lidge-jun/opencodex

opencodex is a local proxy that lets you use any large language model, including Claude, Gemini, Grok, DeepSeek, GLM, Kimi, Qwen, or a local Ollama model, with OpenAI's Codex CLI, app, and SDK, as well as with Claude Code. Normally these coding tools only speak to one provider's API, so opencodex sits in between and translates Codex's own request format into whatever format the chosen provider expects, in both directions, including streaming responses, tool calls, reasoning tokens, and images. Setup is a single global npm install, which also bundles the Bun runtime automatically so nothing else needs installing separately. After that, an interactive setup command writes a config file and wires opencodex into Codex, and a start command launches the proxy that Codex then talks to through a local address on your machine. It runs on macOS, Linux, and Windows, and can install itself as a background service on each so it starts automatically. The project ships with a web dashboard where a user picks from more than 40 built in providers, or types in a custom endpoint, pastes an API key or logs in through OAuth for a few of them, and models are then discovered automatically from that provider's own model list. New providers work immediately without restarting anything. Once configured, a model can be chosen per request using a provider slash model naming pattern, so a single Codex session can be pointed at Claude, Gemini, a local Ollama model, or many others just by changing that name. If no provider is specified, opencodex guesses from the model name itself. For teams or heavy users relying on ChatGPT accounts for Codex authentication, opencodex can also manage a pool of multiple accounts, track each one's usage quota, and automatically route new sessions to whichever account has the most quota left, while keeping an existing conversation pinned to the account it started with so a long running session does not switch accounts partway through.