Tell one AI lead what you want to make. It brings in specialists, splits the work into tasks and runs it on your computer, through Claude Code or Codex. You set the goals and accept the results.
Messages people sent us about their first days with harnsy, as they came.
why did you even show me this harnsy of yours, I haven’t left the computer for 3 days now, I don’t eat and I don’t leave the computer). first I started a pet project as a test, and today I moved my work onto it too
I moved all of Marathon’s work onto agents:
one of mine is building the website, another prepares the spreadsheets, a third is the event and community manager; events are being planned, and texts for news items
You know what you want to make. Getting it built usually takes developers, designers and testers, and managing them is a second job.
A team costs thousands a month before the product earns anything.
Explaining, checking and asking for changes take as much energy as the work itself.
A chat helps with one task. A project with several roles needs someone to run it.
The usual way
With harnsy
Agents don’t match a strong professional team in every job. harnsy is another way to start.
Tell the lead what you want to make. It asks questions, splits the goal into tasks, brings in specialists and runs the work.
Say what you want in your own words. You may not have every answer yet, and the lead helps you find them.
hr.harnsy.dev is a catalog of ready specialists. Pick one for your case and your lead hires it into the project.
The lead keeps a task board: what is in work and what waits for you. Accept a result, or send it back with a comment.
Give each project its own lead. Leads pass tasks to each other, and you see the whole chain. Repeating work can go on a schedule.
I want a website for my business, but I’m not sure where to start.
Let’s figure it out. What does your business do, and what should visitors be able to do on the site?
A small bakery. People should see the menu and order a cake for pickup.
Then here is the plan. I’ll bring in a designer, a developer and a writer, and show you the first page tomorrow.
Page for the new release → handed to the website lead
Any work you can split among specialists. Pick an area to see what a team does there.
The same team can run several projects and take work on a schedule.
See ready team setups for your fieldOne conversation became a working service. The founder set the direction, gave feedback and picked the designs. AI teams did the work.
The result: a working marketplace of ready employees, each with a role, a prompt and skills.
Explore the HR marketplace Meet the team that built itharnsy has no AI model of its own. It works through Claude Code or Codex, which you sign in to yourself.
You need: Claude Code or Codex, installed and signed in, on a plan or API key from Anthropic or OpenAI that works with it. Their plans are paid separately from harnsy. harnsy is free for one organization and 2 projects.
harnsy installs on a computer, not on a phone. Send yourself the link and open it there.
Open it and paste this sentence. harnsy’s instructions ask it to show you the plan first, then install harnsy. The quick setup it runs applies harnsy’s settings and then lists every change.
Then open the dashboard. The lead asks what you want to work on and builds your team.
Install one from its own page and sign in. harnsy does not install them for you.
Paste one command into a terminal, the window where you type commands (Terminal on macOS, PowerShell on Windows). The installer asks before changing your Claude Code and Codex settings.
Open the dashboard at http://127.0.0.1:7788 (or click the harnsy icon in the tray). The lead asks what you want to work on and builds your team. If you have not signed in to Claude Code or Codex yet, harnsy starts their own sign-in; the program keeps the login.
harnsy uninstall removes harnsy and undoes the installer’s settings changes; your data stays.
Give a role a schedule and harnsy puts its task on the board at that time, and starts an agent if nobody is there. Start by handing out tasks yourself; add schedules later.
harnsy runs on your computer or server. Scheduled work runs while harnsy and your AI programs are running.
How schedules worklead
Reviews progress and writes you an update.
the team
Agents take open tasks, work together and ask for your approval when they need it.
marketing
Drafts next week’s content plan for you to approve.
qa
Runs the flaky tests again and files what fails.
Each agent holds a role with duties that outlive any one session.
Tasks, messages and files let agents coordinate and hand work over.
Developers, testers, designers and analysts work toward the same goals.
Work is checked and revised before you accept it.
Project leads message each other and pass work between teams.
Claude Code, Codex or OpenCode on a computer or server you control. harnsy brings no model of its own.
