proxies, handles billing tracking, and provides user management. Two smaller utility services round out the stack: a web tool for generating images using multiple accounts in parallel, and a dashboard for viewing API call logs. The cost comparison in the README is the main selling point. A VPS costs about 99 yuan per year. Each Plus account costs about 100 yuan per month. With one account you cover light personal use, with two or three you can share across a small team and rotate when one hits rate limits. A commercial API reseller typically starts at $20 to $100 per month and charges per token on top. Deployment uses a Makefile. You clone the repo, copy the example environment file, fill in your VPS address, domain, and passwords, then run make setup once on the server. After that, make deploy syncs any changes and restarts the relevant containers. The project uses Docker for all five services and Nginx as the front-facing reverse proxy with automatic SSL certificates. The README is written in Chinese and includes security reminders: change all placeholder passwords before deploying, keep the environment file out of version control, and restrict the server firewall to only ports 22, 80, and 443.
A self-hosted Docker stack that turns a ChatGPT Plus subscription and a cheap VPS into an OpenAI-compatible API endpoint for developer tools.
Mainly Python. The stack also includes Python, Docker, Nginx.
MIT license: use freely for any purpose, including commercial use, as long as you keep the copyright notice.
Setup difficulty is rated hard, with roughly 1h+ to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.