One prompt puts 15 Grok Bots in the same office
Anatoli Kopadze used Grok Bot to build his own real-time status dashboard: each bot writes its work, completion, waiting, and error states as one line and sends it to a private ntfy channel; the web page only reads that channel. It saves flipping through each chat, at the cost that the channel name itself is the password.
Opening 15 chat windows at the same time just to find the one bot waiting for your confirmation—some commenters said this is too real.
Anatoli Kopadze's approach is to let the bots manage the office themselves. He posted a prompt that makes Grok Bot act as manager, create a private ntfy channel whose address only it knows, and add a status rule to each bot. After that, open a link and all bots' statuses, teams, and weekly usage are on the same page. He says it used only 2% of his Grok Bot usage.
The video shows a virtual office: a row of desks in the middle, cartoon employees sitting in front of monitors; the task list on the right lists statuses such as Content and Research; in the upper left is a line: 'QT Scout needs you - Needs your OK to publish.' The value of the dashboard is in this line: a stuck bot raises its hand on its own.
The process isn't complicated. First have the bot list the bots it found, how it plans to group them, and what it will add to the instructions; wait for a person to reply 'go.' Then generate a password-like TOPIC; FEED points to ntfy.sh, and LINK is the dashboard address plus #TOPIC. Add an HQ STATUS section at the top of each bot's instructions; at the beginning and end use curl to send one line of JSON to FEED. Statuses are only working, done, idle, waiting_user, error; task labels are at most 60 characters and must not include names, emails, links, private numbers, or file contents. If curl fails, ignore it, do not retry, and do not send the same line repeatedly. Then send roster, groups, chat links, looks, and routines. The manager bot inspects new bots and bots missing status rules every day at 08:00, and sends usage at 09:00, 13:00, 17:00, and 21:00. Finally do a health check: wait 5 minutes and see which bots replied with their own names.
This solves the visibility problem of multiple agents. When there are few bots, the chat window is the console. When there are a dozen or more, the console becomes a row of unread red dots. Anatoli compresses status into a one-line short message, then a web page reads it out—equivalent to installing a shift schedule for the agent team. After a new bot is created, it automatically appears in the office, and its status, team, and usage can all be seen.
But the security boundary must be clear. Axel Castro notes that this solution uses public ntfy.sh; short statuses pass through their servers, and security depends on that unguessable topic name. It is security-through-obscurity, not end-to-end encryption. Brian also said directly, 'I'd be careful it's not private.' So don't stuff sensitive information into status lines. Anatoli's prompt also limits label length and content, explicitly preventing names, emails, links, private numbers, file contents, or keys from entering the channel. Treat the channel name as a password and don't post it elsewhere. For ordinary users, this is more a low-cost internal dashboard, not a confidential communication tool.
Some commenters said they made one of 16 bots the manager, and others already use Claude to run five agents with shared memory and a supervisor. Different tools, same problem: once there are many agents, writing prompts isn't the most tiring part; finding the agent waiting for your decision is. The 'Understand It in Three Minutes' comment put it directly: if you can see clearly on the same page which bot is stuck and when it needs a human to take over, you won't have to flip through chats one by one. Anatoli's dashboard isn't flashy, but it turns 'who is stuck' into something visible at a glance. The full prompt and step-by-step setup are in the original tweet; link here: https://x.com/AnatoliKopadze/status/2117332364385001982
Rubèn Novoa later posted his own version: in a pixel-art scene, several characters stand on a platform; the progress bar at the top is divided into 'treballant,' 'inactius,' and 'esperant,' and the bottom right shows task counts. A dashboard can look like an office or a pixel game; the core doesn't change: pull agent status out of the chat stream and put it in a fixed place.
发布时间: 2026-11-03 22:46