Claude Sub Proxy
Your Claude limits
Loading...
Long chats go much further before Codex compacts, but big contexts use your limits faster. After changing this, press Update on any SSH machine below so its Codex learns the new size.
When you hit a limit, the request waits (up to 6 hours) and carries on after the reset, instead of failing. Off: your agent gets a "you hit your limit, resets at ..." error.
Codex on this computer not set up
Makes the ChatGPT desktop app's Codex use your Claude subscription, with every Claude model in its model picker. Writes to . Restart the ChatGPT app after.
Codex on an SSH machine checking
For Codex chats that run on another machine (the app's remote SSH mode). That machine reads its own config, so it needs its own setup. Goes over SSH and your tailnet. Set up and Undo restart Codex's engine there, so any chat running on it gets cut off.
Other devices on your tailnet
For Codex on another device: save the model list into that device's ~/.codex folder, then put this in ~/.codex/config.toml there. The first four lines must sit above any [section].
Any other agent
Point anything that speaks OpenAI or Anthropic at these. On this computer the key is optional.
claude-opus-5-5 claude-sonnet-5-5 claude-haiku-5-5