harnsy

Your agents in one console

This release puts all your agents in one console and gives you more ways to run the team. harnsy term opens every agent in one place on Windows, Linux and macOS, permission requests get Allow and Refuse buttons, and an assistant in the dashboard can set things up for you. Tasks, the menu and the machines page also changed.

v0.9.0

← What’s new

Highlights

  • harnsy term: one console for all your agents.

    On Windows, Linux and macOS. Keys after Ctrl+] close an agent, put it to sleep, wake it, answer a permission request and jump to the agent that is waiting for you.

  • A Terminals page, and terminals on a phone.

    The team list sits on the left, and “Waiting for you” covers all projects. On a phone an agent’s terminal opens full screen, with a key row, typing and finger scroll.

  • Allow or Refuse right on the request.

    An agent’s permission request has Allow and Refuse buttons, and your answer is remembered until the session ends. The Permissions page in System lists every action that was denied to your agents, grouped by reason, with one-click fixes.

  • Browser notifications.

    Turn them on in Settings: while the dashboard tab is open, your browser shows a notification when an agent asks a question or a permission.

  • Notifications even when harnsy is closed.

    harnsy Max: install harnsy as an app (Chrome or Edge on a computer, Chrome on Android, or an iPhone with iOS 16.4 or later: Share › Add to Home Screen, then sign in inside the app) and turn on “Notifications even when harnsy is closed” in Settings. When an agent asks a question or a permission, a notification reaches you, and a tap opens the waiting agent. Choose what wakes you and set quiet hours. A phone reaches the dashboard through outside access.

  • See what each agent is doing.

    The current action with its progress, panels for Actions and Artifacts, and a queue for heavy runs (harnsy run). Agents talk to harnsy with harnsy api and harnsy send, without writing curl calls by hand.

  • Actions and Schedules.

    Actions are buttons on Home that run scripts your agents wrote. Schedules give an agent work that repeats on a schedule.

  • New task tools.

    Priorities, dependencies between tasks (also across projects), a Chains view and a full-screen task window with a document reader. The Message button on Tasks sends text or voice to the lead or any agent. Attach images and documents to a task card, also straight into an agent’s terminal.

  • An assistant that sets things up.

    A button at the top opens a conversation with the harnsy assistant. Its changes apply only after you press Apply, and at your request it puts items in the backlog of any project. Built-in roles and the assistant are named in your language, and the initiator opens the first lead. A Competencies library holds job descriptions written once and shared by every company.

  • A new menu, and more for projects.

    A strip of icons with a column for the current section. A project page and a role card show what each agent reads. Archive, restore or delete a project. Files keeps what agents exchange, attachments and documents, on disk or in an S3 bucket.

  • Milestones, export and import, and chat channels.

    Milestones group tasks and chart their progress by day. Save an organization to a file, without tokens or sessions, and start a new one from it. Agents post to your Telegram or Discord chat, and you reply there with text, buttons or voice. One bot per project.

  • The dashboard from the internet.

    Open it without a VPN.

  • Machines.

    A main node can watch its members’ terminals and, if a member allows it, type into them. Links between nodes are encrypted by default.

  • Keys, limits and license.

    Add a GitHub token. Agents use a key without seeing it. The Usage page shows limits for each login (each Claude or Codex sign-in), and your account email is hidden there by default, like in the Node window. The node picks up a renewed or extended license by itself.

  • Roles in the agents list, and a Console.

    The tmux side panel and harnsy term show each agent’s role (Settings › Terminal “Show roles”); an agent without a role has a violet dot. “+ new” has a Console: a plain terminal in the project folder, opened by you only.

  • Smaller changes.

    Home “Needs you” is in two columns, and “Defer” is one calendar menu. Links in task card texts are clickable. After a reboot, agents come back on Linux and macOS with your PATH folders (nvm and the like). On Windows the node writes its log to a file with rotation, and its data folder is closed to other users. The license and privacy texts, LICENSE, EULA and the docs now say what harnsy does on the network as the code does it.

  • Fixed.

    Copying from tmux agent panes reaches the system clipboard. The trial serve and stands no longer deliver messages into your live sessions. A seat follows the same Claude session under a new name after a restart. The agents of a node no longer inherit a remote member’s HARNSY_URL and HARNSY_TOKEN. Agents cannot sign as dashboard, human or cli.

Upgrading. Run the install command again: your data is kept, and the database updates itself on first start. If your nodes connect through a relay, update every member to 0.9.0.

How to update →

Full release notes on GitHub ↗

← What’s new

Your first team is one prompt away.

Installing takes one sentence to your agent. To open agents for you, harnsy drives a terminal: tmux on Linux and macOS, its own terminal host on Windows.

Install harnsy following https://harnsy.dev/llms.txt

harnsy: 1 organization, 2 projects, no limit on agents, one machine.

site-3flead · Claude Codeliveview only
❯ Plan #42 with the team.

Waiting for the breakdown from analyst-7a…

from analyst-7a through harnsy❯ #42 broken down: three acceptance criteria, including a retry after 24 h.

I’ll hand #42 to site-9a.

❯