Proxy for OpenAI Codex and Claude Code, use any LLM with those apps

**OpenCodex** is a proxy that sits in front of **OpenAI Codex** and **Claude Code** so those apps can talk to models other than the ones they ship with. The project lives at github.com/lidge-jun/opencodex and showed up on Hacker News with 5 points and no comments yet. The pitch is simple: keep the Codex or Claude Code client, swap the backend LLM.

A proxy in this setup intercepts the traffic those coding agents already send, then routes or reshapes it toward another model endpoint. That keeps the app’s UI, workflows, and agent loop intact while the model behind the requests changes. Builders can point the same local tools at whatever provider or self-hosted stack they already run, instead of waiting for first-party support for every model.

For engineers, the value is control over cost, latency, and data path without giving up tools they already know. Teams locked into Codex or Claude Code for agentic coding can try cheaper or open models, keep traffic on private infra, or A/B models against the same prompts and harness. The agent UX stays fixed; only the model layer moves.

The market is crowded with coding agents that each assume a preferred model stack. Proxies like this attack the lock-in problem from the side: they treat Codex and Claude Code as fixed front ends and open the back end. That sits next to “bring your own key” and multi-provider SDK patterns, but targets the agent apps themselves rather than raw chat APIs. Early HN signal is thin—5 points, zero comments—so this is still discovery stage, not proven adoption.

What to watch next is whether the proxy covers the full request surface those apps use (streaming, tools, multi-step agent calls) and which model backends actually work day to day. If it stays thin or breaks on agent features, teams will bounce. If it holds under real coding sessions, it becomes a practical way to run any LLM through Codex and Claude Code without rewriting the agent.