Why I wrote ClearRemote

Posted (in Japanese) to X on August 7, 2026. This is a full English translation. See the original post on X →

I have used quite a few remote desktops over the years — Chrome Remote Desktop and several Japanese freeware tools among them — but none of them fit the way I work, so in the end I wrote my own. That is ClearRemote.

Every remote desktop I used was good software. I was the one being helped by them, and I have no intention of talking them down. They just did not fit the way I work.

We are a seven-person machine shop in Komaki, Aichi, Japan. My days go like this: I spin 3D models around in CAD/CAM and program toolpaths while running a 5-axis machining center; I also look after our thirty-odd PCs and servers; I handle the flood of RFQs from customers together with everyone in the company, on Salesforce and the file server; I plan part production; I check the numbers on the finance PC at the head office; in between I develop Salesforce objects and write software; and I read my X timeline.

Working like that means the finance PC at the head office 20 km away, the servers in the server room, the CAM workstation, the production-control terminal, the software-development PC — I operate all of them from the PC in front of me, without ever leaving my seat.

Which is to say: I am a remote-desktop junkie who depends on it every single day.

The office staff often work from home too, so they are remote-desktop junkies in training.

Why so many separate PCs in the first place?

We split PCs by role.

• The finance PC
  It sits at the head office 20 km from my machining site, talks to the production-control system and the CRM, and hammers the database constantly — its memory is pinned at the ceiling, so it cannot double as anything else. It also holds the company's financial data, so it lives where only a few people can touch it (with thorough backups, of course).
• The CAD/CAM workstation
  CAM is hungry for compute, so this machine carries an expensive CPU and GPU. A shop floor with airborne dust and oil mist (ours is air-conditioned and not that bad, honestly) is no place for it — I think most shops keep it in a CAM room. Let a workstation breathe dust and oil mist and the fans and boards gum up, the cables turn sticky, and the painfully expensive GPU dies. So the workstation lives in the CAM room with air conditioning and an air purifier, separate from the PC you keep at the machine.
• The software-development PC
  This is the PC where the AI lives. ClearRemote itself is being built on it, with an AI writing the code. My Salesforce object design happens there too. Editors, tests, AI agents, Chrome tabs and subagents all eat memory without restraint, so sharing it with my main PC got too heavy and I split it off. AI-driven development runs long and unattended — while a build or test run is going I leave it alone and drop in over remote desktop from my own PC to see "is it done yet?". Lots of RAM, no display attached.
• The production-control entry terminal
  The shared PC everyone touches — scan a barcode, pull up a part's data, track progress. The production-control software licence is a bit pricey, so putting it on everyone's own PC is not realistic. A cheap laptop.
• My own everyday PC
  For mail, editing documents on the file server, and reaching every PC above over remote desktop — so it has a whole row of big displays. Onboard GPU (a good one, though); a consumable machine.

Our CAD/CAM (OPEN MIND hyperMILL) network licence lists VPN use as "consult us", so while I technically could run hyperMILL on my own PC, the licence server is at the head office — running it on the workstation inside that network and driving it over remote desktop carries much less risk of the licence dropping mid-job.

There is also the classic systems-engineer arrangement: a beachhead PC of your own placed at a customer's site. Some problems at a customer only show themselves from inside their network — name resolution acting up, shares not visible, one line-of-business system mysteriously slow. Over the phone, from outside, the symptom simply will not reproduce. Keep one terminal of yours inside their network and you can do the first round of triage before moving: sometimes you avoid the trip entirely, and when you do go, you go knowing what to bring.

A common pain at mid-sized companies is the production-control entry terminal. Licences and seat counts are fixed by contract, so only a few PCs can enter data. Few terminals means a walk every time. Production stops so that production control can happen. At Shimada Kiko we reach the entry terminal over remote desktop from our own PCs. It has one screen, so one person operates it at a time — but the walking time is gone, and doing it from your seat makes a real difference. We stay within the "one entry terminal" licence condition, so it is not a licence violation.

Splitting PCs by role, in other words, is something necessity forced on us.

The problem was: "so how do we go touch them?" Every PC you split off adds another walk. The round trip to the CAM room gets old. The head office is an hour round trip just to show up.

With remote desktop, everything is within reach of the PC in front of me. No standing up. I enter the head-office network over VPN and operate the finance PC 20 km away, and it feels no different from sitting in the CAM room at the three-monitor workstation myself. Teleportation, basically. A near future where corporate networks cover the planet and electrons and light race through them — yet not so information-driven that nations and peoples have vanished. Perfect for a homebody like me.

But the remote desktops I used kept leaving me with three wishes. Tools existed that granted one of them each. Nothing granted all three — and I still have not found one that does.

