ProBackend
agentic keyboards input
15 hours ago5 min read

Neverclick's Agentic Labs Approach to Mouse-Free Computing Works Everywhere

Built by an RSI sufferer, Neverclick uses a lightweight local computer-vision model to detect UI elements and perform mouse actions from the keyboard — no accessibility APIs, no cloud AI, works in every Windows app.

Neverclick's Agentic Labs Approach to Mouse-Free Computing Works Everywhere

Here's the thing about keyboard-driven mouse replacements: most of them suck. Not because the idea is bad — it's brilliant, really — but because they're held back by the same old problem. They rely on operating system accessibility APIs, and those APIs only work in apps that bother to expose their UI elements through them.

So you've got this keyboard shortcut that works perfectly in Chrome, great in Word, and absolutely nothing in half the other apps you actually use. Frustrating? Yeah.

Neverclick sidesteps this entirely. Instead of asking apps nicely for their UI structure, it just... looks at your screen. A lightweight computer vision model runs locally on your machine, detects UI elements in real-time, and lets you interact with them using just your keyboard. No APIs required. No app-specific integrations needed.

It's the kind of agentic labs thinking that makes you wonder why nobody built this sooner.

Why Accessibility APIs Hold You Back

Let me paint you a picture. You're setting up your favorite keyboard-driven mouse replacement tool. You configure it, test it out, and suddenly you hit the wall — that one app your team uses daily that doesn't expose accessibility data. Or the game you play after work. Or whatever legacy internal tool your company refuses to modernize.

The problem isn't the tool. It's that most of these solutions are built on top of OS accessibility APIs — things like Microsoft's UI Automation framework. These APIs work great when app developers actually implement them properly. But too often, they don't. Or they implement them incompletely. Or the app uses custom drawing that bypasses the accessibility layer entirely.

Result? You've got a keyboard shortcut that works in 60% of your apps and leaves you reaching for the mouse the other 40%. Which kind of defeats the whole purpose, doesn't it?

This is exactly the problem Neverclick's creator Lazo ran into. And instead of accepting it, they built something different.

How the Vision Model Actually Works

Here's where it gets interesting. Neverclick doesn't ask your apps for permission to see their UI elements. It just... sees them.

The tool runs a quick, lightweight computer vision model entirely on your local machine. It scans what's currently displayed on your screen — buttons, text fields, menus, icons — and builds a mental map of where everything is. Then it overlays hints on top of those elements, showing you which key to press to interact with them.

No cloud AI. No ChatGPT. No Claude. Nothing leaves your computer. Everything runs locally, which means it works whether you're on a corporate network with strict firewalls or sipping coffee at a café with sketchy Wi-Fi.

The accuracy is pretty impressive, though Lazo acknowledges it can occasionally hallucinate — misidentify an element or suggest the wrong interaction. But for day-to-day use, it's reliable enough that most people stop reaching for the mouse almost entirely.

Performance on Real Hardware (Not Just Benchmarks)

Here's what really sold me: Neverclick is designed for office laptops, not gaming rigs or workstations. The developer specifically optimized it to run smoothly on hardware that's ten years old.

On a typical 1080p monitor, it uses about 200MB of RAM and takes up roughly 40MB of disk space. That's nothing, honestly. You could run this alongside a dozen Chrome tabs and not notice the difference.

The hints appear instantly. I mean instantly — there's no perceptible delay between pressing a key and seeing the visual feedback. That responsiveness is crucial because if there's any lag, you'll end up reaching for the mouse out of habit before the tool can react.

And it handles multiple monitors, 4K displays, and even 8K setups without breaking a sweat. The hint labels adapt to your keyboard's language too, so if you're typing in another language, the visual cues match.

Privacy and Local-First Design

Let's talk about privacy for a minute, because it matters.

Everything Neverclick does happens on your machine. Your screen data never leaves your computer. There are no accounts to create, no subscriptions to manage, no trials that expire and leave you hanging.

The FAQ is pretty blunt about this: "Neverclick's computer vision system runs entirely on your own machine. It's not built on ChatGPT, Claude, or any online AI service." And "Everything runs locally on your machine. Nothing you do ever leaves your computer."

In an era where every other "AI-powered" tool wants to phone home to some cloud service, this kind of commitment to local processing feels almost radical. You're not training someone else's model on your screen contents. You're not sending your workflow data to a server farm. It's just you and your computer, working together.

The RSI Origin Story

Lazo built Neverclick while struggling with repetitive strain injury. Recovered years ago, but the tool stayed. Now they use it every day and can't imagine using a computer without it.

That personal connection shows in the design choices. The keybinds prioritize letters in the center of the keyboard — QWERTYUIOP, ASDFGHJKL, ZXCVBNM — because that's where your fingers naturally rest. Less reaching means less strain.

There's an interactive virtual keyboard for configuring keybinds, which makes customization intuitive even if you're not a power user. And the whole thing is free forever. No premium tier, no feature gating, no "buy me a coffee" that actually gates functionality.

It's the kind of tool that feels like it was built by someone who actually needed it, not a startup trying to find product-market fit.

Current Limitations and What's Next

Right now, Neverclick is Windows only. Linux and macOS aren't supported yet, though the developer has mentioned them as future goals.

The installation requires admin-level mouse actions, which is why it installs to Program Files — Windows needs that for the accessibility tool registration. It's not portable, which might bother some users who prefer tools they can carry on a USB drive.

But honestly, these feel like growing pains for a solo developer working on something this ambitious. The core technology works, it's free, and it solves a real problem in a way that feels genuinely innovative.

If you're tired of keyboard mouse replacements that only work in half your apps, Neverclick might be worth a look. It's free, it's local, and it just... works.

Read more about Neverclick on the official site | Original coverage at The Register | Explore Agentic Keyboards & Input

Neverclick's Agentic Labs Approach to Mouse-Free Computing Works Everywhere

More blogs