Run your AI coding agents on an infinite canvas.

An agentic development environment (ADE) for Mac and Windows. Run Claude Code, Codex and Cursor on an infinite canvas.

v1.0.8 · Mac and Windows · free for 3 days · no account

The Starkomand canvas after Tidy, on the Spacewalk backdrop, every agent window on its CLI’s own screen: Nora, Owen and Lena in one handoff run ready for human review, Owen on Auto · Astra · xhigh and Lena on Auto · Fable · high; Marcus and Diana in one squad, issue #2 tucked behind Diana’s window for review; Iris beside her hand-drawn diagram; Felix above a preview of the orders page with the status filter kept on page 2.
A still from the app.

Powered by the subscriptions you already pay for.

Starkomand runs the agent CLIs you have installed, signed in as you, on this Mac or on your own servers over SSH. Auto gives each step to one of your plans and picks its model. It never resells tokens or asks for a key.

One agent is enoughNo Starkomand accountAny other CLI you configure

01 · Issue

Drop the issue on an agent.

Pull a GitHub issue onto the canvas and drop it on a window. The note stays pinned beside the work while the agent starts.

02 · Plan

See the system before the code.

Ask an agent to draw it. The diagram lands beside its terminal, hand-drawn and editable.

03 · Handoff

The right model for each step.

Hand off and Auto picks the model and effort, and says why. One model reviews what the other built. Each agent starts from a focused brief.

04 · Voice

Voice control for your coding agents.

Hold ⌥ and speak to Claude Code, Codex or Cursor. Each line goes to the agent you name, and the windows light up as they finish.

05 · Catch up

Flip a window. See where it stands.

Every header times its turn, and how long it has waited on you. Tap Space and every agent turns over: what it worked on, whether it waits on you, what it said last, and the files, servers and pull requests it left.

Close the lid. They keep working.

Your MacBook stays awake while agents work, screen dark, and sleeps when they finish.

Settings open on Lid closed, with Keep agents working switched on and the line Awake for 4 working agents, beside Lena’s Claude Code window reviewing on Auto · Fable · high.

Every dev server, by name.

Ports lists which agent left what running, flags leaks and stops what you forgot.

The Ports widget on the canvas: 2.64 GB across 3 servers on this Mac; pagination-demo on port 5173, started by Felix on branch fix/filter-on-page-change; marketing-site on port 3000 in amber, up 900 MB in 10 minutes at 2.4 GB; postgres on 5432; a Clean up button offering one server.

Preview beside the agent.

Open what it built in a browser window on the canvas.

Felix’s Codex window above a browser window on the canvas showing the orders page with the status filter kept on page 2.

One workspace per project.

Each keeps its crew, layout and backdrop.

The workspace menu open on Pagination fix, listing Pagination fix, API gateway and Studio website, over that workspace’s agents, diagram and preview.
  • Smart reading

    A small model on your Mac flags turns that stopped short and risky PRs. Optional.

  • SSH hosts

    Claude, Codex or a shell on a server you own, in a canvas window.

  • Announcement voices

    Hear “Marcus done” in an on-device Kokoro voice, or OpenAI or Grok.

  • File guard

    Parallel agents can’t edit or stage each other’s files.

  • Agent messages

    Agents ask each other, across providers.

  • Limit handoff

    Claude ran out? Continue in Codex in one click.

  • PR flight board

    Your PRs, related ones together, on any account.

  • Usage rings

    Runway per provider, with reset times.

  • Review loops

    Bounded rounds that check files before continuing.

  • Persistent agents

    Run one in Herdr and it outlives the app.

  • Hand gestures

    Pinch, pan, clap to tidy. Optional.

  • Backdrops

    Built-in scenes, a curated library, or your own.

Privacy

Stays on your Mac.

No Starkomand account. The canvas, local dictation, hand tracking and the optional decision model run on this Mac. The local server only listens here.

