Cmochance/codex-app-transfer

What it solves

Codex App Transfer is a lightweight desktop configuration and forwarding tool designed for the OpenAI Codex APP. It allows users to seamlessly use third-party inference services (via chat/completions protocols) within the Codex APP by acting as a local gateway that translates Codex's internal Responses API requests into formats compatible with various AI providers.

How it works

The tool starts a local gateway (typically at 127.0.0.1:18080) that intercepts HTTP requests from the Codex APP. It translates these requests into protocols such as Chat Completions, Gemini Native, Anthropic Messages, or Grok Web, and forwards them to the chosen provider. It includes a desktop UI for managing providers, model mapping, and logs. Additionally, it uses the Chromium DevTools Protocol (CDP) to inject custom themes, usage statistics, and draft stashing features directly into the Codex Desktop interface.

Who it’s for

Users of the OpenAI Codex APP who want to use models from other providers (like DeepSeek, Kimi, Gemini, or Anthropic) while maintaining the Codex APP's native features and interface.

Highlights

• Multi-Provider Support: Maps OpenAI model names to real model IDs from various providers including DeepSeek, Kimi, GLM, and others.
• Protocol Translation: Handles complex interactions like tool-use loops, history playback, and reasoning content injection to align with the OpenAI Responses API.
• Patch Application Recovery: Includes a middleware layer that repairs malformed V4A patches generated by third-party models to ensure they are correctly rendered as diffs in Codex.
• UI Enhancements: Injects custom anime themes, real-time token usage statistics (including cache hit rates), and a draft "stash" system into the Codex Desktop client.
• Plugin & Account Management: Supports unlocking Codex Plugins via simulated or real ChatGPT accounts and provides a mirror of the OpenAI Codex plugin marketplace.
• Integrated Web Fetching: Provides a built-in web_fetch tool that can automatically scale from simple curl requests to headless Chrome browsing to bypass Cloudflare and render JS-heavy pages.
• Session Persistence: Implements two-layer persistence (LRU memory + SQLite) with content-addressable deduplication to save space and preserve conversation history.