lucasmantou/codex-proxy

Codex Proxy is a small local server that acts as a translator between OpenAI's Codex desktop application and alternative, cheaper AI models. Codex is a desktop coding assistant from OpenAI, but it is designed to only communicate with OpenAI's API, which can be expensive. This proxy runs on your own computer and intercepts Codex's requests, translating them into a format that other AI providers understand, then translates the response back so Codex receives what it expects. The technical problem it solves is that Codex uses a newer API format called the Responses API, while most alternative AI providers use an older format called the Chat Completions API. These two formats speak different "languages," so you cannot simply swap the API endpoint URL. The proxy handles this translation automatically, including details like streaming responses (where text arrives in chunks as it is generated) and reasoning chain data from models that show their step-by-step thinking. By routing through models with lower pricing, the proxy's author claims you can reduce costs by 30 to 50 times compared to using Codex with OpenAI's own models. The proxy adds only about 50 to 100 milliseconds of extra delay per request since it is only doing format translation. You run the proxy as a background Python process on your local machine, then configure Codex to point at it instead of directly at OpenAI. The proxy is written in Python and licensed under MIT.