$ netstat -an -p tcp | grep 3778
tcp4  0  0  127.0.0.1.3778  *.*  LISTEN

$ curl -s http://127.0.0.1:3778/mcp
{"ok":false,"error":"agent authorization required"}

$ curl -si -H 'Upgrade: websocket' http://127.0.0.1:3778/
HTTP/1.1 403 Forbidden
Your providers
Agents run with your own sign-ins, as in any terminal. Nothing here sends your work to a model on its own: a work summary is written only when you ask.
Loopback only
The server listens on 127.0.0.1. Every connection carries a key kept in your home folder.
Heard and read on device
Local dictation and hand tracking handle audio and camera frames on your Mac. Smart reading’s decision model runs on its CPU. Cloud dictation is a separate choice.

A few counts of launches, trials and setup steps help us improve the first run. Your work is never included. See what is shared.

Rashid in a winter jacket in front of the Developer Timeline title card
A Developer Timeline product

Built by an engineer, for engineers.

I’m Rashid, a software engineer for eight years. I also run Developer Timeline, where developers go to laugh at the job they know too well. Starkomand is the workspace I wanted for my own agents.

Pay once. Keep your subscriptions.

Every feature free for 3 days. One license per Mac with a year of updates. No Starkomand credit meter, ever.

try it

Free trial

$0for 3 days

  • Every feature, every provider
  • Unlimited agents and workspaces
  • No account, no card
  • Keeps your canvas if you buy
Download free

one-time

Starkomand Pro

$49Standard price $169once

Launch pricing, set by seats sold

Checkout shows your price before you pay.

  • Unlimited agents, workspaces and Command Keys
  • Saved task briefs, criteria and handoff history
  • Auto model routing, handoffs, review loops and final QA
  • Lid-closed work, SSH hosts, Smart reading, Ports, diagrams, previews, GitHub tools and voice
  • A year of updates. Keep the version you have after.
  1. Founding $49First 10 users
    Checking availability…
  2. Early $89Next 20 users
    Checking availability…
  3. Launch $129Next 30 users
    Checking availability…
  4. Standard $169Everyone after
    No seat limit
Get Starkomand Pro

First come, first served: launch prices last until their seats are taken.

v1.0.8 · macOS on Apple silicon, Windows 10 and 11. Each provider bills you for its own plan and usage. No refunds, except where required by law.

FAQ

Setup

What is Starkomand?

Starkomand is an agentic development environment (ADE) for Mac and Windows. It orchestrates the coding agent CLIs you already pay for (Claude Code, Codex, Cursor, Gemini CLI and others) on one infinite canvas, with handoffs between agents, bounded review loops and optional voice control. It is a one-time purchase, not a subscription. What an ADE is explains the category, and how this one sits next to your editor.

Is Starkomand an ADE?

Yes. An ADE (agentic development environment) is an app built around AI coding agents rather than around a text editor. Most ADEs give each task its own git worktree and chat pane; Starkomand runs each agent’s real CLI in a terminal window on a canvas and adds handoffs, review loops and diagrams around it. It sits next to your editor rather than replacing it. The agent orchestration guide covers roles, briefs and file ownership.

Which computers does Starkomand support?

Macs with Apple silicon, and Windows 10 and 11 PCs, x64 or Arm64, from the Microsoft Store. On Windows, agents on your own servers (SSH hosts) are not available yet. There is no Linux or Intel Mac version for now.

Do I need several agents to get value?

No. Start with one agent you have installed. Ask it to explain the code, look into an issue or write a plan. Add handoffs and review loops later, when a second agent would help.

Which agents can I use?

Claude Code, Codex, Cursor, Grok, GLM and any other CLI agent you set up. Each one signs in the way it already does. Every agent works in the 3-day trial and with Pro. You keep paying the providers for their plans and usage.

Is voice required?

No. Type, drag and use keyboard shortcuts as usual. Voice and hand gestures are optional. Both come with the trial and with Pro.

Will it use a lot of memory or battery?

Not in everyday use. Starkomand stays quiet in the background. Hide the canvas and it draws less, while your agents keep working. You may notice it when you run many agents and previews at once, with a moving backdrop and local dictation on.

Agents and workflow

What does it add to running agents in terminal tabs?

Each task keeps its goal and what counts as done. You can hand a short brief to another model, and the task keeps the findings and checks. Review loops check for changed files before they go on. The issue, the agents and the preview stay together while you decide what comes next. See how it compares with Conductor, Maestri and other ADEs.

Can I run Claude Code and Codex together?

