A thing the size of your little finger that sits in my pocket and does my most frequent jobs with one button: ask an AI, hand a task to my computer, write a thought into Obsidian, switch off the TV, send a message home, check my server balance. The gadget itself is about 25 dollars from the manufacturer. I wrote the firmware myself.
⚡ The short version — TL;DR
- This is not a finished gadget, it is a blank page. The M5Stack Stick S3 is a programmable box on an ESP32-S3 chip with a screen, three buttons, a microphone, a speaker, Wi-Fi, Bluetooth and an infrared emitter. What it can do is decided by the firmware.
- The value is not in “a smart gadget”. The value is that ten steps on a phone became one button.
- Voice is the main mode. Hold the big button, speak, let go. A simple question gets a spoken answer. A task goes to the computer and happens there.
- “Write this in Obsidian” is my most frequent command. The thought lands in my daily note without the phone coming out.
- The other modes: a TV remote, a ready-made message to the family chat, a teleprompter and phone remote, a rep counter, a server balance, shutting the laptop down.
- The main downside is the battery. Sleep fixes it: after five minutes idle the device drops off by itself, and then a charge lasts the day.
- The second downside, honestly: a phone does almost all of this. The difference is that a phone pulls you in — you unlock it for one thought and leave the notifications twenty minutes later.
- You can put the firmware on your own. Installers for Windows and for macOS on Apple Silicon: plug the Stick in with a cable, press one button. It finds the device itself and backs up what was there.
📺 Video review
The video about this gadget is on the Elton Labs channel. This page is the same story in text, only longer: with what did not fit in the video, and with every setting.
What this is, in plain words
M5Stack is a company that makes small electronics on ESP32 chips. Its best-known models are the Stick and the Cardputer. The Stick S3 is a black box the size of your little finger: a colour screen, three buttons, a microphone, a speaker, a battery, Wi-Fi, Bluetooth and an infrared emitter like the one in a TV remote.
Two things about it appeal to me: the size, and the fact that it is not a finished gadget but a blank page. Out of the box it does nothing interesting. Everything it does was thought up and written by its owner.
This is not a smartwatch and not a speaker. It is an operating system I assembled for myself.
Roughly speaking, I am building my own phone — except I choose the functions, not the manufacturer. Nothing unnecessary physically fits inside it.
So from here on I am not showing you "device capabilities" but the specific jobs it has taken off my hands.
Three buttons, and what is inside
There are exactly three buttons, and that is a deliberate part of the design: there is not much to choose from, so there is not much to think about.
Inside there is Wi-Fi (2.4 GHz only — it will not see a 5 GHz network), Bluetooth, an infrared emitter, a microphone, a speaker, a gyroscope with an accelerometer, and a magnet in the case. Everything else grows out of that set: Wi-Fi gives you voice and the internet modes, Bluetooth gives you phone control, the infrared port gives you the remote.
While it waits, a little face lives on the screen — something like BMO from Adventure Time: it blinks, it looks at you, and when it answers out loud its mouth moves in time with the words. A small thing, but it is the reason you want to pick the device up.
Voice: a question, a task, the news
This is what I use most. The mechanics are the same every time: hold the big button, speak, let go. While you hold it, it records, and a volume bar jumps on the screen.
Then the device works out for itself what you wanted. You do not tell it “question mode” or “task mode” — the model decides from the meaning of the sentence.
Option one — a simple question. “What is the weather in Istanbul” — and a couple of seconds later I hear the answer out loud. Not as text on a tiny screen, but as a voice, while my hands are busy. If the question contains the word “search”, the answer is built from a fresh web search rather than from the model’s memory.
Option two is more interesting — a task for the computer. Behind the gadget sits my local agent on Claude. It wakes the computer and does ordinary things: assemble a PDF, reply to a message, pass my words on to the family chat. I speak into a button, and the work happens on a machine I am not even near.
