Run Codex and Claude with any model including GLM 5.2. No settings file headaches.

https://preview.redd.it/d3r9vk2pyg9h1.png?width=2634&format=png&auto=webp&s=a7d921315936e1d0a6d5823dbbaaec0ce1f13fd7 I shared relay-ai here last week when I first launched the CLI. Since then, we've shipped a few updates to solve some of the most annoying Codex Desktop limitations and add new provider support. If you want to run Codex Desktop or Claude Code using your own API keys, xAI/OpenAI OAuth subscriptions, Gemini, or local models: I built this tool to handle the entire routing layer. You don't have to edit settings files or deal with conflicting env vars. Here are the exciting features we just added: > SuperGrok & ChatGPT Plus OAuth: You can now run SuperGrok and ChatGPT Plus OAuth simultaneously alongside standard API keys. The sign-in flow automatically opens your default browser. > Stop Codex Desktop background crashes: We updated the proxy to catch Codex Desktop's background polls for hardcoded OpenAI model IDs. The proxy now routes these calls to your active model so your sessions don't crash. > Context overflow safety: We write context windows and compact limits to your config. This lets Codex Desktop trigger auto-compaction before hitting the hard limits, which keeps long sessions alive. > Unified OpenAI endpoint: You can connect standard OpenAI clients to any model in your registry using our bidirectional translation adapter. How to get started: npm install -g u/jacobbd/relay-ai npm install -g /relay-ai relay-ai providers add # to add your API/oAuth providers and models relay-ai codex-app # or relay-ai codex / claude-app / claude We put the source, documentation, and a full side-by-side walkthrough video on the GitHub page: https://github.com/jacob-bd/relay-ai If you notice any issues, please submit a GitHub Issue.