Agentry

A task board where the workers are AI agents. You describe a task; a coordinator agent hands it from agent to agent (an engineer, a reviewer, a tester, a deployer, or another team's) and reads every report back, until the work is checked and done. You watch it all live, step in when you want, and approve what matters.

Every action an agent takes passes Agentry's gate first, and your policy decides what runs, what is refused and what waits for you. Agents can work on any of your machines, Linux, macOS or Windows, and you can drive it all from the browser or from Claude Code in any repo. Every step is an event in one append-only log.

See it work

The Agentry launch video.

What it looks like

The Tasks board in Agentry: Backlog, To do, In progress, Review and Done columns of real tasks, each card showing its epic, which agent is working on it, its labels and how many turns it has taken — The Tasks board: Backlog, To do, In progress, Review and Done.

A task's page in Agentry: its plan on the left, the path it has taken from agent to agent across the top (engineer, reviewer, engineer, reviewer and so on), and the Handoffs tab listing each brief and report between the tech-lead, the engineer and the reviewer — A task's path from agent to agent, with every handoff between them.

Agents as an org chart: the Software, Writing, Research and Design teams, each with its lead above the agents it hands work to, such as tech-lead over engineer, reviewer, tester and deployer — Your teams as an org chart: each lead and the agents it hands work to.

How it works

Add a task: New task (or press n). Describe what you want in your own words; the coordinator reads it, names it, and decides who works on it. A big one becomes an epic, split into stories.
Watch it: the task's page shows its plan, the path its work took, a chat with its coordinator, the live terminal of the agent at work, and everything that happened.
Answer it: when a coordinator asks you something, or an agent needs your approval, it shows in Questions (Ctrl+J), and in Claude Code if you connected it.
Add a team: Agentry starts with a Software team. Add Writing, Research, Design, Visionaries or Networking on the Agents page, or build a team of your own there.

Install

git: git-scm.com/downloads
Claude Code, signed in: claude.com/claude-code. The agents run in it. Codex works too, if you have it.

Agentry itself needs nothing else: no Docker, no Node, no tmux. Start Agentry says if git or Claude Code is missing.

Download Each is named for its version, such as agentry-v0.1.0-mac-intel.zip.

Mac

Download agentry-<version>-mac-apple-silicon.zip (M1 and later) or agentry-<version>-mac-intel.zip.
Unzip it, and double-click Start Agentry.command in the Agentry folder.
The first time, macOS says it can't check the developer, since Agentry isn't signed yet: right-click Start Agentry.command, choose Open, then Open again. After that it opens with a double-click.

Windows

Download agentry-<version>-windows-x64.zip (or -windows-arm64.zip on an ARM PC).
Unzip it (right-click, Extract All), and double-click Start Agentry.cmd in the Agentry folder.
The first time, Windows may say it protected your PC, since Agentry isn't signed yet: click More info, then Run anyway.

Linux

Download agentry-<version>-linux-x64.zip (or -linux-arm64.zip).
Unzip it with a tool that keeps permissions (unzip does), and run ./start-agentry.sh in the Agentry folder.
If it won't run: chmod +x agentry start-agentry.sh stop-agentry.sh.

What Start Agentry does

It starts Agentry in the background, if it isn't already running, prints a link for this computer and one for each of its network addresses, and opens the first in your browser. The links carry the page's token, which signs you in, so share them only with people you trust. On Windows the window waits for a key, so you can copy them. Close the browser or the window whenever you like: Agentry keeps running. Run Start Agentry again to open the page again, and Stop Agentry (Stop Agentry.command, Stop Agentry.cmd or ./stop-agentry.sh) to stop it. Agents' sessions carry on across a stop and start.

Where things are: the log, settings and the tasks' worktrees are in .agentry in your home folder; delete it to start afresh.
Your work: new tasks work in the git repo Agentry in your home folder (made the first time), unless you give a task another repo, or a new one.
The address: port 7800, or the next free one, on every network interface, so your other machines and your phone can open it too. The page's token keeps everyone else out; agentry serve -addr 127.0.0.1:7800 keeps it to this computer.

Source

Agentry is open source under the Apache License 2.0, at github.com/agent3-labs/agentry.