lanqian528/chat2api

Chat2API is a Python service that sits between your code and ChatGPT's website. It translates calls made in the standard format used by the official OpenAI API into requests against the ChatGPT web interface, then returns the responses in that same standard format. The practical effect is that any program or tool already built to talk to the OpenAI API can be pointed at this service instead and use ChatGPT's website under the hood. For GPT-3.5, no login is required at all. For GPT-4, GPT-4o, and the reasoning models (O1, O3-mini, and their variants), you supply an access token from your own ChatGPT account. The README notes you can get this token by logging into chatgpt.com and visiting a specific session URL. You can add multiple tokens and the service will rotate through them automatically, retrying with the next token if one fails. Deployment options include running the Python script directly, using a Docker container, or using Docker Compose. A one-click deploy option for the Zeabur hosting platform is also provided. The service listens on port 5005 by default and exposes the standard /v1/chat/completions endpoint. A gateway mode can be enabled that proxies the actual ChatGPT web interface, allowing users to log in and use it through your domain. The README includes a table of environment variables for configuring things like authentication keys, proxy addresses, whether to save conversation history, and how aggressively to retry failed requests. The README is written primarily in Chinese, with configuration tables and feature lists making up most of the content. The project is licensed under MIT.