1. Every key should reach the far machine

With Chrome Remote Desktop, the hankaku/zenkaku key (the Japanese IME toggle) does not work. You cannot switch Japanese input on or off during a session, so every time you need to type Japanese you reach for the mouse and click the IME icon in the far machine's tray. Hold ESC a moment too long and full screen collapses. With another tool, Alt+Tab lands on the local PC: you are working full-screen on the far machine, hit Alt+Tab, and it is your local windows that switch — a moment of "wait, what just happened?" every time — and switching apps on the far side means reaching for the mouse again. That never stopped being annoying.

2. The thin lines of CAD models and toolpaths should not blur

Remote-desktop pictures usually get compressed to save bandwidth, and that faint blur is a constant stress. With CAD and toolpaths you are looking for uncut stock and gouges at a very fine level of detail — blur is a problem, and it is indistinguishable from my aging eyes. Turn compression down or cap FPS for a cleaner picture and the traffic goes up, the latency hurts, and that is stressful too. (Some tools do get this right — fast software with excellent image quality does exist.)

3. The far machine's two or three monitors should stay two or three monitors on my side

This one was surprisingly hard to find done conveniently. Most tools either squeeze everything onto one screen or make you switch monitors with tabs. With tab switching, every time you drag a window across the far machine's monitors you have to switch views — genuinely painful to use.

So: chronically dissatisfied, with no better remote desktop in sight, I figured I might as well write one. That became ClearRemote. Since I am here, let me write up the parts I sweated over.

Take every key away from Windows — but recognize the ones I injected myself

A low-level keyboard hook (WH_KEYBOARD_LL) grabs everything — ESC, Alt+Tab, the Windows key, even F10 — before Windows sees it, and forwards it to the far machine. Keys are sent as scan codes, not virtual keys, so to the far PC they are indistinguishable from real keystrokes. That is why the IME toggle key arrives and Japanese input switches properly. Disconnect is Ctrl+Alt+Shift+Q — a four-key chord, because four keys collide with nothing that applications use.

Lossless by default — compression designed for text and thin lines

The screen is split into 128×128 tiles and only changed tiles are sent. Each tile's content picks its own encoding: a single flat color becomes "this one color"; four or eight colors become a palette; anything else goes as raw data. CAD drawings and office screens are "thin lines and text on white", so this split pays off enormously. On a thin connection it lowers quality automatically, and returns to lossless when the line frees up. Full-auto high quality that never bothers the user (me) — that is the point.

One window per remote monitor

Instead of squeezing everything into one local screen, each remote monitor gets its own local window. The key decision: the connection stays a single one. The far side keeps capturing all its monitors as one picture, and the local side merely divides up which window paints which region. So a drag across monitors just works, and an app straddling two screens is drawn straddling them. You can even line up your local bezels exactly where the remote bezels are.

The picture must not freeze while a big file transfers

The send queue has priority lanes: screen and input first, file transfer lowest. Measured in my automated tests: 24 MB sent in about 1.1 seconds while the screen kept updating throughout. After transfer, files are verified with SHA-256.

Signing in at the lock screen, remotely

When Windows Update reboots a machine overnight, you can find yourself locked out behind the lock screen. ClearRemote handles this by running as a service, placing a SYSTEM-privileged agent in the interactive session, and re-attaching its thread to whichever desktop is receiving input (Winlogon). Only SYSTEM can open the Winlogon desktop — administrator rights are not enough (tested). Chrome Remote Desktop can get into the lock screen too, so my old routine was to sign in via Chrome Remote Desktop first and then switch over to my own remote desktop. Not any more.

Other things I am quietly happy about

• One exe, 342 KB. No installer, no account registration.
  * 342 KB was the figure when this was posted. Features have kept landing almost daily; the current build is about 500 KB.
• No outside servers involved. The two PCs talk to each other directly.
• By default it only accepts connections from LAN and VPN addresses. Even if you accidentally open a router port, nobody gets in that way.
• Drop a new exe into the company shared folder and every PC's ClearRemote updates itself. No more remoting into each PC with Chrome Remote Desktop to run installers.
• File transfer is genuinely easy. To send: drag & drop onto the remote screen. To receive: Ctrl+C on the far machine, Ctrl+V on yours (the direction changes the gesture).

The free version is unlimited in time for non-commercial use (20-minute sessions, one connection at a time). Business use, per-monitor windows, and lock-screen operation need a licence: personal ¥4,800, corporate from ¥29,800. Both are one-time purchases, and moving up a tier later costs only the difference — never a repurchase.

A multitasking remote-desktop junkie in a small machine shop built this to make his own work easier. If it reaches people stuck the same way, that makes me happy.

Shimada Giken (Shimada Kiko, @shimadakiko_ofc) · the original post (August 7, 2026, in Japanese) · 日本語版

Download the free version Back to the product page