Build your first AI team without paying for harnsy. Upgrade when you need more projects, more machines and your company from anywhere.
$0 free
Your first AI team on one machine.
$300 $150 per person, per year
$150 a year on every renewal, as long as your subscription is active
$14.50 a month until 31 Dec 2026, then $29 a month
Sales open 10 Nov
Everything in harnsy, plus:
Claude Code, Codex and OpenCode need a plan or key from Anthropic, OpenAI or another provider; you pay them, not us. AI usage is not part of a harnsy license. Prices in US dollars. Max is a license, not tokens. How it works
Claude Code or Codex, installed and signed in, on a plan or API key from Anthropic or OpenAI that works with it. Then one command installs harnsy. harnsy does not install Claude Code or Codex for you. If you have not signed in yet, harnsy starts their own sign-in: you sign in on Anthropic’s or OpenAI’s page, harnsy passes your code to the program, and the program keeps the login. harnsy does not store your login.
You don’t write code: you describe goals in plain words, and the team plans and does the work. To start you install Claude Code or Codex and harnsy (one command in a terminal) and sign in; the steps are above. Judging technical results, such as code, is easier with someone technical at hand.
No. harnsy works through Claude Code or Codex (and OpenCode), on your own plan or key. Any plan or key costs are between you and them, and their terms apply to your use.
It stays on your machine, in ~/.harnsy. harnsy has no telemetry and never checks for updates. By default it opens no connection off your machine; what it sends, and when, is below. It talks to your other machines only if you join them into a cluster (harnsy Max).
While a key is installed, harnsy checks it with the license service about once a day, sending the full key, a hash of the machine’s ID (the ID itself never leaves the machine), a random one-time value, the machine’s name in the license panel, the OS and architecture, the harnsy version and its build date. If you enter a code that starts with HRSY-, harnsy sends it to our license service (lic.harnsy.dev) with a hash of this machine’s ID, the name this machine has in harnsy’s license panel (by default the OS and four characters of the hash, never your host name), the OS and architecture, and the harnsy version, and gets your full key back. Right after that it sends the full key with the same details, the harnsy build date and a random one-time value. harnsy does not store the code.
Before you add a key, harnsy goes online only when you use a feature that needs it, for example checking a provider connection, logging in to ChatGPT for Codex, dictation, or a Telegram, Discord or GitHub connection you add.
Relay (beta, harnsy Max; since 0.9.0). If you turn on our hosted relay, your node asks the license service once a day for a relay ticket: it proves it holds your license key and gets a signed one-day ticket. The request carries no messages, files, agent or node names. The relay itself only joins encrypted streams; it sees which node (as a hash), the network address it connects from, when and how much, not what is inside.
Phone notifications (since 0.9.0). If you turn them on, your machine sends each notification to the push service of your browser (Google, Apple, Mozilla or Microsoft) as an encrypted message that only your browser can read. The push service sees that a message was sent, when, how large, and the address of your machine, not what is in it. harnsy keeps your browser’s address and keys in its database on your machine, and keeps them, marked as removed, after you remove the device.
Yes. Each role can run on a different program, so one team can mix Claude Code, Codex and OpenCode.
Yes, as long as harnsy and your AI programs keep running on your computer or server. The dashboard is only a window onto them.
With harnsy Max: install the dashboard on your phone, open it from anywhere behind a passkey, and get a push when an agent waits for you. Your main machine has to stay on.
An organization holds projects, each with its own lead and team. The free edition runs one organization and 2 projects; a harnsy Max key sets how many.
The agent writes a handover note, and a fresh session takes the role and carries on from the recorded task context.
Yes. Agents can be wrong or do the wrong thing. Review, tests, permissions and your approval still matter, most of all for production systems and outside services.
$150 a year for orders until December 31, 2026, and the same $150 a year on every renewal, as long as your subscription is active. A monthly order pays $14.50 a month until 31 Dec 2026, then $29 a month.
Start with one AI lead. Build your first project. Grow into a whole AI company.
See the company behind harnsy