opencodex: a universal provider proxy that enables any LLM to work with OpenAI Codex and Claude Code

What it solves

OpenAI's Codex CLI, App, and SDK are typically locked to OpenAI models. This project breaks that limitation by allowing users to use any LLM provider (like Anthropic, Google Gemini, Grok, or local models via Ollama) within the Codex ecosystem and Claude Code without waiting for official support.

How it works

It acts as a lightweight local proxy that intercepts Codex's Responses API and translates it into the protocol used by your chosen provider. It supports streaming, tool calls, reasoning tokens, and images in both directions. It also includes a management dashboard to handle multiple ChatGPT account pools, automatically routing new sessions to accounts with the lowest usage to manage quotas.

Who it’s for

• Developers using the Codex CLI, App, or SDK who want to use non-OpenAI models.
• Users of Claude Code who want to route requests to various LLM providers.
• Power users managing multiple AI accounts to maximize usage quotas.

Highlights

• Universal Compatibility: Supports 40+ providers including Anthropic, Gemini, xAI, DeepSeek, and local providers like Ollama or vLLM.
• Claude Code Integration: Can launch Claude Code fully wired to the proxy, allowing model discovery for routed models.
• Account Pooling: Manages multiple ChatGPT/Codex accounts with affinity (keeping threads on the same account) and automatic failover for quota limits.
• Web Dashboard: Provides a GUI to add providers, manage OAuth logins, and view a live request log with token counts.
• Cross-Platform Support: Runs natively on macOS, Linux, and Windows as a background service or on-demand shim.