icebear0828/codex-proxy

What it solves

Codex Proxy is a lightweight local gateway that allows users to access Codex Desktop's programming models through standard AI protocols. It solves the problem of Codex Desktop's proprietary API being incompatible with popular AI clients like Cursor, Claude Code, and Continue, by translating those requests into a format the Codex backend can understand.

How it works

The project acts as a translation layer between AI clients and the Codex backend. It converts requests from OpenAI, Anthropic, Gemini, and Ollama formats into the Codex Responses API. It uses a Rust-based native TLS addon to ensure the TLS fingerprint matches a real Codex Desktop client to avoid detection. It also manages multiple ChatGPT accounts, providing intelligent rotation strategies (like round-robin or least-used) and automatic token renewal.

Who it’s for

Developers who use Codex Desktop or have a ChatGPT account and want to integrate these high-performance coding models into their preferred IDE extensions, CLI tools, or AI agents.

Highlights

• Multi-Protocol Support: Compatible with OpenAI, Anthropic, Gemini, and Ollama protocols, supporting SSE streaming, function calling, and structured outputs.
• Anti-Detection: Uses Rust Native TLS and client profile presets to mimic real Codex Desktop fingerprints and headers.
• Account Management: Features OAuth PKCE login, multi-account rotation, and plan-based routing to automatically assign models based on account type (Free/Plus/Team/Business).
• Image Generation: Supports image generation via the image_generation tool for Plus accounts.
• Model Routing: Includes a web dashboard for managing API keys, account usage, and custom model aliases.