A musician doesn't own one instrument. They own the ones that suit different songs, and the studio doesn't care which one is plugged in. That's the idea behind this release. Frontier has always run Claude Code on your behalf, and it still does. Now it runs the Codex CLI with the same fluency: every run, every tool call, every terminal session you sync back into a workspace, every one of Frontier's own tools reachable from inside the conversation. Nothing about your workspaces, agents, or memory changes when you switch. Only the voice doing the work.
We didn't build this because one is better than the other. We built it because the question "which one?" has a different answer on different days, and a tool that makes you choose once, forever, is a tool that's quietly narrowing what you can try. Pick a default for the whole app. Override it for one workspace. Keep a builder on one and a reviewer on the other, and see what each catches that the other doesn't. What becomes possible when the choice stops being a commitment?
Getting either one connected used to be a small chore spread across a few screens. It's one screen now. Every supported CLI shows up as a card with its logo; the ones installed on your machine are ready to tick, the ones that aren't sit greyed out and tell you where Frontier looked for them. Tick what you want wired up, choose which runs by default, and continue. Frontier handles the terminal-session hooks and the MCP registration for each one, and skips anything you'd already done. It's the "Set up your coding agents" step of every new context, and the same screen lives in Settings so you can come back to it the day you install something new.
Codex asks one thing of you that Claude doesn't: it wants to see a new hook once before it trusts it. Open any Codex session after setup, run its /hooks command, and approve Frontier's. We'd rather tell you here than let you find out by wondering why a session never appeared.
While we were in the guided setup, we gave it a Back button, which it had been missing longer than we'd like to admit. And the question about handing work between agents now explains why it's worth the setup: a single agent has to hold the whole task in its head at once, and on bigger jobs that head fills up. Splitting the work into stages, each carrying only what it needs, is usually the more accurate path. That's not a feature. It's a way of working we'd like more people to discover.
Every generation gets new instruments. This release just makes sure you're not limited to one of them. Full details on the Release Notes page.