explaingit

solrizon-ai/open-codex

★ 26TypeScriptAudience · developerComplexity · 3/5ActiveSetup · moderate

TLDR

macOS desktop app that wraps the OpenAI Codex CLI and Anthropic Claude Code CLI in a three-pane GUI with chat, file diffs, terminal, and project sidebar.

Mindmap

mindmap
  root((open-codex))
    Inputs
      Codex CLI
      Claude Code CLI
      Project folder
      Env vars
    Outputs
      File diffs
      Chat history
      Terminal output
    Use Cases
      Drive Codex with a GUI
      Drive Claude Code with a GUI
      Compare two backends
      Review diffs visually
    Tech Stack
      TypeScript
      Node 22
      pnpm
      Electron

Things people build with this

USE CASE 1

Drive the OpenAI Codex CLI from a graphical workspace instead of a bare terminal

USE CASE 2

Drive the Claude Code CLI with a sidebar of past projects and a built-in diff view

USE CASE 3

Switch between Codex and Claude Code backends on the same repo without mixing their settings

USE CASE 4

Side-load a custom Codex or Claude binary for testing via the documented environment variables

Tech stack

TypeScriptNode.jspnpmElectronmacOS

Getting it running

Difficulty · moderate Time to first run · 1h+

Needs macOS, Node 22+, pnpm, and a working installed-and-signed-in Codex or Claude Code CLI before the app has anything to drive.

In plain English

Codex Reverse is a desktop application for macOS that wraps coding agent command-line tools in a graphical interface. The two tools it currently supports are OpenAI's Codex CLI and Anthropic's Claude Code CLI. The app does not ship its own AI model or account: users install and sign in to those CLIs first, and the desktop client then drives them as background processes. The interface is a three-pane workspace with chat on one side, a file diff view on the other, and a terminal below. A sidebar shows projects and an active conversation history. Slash commands from the underlying CLI are surfaced through a command palette in both modes, with built-in commands like /model and /permissions in Codex mode and Claude Code's own built-ins plus any installed skills or plugins in Claude Code mode. The two backends are kept separate so Codex-only settings, such as reasoning effort or permission labels, do not bleed into Claude Code sessions. Running the app requires macOS, Node.js 22 or newer, and the pnpm package manager. After cloning the repository and running pnpm install, pnpm dev launches the app. Environment variables let the user point at a specific Codex or Claude binary, use a separate config home directory for testing without disturbing the normal CLI setup, pre-open a project, or select which backend to start in. For Codex mode the app spawns codex app-server in the background. For Claude Code mode it reads the user's existing Claude Code project history from ~/.claude and starts new claude CLI sessions inside the selected project directory. Git and diff views stay scoped to that project. The repository includes a troubleshooting section for common startup problems and a packaging command for building distributable artifacts.

Copy-paste prompts

Prompt 1
Clone open-codex, run pnpm install and pnpm dev, then open my current project in Claude Code mode
Prompt 2
Show me the environment variables this app reads to point at a custom claude binary and a sandboxed config home
Prompt 3
Add a third backend (say, Aider) to open-codex following the same pattern Codex and Claude Code use
Prompt 4
Walk me through how Codex mode launches codex app-server as a background process and pipes its output into the chat pane
Prompt 5
Package open-codex into a distributable .dmg using the repo's packaging command and tell me what signing setup is needed
Open on GitHub → Explain another repo

Generated 2026-05-22 · Model: sonnet-4-6 · Verify against the repo before relying on details.