0xSero

0xSero / codex-shim

Public

Local Responses-API shim that exposes Factory BYOK models (and optional ChatGPT GPT-5.5 passthrough) to Codex Desktop.

108
11
69% credibility
Found May 22, 2026 at 108 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

Codex Shim is a local helper program that lets you use Codex Desktop with AI models that aren't officially supported. It reads your existing model settings, creates a private bridge on your computer, and makes your custom models appear in Codex's model picker alongside the built-in ones. When you select a model, the bridge forwards your requests to the right AI service and translates the responses so everything works smoothly. You can use models from OpenAI, Anthropic, DeepSeek, and other providers using your own API keys, all without changing Codex's core settings.

How It Works

1
πŸ’­ You want more AI models in Codex

You've been using Codex Desktop for coding help, but you wish you could use your favorite AI models instead of just the ones OpenAI officially supports.

2
πŸ”§ You set up the shim

You install a small helper program on your computer that acts like a bridge, reading your existing model settings and making them available to Codex.

3
✨ Your models appear in Codex

Suddenly, all your custom AI models show up in Codex's model picker just like the built-in ones, ready to use.

4
You choose how to launch
🏠
Set it up permanently

Your models stay in Codex every time you open it, working alongside your existing setup.

⚑
Try it just once

You launch Codex with your models for today only, keeping your original settings untouched.

5
πŸ€– Codex routes your requests

When you pick a model and ask a question, Codex sends it through your local bridge, which forwards it to the right AI service.

6
πŸ”„ The bridge translates responses

Different AI services speak different languages, but your bridge converts everything so Codex understands the replies.

πŸŽ‰ You get help from any model

You can switch between Claude, DeepSeek, Gemini, or any model you've set upβ€”all from within Codex's familiar interface.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 108 to 108 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is codex-shim?

codex-shim is a local Python proxy server that tricks Codex Desktop into supporting any model you've configured in Factory's BYOK system. It listens locally, translates Codex's Requests API calls into whatever shape your upstream provider expects (OpenAI chat completions, Anthropic Messages, or generic endpoints), and streams the response back. It also includes an optional passthrough that routes GPT-5.5 through your existing ChatGPT subscription instead ofζΆˆθ€—δ½ ηš„ API quota.

Why is it gaining traction?

Codex Desktop ships with a server-side allowlist that gates which models appear in the picker. If you've paid for API access to Claude, DeepSeek, Gemini, or anything else, you can't use them as first-class models without this kind of workaround. The shim bridges that gap by generating a catalog from your existing `~/.factory/settings.json` and injecting it at launch. You point Codex at it, generate the catalog, and your custom models appear in the picker alongside the defaults.

Who should use this?

Developers who pay for multiple AI providers and want to consolidate access through Codex without juggling separate tools. It's especially useful if you have Factory.ai configured with BYOK keys and want to use those models as the active Codex engine rather than switching to the web interface. macOS users will need to run the ASAR patch for the picker to show the catalog entries; Linux and Windows users can skip that step entirely.

Verdict

This fills a real gap for power users stuck behind Codex's model restrictions, and the implementation is thoughtful (keys never leave your settings file, config is opt-in per launch). At 108 stars and v0.1.0, it's a young project with minimal test coverage, so expect to debug edge cases. If you need this right now and don't mind the rough edges, it's the only game in town for this specific workflow.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.