codex-shim

Why we included this project

Codex Desktop only shows models its server-side config allows, so if you want to use your own API keys with Anthropic, DeepSeek, Gemini, or a local proxy, this shim puts them in the normal picker without rebuilding the client. It runs a small Python/aiohttp server on loopback that speaks the OpenAI Responses API, and translates each request to the matching upstream while keeping the agent loop intact: function calls, tool outputs, reasoning blocks, and streaming SSE survive the translation instead of being flattened to text. An optional passthrough routes Codex's native traffic to your ChatGPT subscription's Codex model, and the auto-router can send each task to the cheapest model that can handle it. If you already run BYOK models or a local proxy in front of your tools, this is a practical way to keep Codex's native UX while moving model routing off the server.