Option three — the news. I say “news” or “what is going on today”, and I get the five main things of the last day, picked for my topics: AI model releases, product launches, what makers are talking about, what is new around note-taking. Each item is two sentences: what happened and why it matters to me. At the end: which of it is worth a closer look today.
The point is not that “the AI is in my pocket now”. The point is that between the thought “what is in the news” and the answer there is no feed for me to sink into.
"Write this in Obsidian"
My favourite command. I keep every thought, idea and plan in Obsidian — and to write down one line I used to have to get the phone out, unlock it, open the app, wait for a sync. Four steps for ten words.
Now: hold the button, say “write in Obsidian, an idea for a video about…”, let go. The gadget confirms that the thought arrived, and the line appears in today’s note — with a timestamp, in the right section.
This works through the connector in my plugin: the gadget hands the text to the server, the server to the plugin in Obsidian, the plugin appends a line to the daily note. There is no separate sync to set up — if you already have Elton AI, the channel is there.
The other modes
TV remote
The Stick has an infrared emitter — the same part an ordinary remote has. So the “TV” mode turns it into a remote: volume, channels, power, menu. Samsung codes are built into the firmware, but that is not really necessary: it has a receiver too, and it can learn from your own remote — point, press, remembered.
A separate pleasure: you can command the TV by voice. You say “turn it down”, the model recognises that as a TV command, and the gadget sends the right infrared code itself.
A message home with one button
The simplest mode and the most used one. Press it and a ready-made message flies into the family chat: “I am home”, “leaving now”, “ten minutes away”. No words, no phone, no “hang on, let me unlock it”.
Phone remote: teleprompter, volume, shutter
Over Bluetooth the gadget pretends to be a mouse and a keyboard for the phone. Then the useful part starts: the phone sits on a tripod recording video, I read the text off its screen, and I scroll it with the gadget in my hand — hold the big button and the screen moves by itself, at a steady pace. The same button turns the volume and fires the camera shutter.
Counter
For counting repetitions when losing track matters: sets, laps, prayer beads. One press is plus one, the number fills the screen.
Server balance
This is what started the whole thing. The “Server” mode shows, with one press, how much money is left in my service and how many customers there are right now. I used to open a browser for that number — which meant opening the laptop and landing in my inbox.
The "More" section
Everything that did not fit on the main screen lives in a separate list: a message to the chat, the remote, phone mode and, for instance, shutting the laptop down — handy when a long render is going and you do not want to walk back to the desk just to press a button. This is the kind of thing where a new mode gets added in an evening, because it turned out to be needed.
How it works under the hood
This was not in the video, but without it there is no understanding why the thing is quick.
One small thing that saves seconds: the moment you enter “AI” mode, the device quietly pokes the server so the secure connection is established in advance — before you ask anything.
What it has that I have not touched yet
On top of the Stick there is a sixteen-pin header, and a Grove socket on the side. That means an extra module can be put on top rather than soldered: a temperature and humidity sensor, a motion sensor, GPS, a joystick, a tiny keyboard, a strap to wear it as a watch. All of it is sold separately and costs very little.
There is also a gyroscope with an accelerometer inside: you can drive the gadget by tilting it, like a mouse in the air. And a magnet right in the case — it sticks to a fridge, to a computer case, to any piece of metal. For household buttons that is more useful than it sounds: hang an “I am leaving” button on the fridge and it lives there.
The downsides — honestly
And one more honest thing, not a downside this time. For me it is not only about the functions: it is a separate device that is always to hand, does not drown in notifications and is simply nice to hold — it reminds me of nineties toys with Tetris on them. It is also where I test ideas quickly: before building a dedicated device for a job, I put the job on this one in an evening.
Before you buy: what matters
This is the most useful part of the page, because half of these things are not in the product description and you find them out at home.
