unofficial integration

claudex — Claude Code on GPT-5.6

Claude Code stays your harness; GPT-5.6 becomes the model, through a localhost-only CLIProxyAPI sidecar signed in with your own ChatGPT/Codex subscription. One guided setup, an isolated session home, and your normal claude stays untouched.

1Watch it work

real machine · real release (ovm 0.1.0) · unedited commands · 57s

What you're seeing: the real installer, ovm claudex doctor's health checks, then ccxy launching Claude Code against the proxy — and the model answering, in its own words, "I'm GPT-5.6 Sol, running in Anthropic's Claude Code CLI harness."

2Set it up

curl -fsSL https://ovm.sh/install | sh
ovm claudex setup     # guided: proxy, sign-in, isolated home
ccxy                  # Claude Code, GPT-5.6 Sol, yolo

Setup checks before it does anything: it installs Claude Code if missing, stages a checksummed sidecar, connects your Codex account via browser OAuth, and seeds an isolated Claude home under ~/.ovm/claudex/ — your existing history, settings, and login are never mixed. Launch variants: claudex / ccx, ccxy (yolo), ccxf (fast tier), ccxyf.

3The honest part

This recipe was shared publicly by OpenAI's Codex lead. It is an unofficial integration — no endorsement by Anthropic or OpenAI is claimed or implied, and you use it at your own risk under your own accounts' terms. The sidecar binds to localhost only, with a key generated per machine; requests leave through your ChatGPT/Codex subscription, never a shared service.