Analysis updated 2026-05-18
Point Codex Desktop at DeepSeek or another OpenAI-format API instead of OpenAI
Manage API keys and models for Codex from a local web dashboard
Let a text-only model handle screenshots via the Vision Bridge component
Automate mouse, keyboard, and window control on macOS through the Computer Use engine
| aitabby/opencodex | jdevalk/specification.website | dataants-ai/cutscript | |
|---|---|---|---|
| Stars | 92 | 92 | 93 |
| Language | TypeScript | TypeScript | TypeScript |
| Setup difficulty | easy | easy | moderate |
| Complexity | 2/5 | 2/5 | 3/5 |
| Audience | developer | developer | general |
Figures from each repo's GitHub metadata at analysis time.
Requires macOS, Node.js 18+, and an existing Codex Desktop installation.
OpenCodex is a local server that runs on your Mac and acts as a bridge between Codex Desktop (an AI coding tool from OpenAI) and third-party AI APIs. By default, Codex Desktop only works with OpenAI's own services. OpenCodex patches the Codex configuration file automatically when it starts up, letting you point Codex at any API that follows the OpenAI format, such as DeepSeek or other providers. The project includes a web dashboard accessible at localhost:8765 in your browser. From there you can add API keys, set custom API endpoints, add or remove models, choose which models appear inside Codex, and restart Codex with one click. A live log stream shows what is happening in real time. There is also a one-click option to restore the original Codex configuration if needed. Two additional components extend what Codex can do on macOS. The first is a Computer Use engine that can control the mouse, keyboard, and windows using the system's native event API, and can take compressed screenshots. The second is a Vision Bridge for AI models that only handle text and cannot process images on their own. When Codex needs to work with a screenshot but the active model does not support images, Vision Bridge compresses the screenshot, sends it to a separate vision-capable model to produce a text description, and injects that description into the prompt instead. This allows text-only models to participate in tasks that would otherwise require image understanding. Installation requires macOS, Node.js version 18 or later, and an existing Codex Desktop installation. Setup is three commands to clone, install dependencies, and start the server. The dashboard opens automatically in the browser.
A local macOS server that lets Codex Desktop connect to third-party AI APIs instead of only OpenAI's own service.
Mainly TypeScript. The stack also includes Node.js, macOS.
The README does not state a license.
Setup difficulty is rated easy, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.