Your ChatGPT account, repackaged as a standard coding API

What it does

Codex Proxy is a local translation layer that sits between your AI clients and OpenAI’s Codex Desktop backend. It converts the proprietary Codex Responses API into standard /v1/chat/completions, Anthropic /v1/messages, and Gemini formats, letting you point Cursor, Claude Code, Continue, or standard SDKs at your own instance. You authenticate with a regular ChatGPT account—free tier works—or route through third-party API providers if you prefer.

The interesting bit

Most protocol bridges just rewrite JSON bodies; this one treats impersonation as infrastructure. It ships a Rust native addon that locks reqwest and rustls versions to reproduce the exact TLS fingerprint of the real Codex Desktop client, then mirrors headers, cookies, and version strings down to the build_number. It even polls Codex Desktop update sources to keep its disguise current, which is either admirably thorough or slightly paranoid depending on your threat model.

Key highlights

• Front end speaks OpenAI, Anthropic, Gemini, and optional Ollama protocols, translating bidirectionally to Codex’s native Responses API
• Multi-account orchestration with plan-aware routing—free, Plus, Team, and Business accounts automatically map to their respective supported models
• Account rotation strategies including least_used, round_robin, and sticky sessions, plus JWT auto-refresh and ban detection
• Per-account proxy pools with health checks, latency measurement via ipify, and automatic exclusion of dead routes
• Web dashboard for account management, usage statistics, and bilingual UI

Caveats

• Non-commercial license only, which limits production or business use
• Explicitly maintained as a solo side project; the author declines custom support or feature-on-demand obligations

Verdict

Worth a look if you want Codex models inside your existing OpenAI- or Anthropic-speaking toolchain without swapping clients. Skip it if you need a commercial license or hand-holding support.