How to See Every OpenClaw AI Agent You're Running in One Room

If you run more than two or three agents on OpenClaw, you already know the problem. Work is happening. Gateways are online. Sub agents are spun up under them. But there is no single surface that tells you what is actually running right now.

Claw Crib is the answer to that, and it is now installable via NPM.

It renders your OpenClaw agents as characters inside a 3D loft workspace. When an agent picks up a job, you watch it happen in the room. That is the entire premise, and it took three full revisions across several months to get to something this small.

The version worth shipping was the simplest one

The honest origin story here is a builder's failure mode: overcomplicating. Too many features. Too many different things bolted on. Claw Crib went back and forth through two prior versions before landing on the one being demoed.

The third revision is the good one precisely because of how simple it is. Not because more was added. Because more was cut.

That is worth sitting with if you build tools. The instinct on revision three is usually to layer on capability. The win here was resisting that, and the reason it shipped at all was a cadence rule: get in the habit of shipping one thing per month. Live and imperfect beats polished and unreleased.

The demo in the video includes a live bug. The view fails to update, there is a hard refresh on camera, and the narration is just "yeah, there's a little bug right now." It stayed in the video. The tool being in people's hands matters more than the tape being clean.

Installing Claw Crib

Installation is one NPM line, and there are two paths depending on where OpenClaw lives for you:

• SSH into your OpenClaw server and run the install line in the terminal.
• Or, if you run OpenClaw locally on your PC, run the same line in your local terminal.

Then follow the steps.

If you skip the CLI steps, you are not stuck. The app walks you through initial onboarding the first time it loads, so you can configure from the browser instead.

Once it is running, you land on your team as a blank canvas. In the demo, that first view shows the gateways that stay online 24 hours a day, each with a stack of sub agents tied to it.

The floor plan is the settings screen

There is no separate configuration panel in Claw Crib. The blueprint is the config.

Open Plan in the top right and you get a blueprint-friendly floor plan. Whatever you do there is reflected directly in the 3D view. Stretch the plan super wide, then pull it super narrow, flip back to 3D, and the room you are looking at has changed shape to match. Nothing to save and apply separately.

From there you can create multiple rooms, multiple buildings, and multiple floor plans. They can be saved and reverted, so an experiment that does not work is not a one way door.

The workflow across the modes goes like this:

• Plan sets the overall floor plan and shape.
• Room creates a new room and names it.
• Organize assigns which agents live in which room.
• Build resizes and repositions the room itself.
• 3D is where you check the result.

Building an exec room, live

The demo builds one room end to end. A new room gets created, it drops in on the bottom side of the plan, and it gets named "exec." Four agents get assigned to it: the main agent, Gary, Elon, and Warren.

Then Build widens the room and moves it up into position, and the view flips back to 3D.

And the main agent is in the wrong place.

The debugging happens on camera. Plan. Build. Room. Organize. Exec shows four agents assigned, but they keep landing on the open floor instead. Then the realization: "I wonder if it's because of the size. Yeah, so it's the size of the room dictates how many agents can be in there."

That is a real constraint, not a rendering quirk. Physical room dimensions determine agent capacity. If your agents are not showing up where you assigned them, your room is too small to hold them.

Back to the floor plan, shrink it slightly, back to 3D, and voila. Executives in one room, everybody else out in the bullpen. A super loft inspired workspace where the layout carries the org structure.

Watching agents actually work

The payoff moment is not the layout. It is what happens when work comes in.

A message goes out to Muddy in Telegram, and the dashboard shows Muddy dispatching. Moments later, Elon gets dispatched work while Muddy is still reasoning. The dashboard proves itself mid demo, without any staging.

That is the point of the whole thing: visually seeing what agents are doing what, when, and where.

What Claw Crib is not

This part is stated plainly in the video, and it is the most useful thing to know before you install it.

Claw Crib is not a control panel. You do not come into it to control your OpenClaw setup. It was always meant to be an always on AI agent activity screensaver.

The origin explains the scope. It was built to live on a TV in the background and stay on 24/7. Second screen, wall display, ambient awareness. Put it somewhere you glance at, not somewhere you operate from.

Setting that boundary out loud is what keeps the tool honest. It does one thing, and it tells you upfront that it does not do the other thing.

Known limitations and how to work around them

One is already flagged: the main agent naming. Whatever you name your main agent, the dashboard still labels it "main." A customizable field may come later.

In the meantime, the workaround is explicitly permitted. Dissect the NPM package and make edits as you see fit. You are more than welcome to do that.

The roadmap belongs to the comments

The line that frames the whole release: "It can't improve with your feedback."

So the roadmap is handed over. Bugs and feature requests dropped in the comments get folded into the next NPM push. Maintenance is committed at least once a month, and possibly a couple of times a week depending on how much traction it gets.

Some directions already on the table for future iterations: making it a bit more complex, adding a garage, creating separate rooms that tie into each other.

And if you want to watch the sausage get made, a lot of this was built live. The streams run three days a week, generally Monday, Wednesday, and Friday at 9:00 a.m. Pacific.

Try it

Install it with the single NPM line, either over SSH on your OpenClaw server or locally in your terminal. Size your rooms generously. Group your executives together, leave the rest in the bullpen. Then leave it running.

Watch the full demo, including the live bug and the room capacity debugging:

https://www.youtube.com/watch?v=gyHaVqOCOKE