Yes. Run Claude Code and Codex side by side, plus Cursor and any other CLI agent you set up. Each one gets its own terminal window. You can open several windows of the same agent, too. Each signs in the way it already does.

Does an approved review mean the work is finished?

A review helps you decide. It does not decide for you. You can read the findings, the checks it reports and the revision it saw. Before it goes on, a review loop checks that the files still match that revision. If they changed, it stops. Ready for human review still leaves the final call to you.

How does Auto pick a model?

From the step and the work. Planning and reviews get your strongest model. A build gets one sized to the change, and anything touching auth, payments or data goes Thorough. With usage meters on, the next handoff skips a plan near its limit and goes to your other subscription. Each choice says why in one line, and one click changes it. No cloud model decides. With Smart reading on, a small model on your Mac checks whether a word like “payments” is the actual work or only a mention.

How do I know which agent needs me?

Look at the window headers. Each one shows what its agent is doing and for how long, turns amber with a timer when it waits on you, and says how long a finished turn took. Turn on Settings → Smart reading and the state gets more exact: “Want me to add tests?” reads finished, “Pick staging or prod.” reads waiting, and a turn that stopped short says “did not finish”. Grok, Cursor and other CLIs that report nothing get the same states from a read of their screen. An optional 1.1 GB decision model on your Mac answers what the rules can’t tell, and anything it decided says so.

Can an agent run on my own server?

Yes. Add the server in Settings → SSH hosts as user@host. The launch menu then lists Claude, Codex and a shell on it under On your servers. The agent runs on the server, inside tmux when it has tmux, so a dropped connection picks up the same session. Diagrams, previews, handoffs and the window’s back work as they do for a window on your Mac. Starkomand keeps the address only; your own ssh keys and config sign you in.

Can two agents work in the same repository?

Yes. Each window says which files it changed and whether another agent touched them. Claude, and Codex once you trust its hooks, check with Starkomand before every edit: an agent can’t edit a file another agent is changing, or stage, stash or revert another agent’s files. It is told why and what to do instead.

Can I switch agents after hitting a usage limit?

Yes. When Claude or Codex hits its limit, press “Continue in Codex” or “Continue in Claude” on the amber chip in its window header. Starkomand never moves a stuck step on its own, and the stuck agent does not need to answer. First stop any running turn and clear unsent drafts. You need to be signed in to the agent you switch to. Here is what to do when Claude Code hits its usage limit.

After a usage limit, does the next agent see what the first one changed?

Yes. It opens in the same project folder, so every edit is already there, including uncommitted ones. It also gets the recorded conversation, shortened if the session was long, and is told to check the files before continuing.

Do agents keep running when I leave?

Yes, while the local server runs. Hide the canvas and your agents keep working. Turn on Settings → Lid closed and a MacBook stays awake with its lid closed while agents work, screen and keyboard dark, then sleeps a couple of minutes after the last one finishes. macOS asks for your password once to allow it. When you come back, tap Space to flip every window and see where each agent stands. A restart of your Mac ends live sessions, but your canvas, notes and task history stay.

Can I separate my projects?

Yes. Each workspace has its own agents, folder, layout and backdrop. The trial and Pro both include unlimited workspaces and agents. Switch between workspaces with the arrow keys.

License and privacy

What happens after the 3-day trial?

The trial includes every feature. After 3 days, you need Pro to keep using Starkomand. You pay once for one computer and get a year of updates. After that year, you can keep using the version you have.

Does Starkomand replace my editor or AI subscription?

No. Starkomand lays out your agent sessions and the context around them. Keep your editor and your provider accounts. Pro pays for the workspace. The providers still bill you for your AI plans and usage.

What stays on my Mac?

By default, your workspace, local dictation and hand tracking. Window backs read your agents’ own session files, and Auto routing decides locally; neither calls a cloud model. Smart reading’s decision model and the Kokoro announcement voice run on your Mac too. Agents and connected services talk to their own providers. Cloud dictation is optional and sends audio to OpenAI; an OpenAI or Grok announcement voice sends the line it speaks. A work summary sends terminal context to the provider you pick in Privacy settings, and only when you ask for one. See the Privacy Policy for details.

I lost my license key.

Open Find your license key and enter the email you used at checkout. Dodo Payments, our payment provider, emails you a link to your purchases. The key is listed there and in your receipt email.