lidge-jun/opencodex is a local proxy that makes it possible to use multiple model providers together in Codex and Claude Code. It converts OpenAI Responses API requests into each provider's format, so other models such as Claude or Gemini can be used through the Codex CLI, the app, and SDK paths. The GitHub repository record lists the latest release as v2.89.0. It is written in TypeScript and follows the MIT license.
A proxy is an intermediary program that relays requests on your behalf. This project converts incoming OpenAI Responses API requests from Codex, Claude Code, Claude Desktop, and Grok Build into each provider's format. The README explains that conversion in the reverse direction is supported as well.
It supports streaming, tool calls, reasoning tokens, and image passing. Providers and models are configured in a dashboard where accounts are managed, and automatic routing based on Codex account usage is also possible. It offers both a desktop app and a command-line way of running it. The repository shows 17663 stars, and the repository creation time is 2026-06-18T23:03:18Z.
The changelog for this release is not included in the provided information. Only the version number is known, so what v2.89.0 changed cannot be confirmed. Beyond the feature scope described in the README, there is no comparison data on per-provider compatibility or conversion latency either. The novelty score is 7.0 and the momentum score is 8.6, while the technical and quality scores are 8.0 each. The structure that bundles multiple tools into a single provider specification is notable. Whether to adopt it is better decided after checking the per-provider compatibility range and the version-by-version changes directly.