ClawWhy a piece of hardware, at all
The case against a desk device is obvious and worth stating first: you already have a screen. Two, probably. Adding a third one that does less than either seems like the definition of a gadget.
The case for it comes down to a cost nobody measures, which is the cost of checking. When watching your agents means opening a tab, that check has a price: the context switch, the reload, the ten seconds of reading. So you do it less often than you should. And the moment the checking is expensive, a stuck agent stays stuck, because the cheapest thing to do is assume it is fine.
Ambient displays fix that by making the check free. You do not open a wall clock. You glance at it, from across the room, while doing something else. That is the whole design goal here, and every pro and con below follows from it.
The pros
1. Today's spend stops being a monthly surprise
The single most useful number on the glass is what you have spent today, across every agent, at API rates. Not this month. Today. A runaway loop that would have shown up on an invoice three weeks later is instead a number that looks wrong while you are still drinking your coffee.
This matters more than it sounds because agent cost is spiky by nature. One misconfigured loop can outspend a normal week in an afternoon. The value is not the precision, it is the latency: you find out in minutes instead of in billing.
2. It catches the stuck agent, which is the expensive one
Agents rarely fail loudly. They fail by retrying the same build step 38 times, or by making tool calls that go nowhere while the token counter climbs. Nobody watches for that, because watching for it is boring and constant, which is exactly what a machine should do instead of a person.
The device surfaces the one agent that is looping or has stopped making progress, at the moment it happens rather than the next time you happen to look.
3. Unblocking takes one tap instead of a tab hunt
An agent that paused to ask permission is not just idle, it is often holding up everything downstream of it. The gap between "it asked" and "you answered" is dead time you are paying for in wall clock, and it is usually long because the request is buried in a terminal you are not looking at.
On the device the request appears on the glass with the exact command it wants to run, and you tap Approve or Deny. The agent unblocks immediately. This is the feature people report as the one that changed their day, more than the cost number.
4. It is runtime-neutral, so it does not go blind next month
Most desk gadgets in this space are built for one assistant, and stop being useful the moment you add a second tool. This one shows whatever ClawMetry supports: OpenClaw, NVIDIA NemoClaw, Claude Code, Codex, Cursor, Aider, Goose, opencode, Qwen Code, Hermes, PicoClaw, NanoClaw, Pi, Deep Agents, n8n, Antigravity, GitHub Copilot and more. Given how fast people switch agent tools, that is the difference between a device with a one year life and one with a five year life.
5. Calm by default
Most of the day it shows a resting screen that says everything is fine. That is deliberate, and it is what makes the loud states legible. A display that is always busy teaches you to ignore it, at which point it is decoration.
6. Privacy is structural, not a promise
Your agents' conversations and activity are encrypted on your machine before they leave. The device holds the key and decrypts on the screen itself, so the cloud stores ciphertext it cannot read. Totals such as cost and health status travel readable so the glance is instant. You do not have to trust a policy document; the key was generated on the device and never touches our servers in readable form.
7. No subscription for the hardware
$49 once, and there is no device subscription, ever. Firmware updates arrive over the air for free, with automatic rollback if an update fails to boot, so you never plug it into a computer again after setup.
The cons
1. If you run one agent, you do not need this
Be honest about your own situation. With a single agent in a single terminal, that terminal already is your dashboard, and it is bigger and more detailed than a 4 inch screen. The device starts earning its place somewhere around three or four concurrent agents, and the value climbs steeply from there. Below that it is a nice object, not a useful one.
2. It needs the cloud
The ClawMetry software self-hosts, and plenty of people run it that way. The device does not: it pairs and reads through ClawMetry Cloud. If your requirement is a fully air-gapped setup, the device is not for you today, and no amount of wanting it to be changes that. Use the self-hosted web app instead.
3. It is a glance, not an analysis tool
You cannot read a transcript on it, diff two sessions, write a policy, or dig through last month's costs. All of that lives in the web app, and should. If what you actually want is better analysis, the device is the wrong purchase; the software is free to try and does that job.
4. Not every control works on every runtime
This is the part most likely to disappoint if nobody tells you first. Approve and Deny come from runtimes that pause and wait for a decision, which today means OpenClaw and NVIDIA NemoClaw. Pause, Resume and Stop work on Claude Code, Codex, Goose, opencode and Aider as well. Cursor is the exception: the editor runs all of its conversations inside one process, so there is no single session to stop. The device says so on the glass rather than showing a button that quietly does nothing, but the limitation is real.
5. The software tier is a separate decision
The $49 buys the hardware. On the software side, OpenClaw and NVIDIA NemoClaw are covered on the free plan, but watching Claude Code, Codex, Cursor and the rest needs Starter at $9 per node per month. There is a free trial and no card required, but if you were expecting the device price to cover everything, it does not, and you should factor that in before ordering.
6. 2.4 GHz WiFi only, and it wants to stay plugged in
The ESP32-S3 radio does not do 5 GHz. On most routers this is a non-issue, but if you run a 5 GHz-only network you will need to give it a 2.4 GHz network name. There is also no battery and no power button, by design: a status display that sleeps is one you stop trusting. It needs a USB-C power source within cable reach of where it sits.
7. It does not talk yet
Every unit ships with a microphone and speaker onboard, so if voice ships it arrives as a free over-the-air update with no new hardware. We are not promising a date, and you should buy it for what it does today, not for that.
8. It will not decide anything for you
Depending on what you wanted, this is either a pro or a con. The device never stops an agent on its own. Watching is automatic; stopping is always a deliberate human tap. If you were hoping for a box that autonomously kills runaway agents while you sleep, that is policy enforcement, and it lives in the ClawMetry software with its own explicit opt-in, not on the glass.
Who should buy it, and who should not
Buy it if: you run several agents at once, across more than one tool; your agent spend is large enough that a bad afternoon is a real number; you are the person other people's agents wait on for approvals; or you have already noticed that you check on your agents less often than you know you should.
Skip it if: you run one agent; you need an offline or air-gapped setup; you want deeper analysis rather than faster awareness; or your agent bill is small enough that a runaway would not hurt.
If you are unsure, install the free software first. pip install clawmetry gives you the same data in a browser. If you find yourself keeping that tab open all day and still missing things, that is the signal the device is for you.
Ready to set one up?
Already have a device in hand? Take it from the box to a live glance in about five minutes.
Device setup guideWhat setup actually involves
Worth knowing before you decide, because "it needs setup" is a fair thing to weigh. For a device ordered from us, which arrives pre-flashed, it is four steps and about five minutes: power it on, join it to your WiFi through its own setup hotspot, install ClawMetry on the machines your agents run on, and pair it to your account by scanning a QR code. The full walkthrough, including the fixes for the problems people actually hit, is in the device setup guide.
If you would rather source your own Waveshare ESP32-S3 Smart 86 Box, you can flash our firmware onto it from the browser in about three minutes at clawmetry.com/flash, with no toolchain and no command line.
Once it is running, how to use your device covers every screen it can show, what each tap does, and the limits above in more detail.