Which model exactly
You need the M5StickS3 — on the ESP32-S3 chip. Not the “StickC Plus2”: that is the previous generation, discontinued, with a different processor and neither a proper microphone nor a speaker of this class. And not the Cardputer or the Atom — different case, different buttons, different screen. My firmware is written for the S3 and will not install on other models.
From the manufacturer it is about 22 dollars. Resellers charge noticeably more, often double: that is shipping and a middleman’s margin, not a different product. The box contains the stick only — cable, modules and strap are separate.
Five things people trip over
What is not inside
There is no memory-card slot — everything lives in the built-in 8 MB. No GPS, no cellular, no NFC: those come as modules on top, and a cellular module will want its own power supply. There is no separate charging indicator either — the battery state is shown on screen, and only because the firmware says so.
What is nice and not obvious
The infrared port works both ways: instead of hunting for codes online you can teach it from any remote you own. There is a magnet on the back — the stick sticks to a fridge or a computer case. On top there is a sixteen-pin header and a Grove socket on the side: a module fits without soldering, but power over Grove is limited to about 0.38 A, so hungry peripherals do not belong there. Inside there is a BMI270 gyroscope with accelerometer, a microphone and an amplifier with a separate audio codec — meaning the voice side is a proper signal path, not a buzzer.
The dimensions are 48×24×15 mm and it weighs 20 grams: genuinely the size of a little finger, and you do not feel it in a pocket.
What to buy alongside it
A USB-C cable with data, if you do not have a spare. A Grove cable — it will come in handy for any sensor. Buy everything else later, for a specific job: until you have lived with the device you will not know which module you need.
How to put the firmware on your own
I have taken the firmware to the point where I am not embarrassed to hand it over. It does not install the way microcontroller software usually installs — no development environment, no command line, no picking ports.
Updates arrive in the same place: "More" → "Update firmware", on the same key. Only the program is updated; your network, key, model and theme stay where they are.
Ready-made firmware · installers for Windows and macOS Put the firmwareon your own Stick One installer, one button, two minutes. AI key and starting credit included. $29.99 once, no subscription Open the page →
Four screen themes
The screen can be dressed differently: “Face” — a dark background with an orange accent, as in the video; “Paper” — light and contrasty, the best one to read in sunlight; “Terminal” — green on black, like old displays, and the lightest on the battery; “Midnight” — dark blue, calm for the evening. You pick one in the installer and change it in the stick’s own settings.
Where this could go next
Here is where I am thinking of taking it — and where you could take it, if you build yourself one:
Common questions
Do I need to know how to program?
To install the ready-made firmware, no. The installer does everything; programming starts where you want a mode of your own. The sources are open, and modes are built so that a new one is a single file.
Will it replace a phone?
No, and it should not. It takes a few frequent jobs off the phone so you do not have to get the phone out for one line of text. Everything else — camera, maps, messaging — stays with the phone.
What does it cost to use? Is there a subscription?
The device itself is a one-off purchase. The voice modes spend money on the AI model: pennies per request, paid straight from your own key. There is no mandatory subscription for the firmware to work.
What about privacy? Is it always listening?
No. The microphone only wakes while the big button is held — like a walkie-talkie. Nothing is stored: on the server the audio becomes text, the text goes where it is going, the audio is thrown away.
Does it work without the internet?
Partly. The TV remote, the counter and the phone control over Bluetooth all work. Everything involving voice and AI needs Wi-Fi.
Can I put the firmware on another M5Stack model?
The firmware is written for the Stick S3 — for its screen, buttons and wiring. It will not install on a Cardputer or an Atom without rework.
What if I do not like the firmware?
Before writing anything, the installer saves a full copy of the device’s memory. “Restore from backup” puts everything back exactly as it was, settings included.
Tell me in the comments under the video which mode you would hang on the free button. The good ideas go into the firmware — that is how the news mode and the laptop shutdown got there.
This is not an ad and not a sponsorship: I bought the gadget with my own money and wrote the firmware myself. In the age of AI it seems worth trying the things that actually make life easier.