Agent Workroom

A small 3D office for the AI agents you hand work to. Each agent has a desk with its current task on the monitor; the wall holds every task as a sticky note in to do, in progress, in review or done. Type or say what needs doing, and pull real progress from a GitHub repository's pull requests.

0 tasks Drag or ←→ to look · WASD to walk · / to type

Tasks

#TaskAgentStatusProgressHand off

No tasks yet. Type one in the bar inside the room, or load the example.

Set up

Agents

Progress from GitHub

Pull requests from a public repository become tasks: draft → in progress, open → in review, merged → done.

Your data

Saved in this browser only. Export to back it up or move it.

Things you can type or say

fix the login redirect for BuilderNew task on Builder's desk. Anything that isn't a command below is a new task.
add task: rotate deploy key @Scout #ops !Typed shorthand: @agent, #tag, ! for high priority (or say "urgent").
start 2 · finish 2Move task 2 to in progress, or to done.
move login to review · mark task three doneRefer to a task by number or by the start of its title.
assign 4 to ReviewerHand a task to another desk.
add agent called AtlasA new desk.
delete 5Remove a task.

A command that could mean two tasks does nothing and says so, so a misheard word never moves the wrong card.

Questions

Is this real VR?

No headset needed and none supported yet. It is a 3D room drawn with CSS that you walk through with the keyboard or by dragging, on a laptop or a phone. It feels like a small virtual office; it is not a WebXR app.

Where are my tasks stored?

Only in this browser's local storage. Nothing is sent to a server of ours. Export the board as JSON to back it up or move it to another device.

Does voice input send my audio anywhere?

Voice uses your browser's built-in speech recognition. Some browsers, Chrome among them, send the audio to the browser maker's speech service to transcribe it. The page only receives the text. If you'd rather not, type instead: the commands are the same.

How does the GitHub import work?

Give it a public repository. The page asks GitHub's public API for the 40 most recently updated pull requests: drafts become in-progress tasks, open ones in-review, merged ones done. Branches named claude/…, codex/… or copilot/… go to a desk for that agent; other pull requests go to their author. No token is used, so GitHub allows about 60 requests an hour from your address.

How do I hand a task to an agent?

Each task row has Copy brief, which copies a short prompt to paste into your agent's chat, and, once a repository is set, Draft issue, which opens GitHub's new-issue form prefilled. You review and submit it there; this page never posts anything for you.