Run your loops on Codex

Two additions this release. Codex joins Claude Code and OpenCode as a supported harness, and you can now pay for runs with the ChatGPT subscription you already have.

Codex is now a supported harness

Trevize can now run your agent loops on Codex, OpenAI’s coding agent. Pick it for a single session or set it as the harness in a blueprint, and it plugs into everything you already use: signal-triggered loops, live previews, multiplayer steering, and forked sessions to try approaches in parallel. Runs behave the same on every harness. They stop at a pull request, and a human merges.

If your team already works with Codex locally, your runs in Trevize will feel familiar. Same agent, same style of output, now running around the clock on your issues, errors, review comments, and failing CI.

Use your ChatGPT subscription

You no longer need a separate OpenAI API key to run Codex or OpenCode. Connect your ChatGPT subscription once and both harnesses can use it:

npx trevize@latest connect openai

The command walks you through signing in with your OpenAI account. After that, runs on Codex and OpenCode draw from your subscription instead of metered API billing. If you’d rather keep using an API key, that still works exactly as before.