Claude Code, with a window
The Claude Code GUI that runs Codex too
When the first-party app is the right answer
Claude Code Desktop was rebuilt around parallel sessions in April 2026. If you run Claude and only Claude, on one Mac, and you mostly want to see a diff without squinting at a terminal — use it. It is free, it is made by the people who make the model, and nothing here beats “already installed”.
Foreman earns its place on a narrower claim: that the job stops being “talk to an agent” and starts being supervising several, and that the first-party apps are each locked to their own vendor by design.
The three things a second window actually buys you
Claude Code and Codex, one surface
Both CLIs run side by side — one permission model, one project memory, one cost view across both. You route each task to whichever agent is better at it instead of keeping two windows, two habits and two bills. Neither vendor’s own app will ever run the other’s agent.
Sessions that message each other
A session can hand work to another session by name — including ones you started yourself in a terminal outside Foreman. The agent that just finished the migration can ask the agent that owns the tests to check it, without you being the messenger between two windows.
Fan-out that keeps the attempts it loses
Race the same task across up to four agents, each in its own isolated git worktree, varying model or effort per lane. Compare the finished diffs side by side and keep the winner — and the attempts you didn’t keep are folded away rather than deleted, so “actually, the second one was right” is a click and not a rerun.
What the window itself gives you
The rest is the ordinary work of watching agents, done properly rather than novelly:
- A canvas instead of a list — every project a box, every session a card with live status, git state and cost. A glance answers “what is everyone doing?”
- Approvals in the chat, not in a buried terminal prompt: per-tool permission cards, plan review before a single file changes, and question cards with one-click answers.
- Real git — line-by-line diff review, comment on a line and send it back to the agent, merge conflicts resolved in the app, branches and the commit graph.
- Live cost per session, per project and per model, across both agents and every machine — one number rather than two dashboards that don’t add up.
- Steer a running session from your phone over an end-to-end encrypted relay, while the work keeps running on your Mac.
- Your other Macs and build servers join the same canvas, by account pairing or plain SSH.
What it costs, and what it doesn’t include
Free on one machine, forever — not a trial. Pro is $19/month and adds every machine you own, unlimited concurrent sessions, and the fleet view.
Model usage is not included and never will be. Foreman drives the claude and codex binaries you already have, signed in to your own accounts, on your own rate limits. It is not an API wrapper and it never lifts your subscription token into a client of its own.
Where it won’t suit you
macOS on Apple Silicon only — no Windows or Linux build, and none promised. It is a supervision layer, not an editor: you will still have your IDE open for the code you write by hand. And if you only ever run one agent at a time, most of what is above is weight you won’t use.
Free on macOS. Your agents, your keys, your machine.
Download for macOSComparisons on this page are accurate as of August 2026 — check each product’s own site for its current state.