A 4-inch touch screen that sits next to your keyboard and shows what you'd otherwise keep a browser tab open for — your calendar, the weather, your build, the thing you meant to do at four o'clock.
It listens, too. Speak and it types. Tap and it records the meeting. And when you need it, the whole screen becomes twelve shortcut keys.
Google Calendar
Home Assistant
OpenAI
OllamaEvery view, always current. You don't start from an empty screen. These nine ship with the device — the meeting you're about to miss, the portfolio, the deploy you keep checking a tab for, the thing you promised yourself you'd do at four o'clock. Swipe through what ships on it.
Dictation, a quick command, a real question and an hour-long call are four different jobs — so each one starts its own way, and none of them starts on its own.
Talk at your normal pace and the words build on the display as you say them, so you can watch the sentence arrive. Tap when you're done and it lands at your cursor — your editor, Slack, the email you've been putting off.
Change what's on the display, open an app, start a timer — hands still on the keyboard, or nowhere near it.
Press once, ask the messy three-sentence version, press again. No countdown to race, no rephrasing for the robot.
Tap record when the call starts, tap Mark when something matters. Afterwards: the decisions, the action items, and the full transcript in the language you met in.
DeskProxy's desktop app isn't a settings page — it's the brain. Configure every tile, remap every key, pick your AI, watch it all update on the device live.
Pick what earns a place on your desk and in what order. Every setting shows up on the display as you type it — the mirror beside the form is the device's real screen, not a preview of one.
Same device, different glass. What earns the space is entirely up to you.
The build goes red at 4:51pm. You see it on the desk, not in a tab you forgot to open. Pipeline status, queue depth, error rate — and a tap that acks the alert or kicks the re-run.
Standup starts in four minutes. The next meeting and its join link sit on the display all morning. Tap record when the call starts; the decisions and action items are waiting when it ends. One key mutes you from across the desk.
The garage door is still open. Home Assistant entities, bench sensors, air quality and the relay you actually need to flip — on a screen you can reach without unlocking a phone first.
A tile is a mini-app with a screen of its own — and every screen on the device is one.
Most desk gadgets ship a fixed set of screens and leave you there. DeskProxy ships an SDK instead: every screen on the device is a tile — a mini-app written in plain HTML and CSS. Build the exact readout your work needs — your build status, your queue depth, your shop's orders, your bench sensors — and put it on glass you can reach.
Nine is where you start, not where you stop. Ten more tiles are in the store today, and it opens with twenty-plus when the device ships — smart home, focus timers, air quality, currencies, CI status, whatever we and the community have built by then. Browse it from the desktop app, hit install, and it's on the display in seconds: no firmware flashing, no cables, no restarts. Every one is a tile like the ones you can write yourself, so the catalogue grows faster than we could fill it alone.
Most AI gadgets route your life through their cloud and charge rent for it. DeskProxy is built the other way around.
Plug in your own OpenAI key, or point it at Ollama and keep every token on your own machine. You pay the provider directly, at their price — no markup, no monthly AI fee, and no key of ours on your device.
API keys are stored in the macOS Keychain, encrypted by your OS — never in config files, never synced to us. We couldn't read them if we wanted to.
The device listens for the wake word and nothing else — no audio leaves it until you've said “Hey Desk”, and the command that follows becomes text on your own machine with Whisper. Dictation and meeting recording are the honest exception: while they're running they are listening continuously, and they stream to a cloud transcriber under your own key. Both start only when you press something.
You register the device once so we know which unit is yours — for warranty, support and firmware updates. That's the whole account. There is no analytics SDK and no usage tracking: what you put on the screen and what you do with it stay yours. Your calendar, stocks and habits sync straight from the sources with your own credentials, and never pass through us.
Every image on this page is an actual 480×320 frame rendered by the device pipeline.
The other good hardware in this space each picks one job and does it well. DeskProxy is the programmable one. This is about fit, not fault.
| DeskProxy | Macro key pads | Pixel-art displays | |
|---|---|---|---|
| Full-screen dashboards you write yourself | HTML + CSS tile SDK | Key icons, plugin SDK | Apps via their cloud |
| Touch on the display itself | Yes — tap, swipe, drag | On LCD-key models | No |
| Runs Mac shortcuts and macros | Yes | Yes | No |
| Works with no subscription | Yes — no cloud tier, ever | No subscription | Often a paid tier |
| Dictation, voice commands and meeting transcription | Yes | No | No |
| Your API keys stay on your machine | OS Keychain, injected on the way out | n/a | Held in their cloud |
A pre-launch page that lists only what's finished is hiding something. Here are both columns.
We're not posting a number until we can stand behind it. The prototype run and the manufacturing quotes come first, and a price we have to walk back later costs more trust than the wait costs attention. The waitlist gets the figure before it goes public.
The hardware is validated and the software is running daily on our own desks. What's left is the production enclosure and the manufacturing run. We're not going to invent a date to look further along than we are — the waitlist gets the schedule the moment it's real.
macOS today, on Apple silicon and Intel. Windows is in active development but will not be ready at launch. The device speaks the same USB protocol on any host; it's the desktop app that needs porting, and we'd rather ship it late than ship it broken.
No. Nine tiles ship on the device and ten more are a click away in the store — clock faces, weather, calendar, stocks, media, habits, reminders, Mac stats and the macro pad. Writing your own is the ceiling, not the floor.
Press dictate and talk. The words appear on the display as you speak — the whole panel becomes the text, so you can see what's been heard. Tap when you're finished and it's tidied and dropped at your cursor in whatever app has focus. Tidied means filler words, false starts and self-corrections are stripped and the punctuation is added; it is never reworded, so your grammar and your phrasing come out the way you said them. It streams to a cloud transcriber under your own API key, which is what keeps it fast enough to follow you.
Not for most of it. Clocks, weather, calendar, stocks, habits, reminders and every macro work with no AI at all. Where AI is involved it's your own key, billed to you at the provider's price with no markup — or free on your own local Ollama. Dictation and long-form meeting transcription are the parts that need a cloud key; they run at a latency nothing local matches yet.
The device talks to your Mac over USB, not over your network. Individual tiles reach the internet only for their own data — weather needs a forecast, your calendar needs your calendar. A clock face works on a plane.
No. There's no cloud tier you get pushed onto and no monthly AI fee — the only account is the one that registers your device to you. First-party tiles are free for as long as the product exists. When you use AI, you pay the provider directly at their price.
A fair question to ask anyone selling pre-launch hardware. The tiles are plain HTML and CSS, the SDK and the USB protocol are publicly documented, and if we ever wind the project down we will publish the firmware and protocol spec so nobody is left holding a brick.
A 4-inch IPS capacitive touch panel at 480 × 800, a dual-core RISC-V SoC, a far-field microphone, an RGB notification LED, a battery-backed real-time clock, and a single USB-C cable carrying both power and data. About a watt, no fan, no wall wart.
Leave your email and we'll tell you the moment it's available — pricing, availability, and first access for the people on this list. One message when it matters, nothing else. Ever.
Pricing and ship dates go to this list before they go anywhere else. No spam and no newsletter — one email when early-bird pricing opens, one when it ships.
You're on the list. Talk soon. ✓