This project is a local API shim that allows Codex Desktop to use self-hosted 'bring your own key' (BYOK) models and optionally passthrough requests to ChatGPT (GPT-3.5, not 5.5 as listed). It effectively enables Codex Desktop users to leverage alternative LLMs, giving them more control over their AI coding environment.
A starter prompt for Claude Code, what you'll need, and how to reach them.
You are an expert full-stack developer. Create a local API shim that allows Codex Desktop to connect to various LLMs, including self-hosted 'bring your own key' (BYOK) models and an optional passthrough to OpenAI's GPT-3.5 API. The project should be built using Next.js 16 App Router, React 19, Tailwind v4 for any minimal UI, and Node.js for the backend API. For local model integration, consider a simple proxy for common local inference servers. The MVP should focus on: 1) A Node.js backend that exposes an API endpoint mirroring the expected Codex Desktop API. 2) This backend should have a configurable option to forward requests to OpenAI's GPT-3.5 API using an API key. 3) A minimal client-side config page (Next.js) to input the OpenAI API key and potentially other BYOK model endpoint URLs. 4) Clear instructions on how to set up Codex Desktop to point to this local shim. Ensure the solution is robust, handles common API request/response formats, and provides basic error handling. Emphasize a clean, secure, and maintainable codebase. The build/verify gate for this MVP is successfully routing a simple code generation request from Codex Desktop through the shim to GPT-3.5 and receiving a valid response.
Reach out to developers using Codex Desktop who are looking for more control over their AI models and offer a robust, flexible shim for BYOK and multi-provider access, highlighting how it complements existing AI usage monitoring and evaluation tools.
Local Responses-API shim that exposes Factory BYOK models (and optional ChatGPT GPT-5.5 passthrough) to Codex Desktop.
Open an issue on the 0xSero/codex-shim GitHub repository to propose a community-driven fork or alternative implementation, or send a direct message via GitHub if their profile allows.
“I've built a more robust and extensible `codex-shim` in Next.js/Node.js with support for a wider range of BYOK models and an updated GPT-3.5 passthrough. Here's a demo; I believe this could benefit the community and potentially evolve into a more feature-rich tool if there's interest.”
Open the original ↗