what is openai-codex-claude-code-proxy fr?

This project lets you run Claude Code, Anthropic's coding assistant tool, using GPT models instead of Claude models underneath. It works by starting a small local proxy server that translates the requests Claude Code normally sends to Anthropic into requests that OpenAI or Azure OpenAI can understand, and translates the responses back so Claude Code does not notice the difference. It ships as two launcher scripts. One connects Claude Code to GPT-5.6 through an OpenAI or Codex subscription using a browser login, and the other connects it to a GPT-5.6 deployment hosted on Azure OpenAI using an API key. Both scripts start the local proxy automatically, keep your actual API credentials out of Claude Code itself, and use separate model names for each provider so a request cannot accidentally bill the wrong account. A notable feature is how it handles subagents, the smaller helper AI tasks Claude Code spins up for specific jobs. The main conversation runs on the largest model tier, while subagents can be routed to smaller, cheaper GPT-5.6 tiers depending on whether the task calls for careful review or quick exploration. This routing can be controlled per session or forced onto a single tier. Setting it up requires macOS or Linux, Claude Code itself installed, a separate open source tool called CLIProxyAPI, Python 3, and either an OpenAI or Codex subscription or an Azure OpenAI account with a deployed GPT model. Setup involves cloning the repository, filling in a configuration file with credentials and a generated local access token, and authenticating through a one-time browser login for the OpenAI OAuth path or by pasting Azure keys for the Azure path. The project also includes scripts to test that the connection and Azure's prompt caching feature are working correctly, along with security notes about keeping the local proxy off any public network.