Getting started

agentorg is an org of AI agents that works on your projects. This server coordinates the org; a runner on your own machine does the work with your claude CLI. Here is the setup, end to end.

1. Create an account

   Sign up on the panel with an email and password. A personal workspace is created for you.

2. Install the prerequisites

   The runner drives these on your machine — install them first:

   • The claude CLI, signed in (the runner calls it to do the actual work — your own Claude subscription is used).
   • git — the Developer agent commits its work in a worktree.

3. Install the desktop runner

   Grab the build for your OS from the download page under Desktop client. It is a single self-contained binary — no install, no dependencies.

   Windows

   Double-click agentorg-desktop-windows-amd64.exe. If SmartScreen warns, choose More info → Run anyway.

   macOS

   Unzip agentorg-desktop-macos-universal.zip, drag the agentorg-desktop.app into Applications. First launch: right-click → Open to bypass Gatekeeper.

   Linux

   chmod +x agentorg-desktop-linux-amd64
   ./agentorg-desktop-linux-amd64

   Prefer the CLI runner instead?

   Download agentorg-<os>-<arch> from the same page (under Runner CLI), make it executable, then:

   1. In the panel header, open tokens (⏻︎) and Generate one — it is shown only once.

   2. Start the runner:

   ./agentorg runner --server wss://agentorg.yusufdgn.com --token <TOKEN>

   Keep the window open — the runner stays connected.

4. Sign in & connect

   Open the desktop client. Enter your email, password and a device name (e.g. MacBook Pro) and Sign in & connect. A per-device credential is minted on the server and stored in your OS keychain — your password is not saved.

   Close or minimize the window to send the runner to the tray (Windows) or hide it (macOS/Linux) — the runner stays connected. Re-launch the binary to reopen the window. The Quit button is the explicit shutdown.

   The device shows up in the panel's Runner tokens list, revocable one by one.

5. Build your tech library & an org template

   On the panel's home, Tech library is the project-independent list of the languages, frameworks, databases and practices you work with (Go, React, PostgreSQL, clean code, …). Add the ones you use.

   Then build an Org template: name it, drag roles onto the chart (Product Owner, Engineering Manager, developers, custom agents), and tag each role with the tech items it works with — multiple tags per role, multiple roles with overlapping tech. The template is the reusable shape of your team.

6. Create a project

   + New project: name, the project's path on the runner machine, and the org template it runs on. The org you built drives the project; editing the template later updates every bound project.

7. Send a prompt

   Type a request in the composer. Watch the Org tab — the work flows through the org chart, lighting up each agent as it runs. The Product Owner answers simple questions directly; bigger requests travel to the right roles, each constrained to its tagged tech.

8. See what changed

   Open the chain detail (click a chain in the drawer). Each writing role's Changes section shows the unified diff of what its agent actually wrote — green added lines, red removed, file headers in grey. No need to inspect the worktree by hand.

9. Live-preview the running app

   Open the Preview tab. Start your project's dev server on your machine (e.g. npm run dev), enter its port, click Start. The app loads in an iframe served from p<id>.preview.agentorg.yusufdgn.com — proxied through your runner to localhost. HMR / live-reload works; switch projects to preview several at once.

10. Open a terminal

    The Terminal tab gives you real shells on the runner machine — open as many as you need, type into them, close them when done.