A cockpit for your fleet of Claude Code agents.
Run a dozen agents at once — each on its own repo, branch and worktree. Helmry shows who is working, who is blocked on you, and whose work is ready to ship. All of it on your machine.
Requires your own Claude subscription — billed separately by Anthropic.
Not eight terminals. A fleet.
One agent in a terminal is a tool. Eight of them in eight terminals is a mess — you spend the day polling windows to find out who needs you. Helmry gives the fleet a shape: one rail, one status per agent, one place to review.
Statuses that mean actions
The rail doesn't say what an agent feels — it says what you should do next. A question and a blocked command are different moves, so they get different colours.
Review like a pull request
Work lands as a diff, per worktree, with one commit-and-push flow. You read code, not chat logs.
One worktree per agent
Own branch, own dev server, own tree — five agents build and rewrite at once, and your checkout is never touched.
It has to pass before you see it
Point Helmry at your gates. A red one goes back to Claude with the output attached, and it fixes its own work.
A whole board at once
Pin a Jira ticket to an agent and it knows what it is working towards. No cap on how many run in parallel.
A real terminal underneath
Every worktree keeps a live shell, its ports and its environment. When you want the raw machine, it is one tab away.
A fleet is navigated, not clicked.
Jump to any agent, walk a diff file by file, queue a reply, stop a run — without reaching for the mouse. ⌘ is Ctrl on Windows and Linux.
One wire out. Everything else stays here.
Helmry is a desktop app, not a service. The case for it is built on what is not there — no backend of ours, no telemetry, and none of your credentials on anybody else's server. The sign-in it asks for is created and kept on your own machine.
The desktop app talks to your repositories and your terminal. There is no Helmry backend to route through — and none to breach.
Your own claude process, authenticated as you. Anthropic bills you directly; we never see the traffic.
No analytics, no crash uploads, no usage pings in the app — and this page loads zero third-party resources, fonts included.
One installer, then an account you create on the machine itself — it lives in ~/.helmry, not in a cloud. Helmry finds the Claude CLI you already use.
Pick a repo and an environment — local or WSL. Tick “isolated worktree” and the agent gets its own branch and checkout.
Give each one a task or a Jira ticket. Come back when the rail says a diff is ready.
One plan. First month free.
Pro is $20 per person per month — every feature, unlimited agents and repositories. New subscriptions start with a 30-day free trial: cancel during it and nothing is charged. And billing never interrupts work: an agent that is already running always finishes its turn.
How billing works, in writing →Put your fleet on one screen.
Download, point it at a repo, start an agent. The first diff is minutes away.
Requires your own Claude subscription — billed separately by Anthropic. Agents edit files and run commands on your machine.