Software, rebuilt from the machine up.
What if an app carried only itself? No browser. No framework. No 300 MB of machinery. Just your idea, compiled and running natively on the Yantra engine.
Every era added a layer. Nobody took one away.
- 1991Pages. HTML shares documents.
- 1995Scripts. JavaScript makes pages move.
- 2004Web apps. Mail, maps and docs move into the browser.
- 2010Frameworks. Countless JavaScript frameworks, libraries, virtual DOMs and build tools, each one a layer on the last.
- 2013A browser in every app. Desktop apps become web pages, each carrying its own browser.
- TodayAI writes the apps. The app itself is the little sketch on top. Everything under it is weight.
The web was built for pages. Yantra is built for everything after.
Your chat app is carrying a whole browser.
So is your editor. So is your music player. Each ships its own copy of Chromium, 364 MB of machinery to draw a few buttons, and your laptop runs a dozen of them at once.
Measured on the founder's laptop, 25 Sep 2026: Chrome, 40 processes, 3.8 GB · a chat desktop app, 15 processes, 2.5 GB · a code editor, 6 processes, 1.0 GB.
Guess how much of this is the actual app.
We took a shipping Electron app apart, file by file, and measured every piece.
1.7 KB. The little sketch on top. The other 364.3 MB is machinery.
Electron 43 dist folder. Every size is measured; the bars beside the labels are to scale.
Keep the app. Lose the machine.
Everything else turns to dust. What's left is one compiled module: the whole app.
Runs today · examples/counter.wasmBack to the limits of the machine.
Games still fight for every frame. Apps stopped trying. Yantra compiles your app ahead of time and runs it natively on the Yantra engine, drawn straight on the GPU. No browser to boot. No DOM. No virtual DOM.
Today the engine still draws through a temporary Qt bridge; our own GPU renderer replaces it next. Speed and memory results will be published with the raw data.
Ten apps. One engine.
Yantra's shared engine is 112 MB today, mostly a temporary Qt bridge and Unicode data. It is paid once; each app adds about 87 KB. The engine shrinks as the bridge is replaced.
One language for everything on a screen.
Yantra is built on Nim: typed, compiled, readable. Everything is imported for you. State, data, motion, documents, sheets, film and physics are part of the language, not libraries you glue together. Tap a marked line to see what it does.
From a button to a film, one language.
Every app asks first.
A normal desktop app can read your files the moment it starts. A Yantra app can't touch your files, camera or network until you say yes, with one line in its manifest.
In developmentChange the app while you use it.
- Live. Describe a change and AI rewrites the running app in place. No rebuild, no reload, nothing you were doing is lost.
- Checked. Every edit is type-checked and tested before it lands. When something breaks, Yantra finds the error and fixes it.
- Still yours. The code stays short enough to read, so you can change anything by hand. AI helps. You never depend on it.
Apps today. Documents, sheets, films and worlds next.
Written with AI. Checked by machines.
Every change passes the full check before it lands. Tests can't be weakened or deleted. When an AI agent faked a result, the check caught it and we logged it.
Now, next, and everything after.
Now
The language core, native and web runners, 87 KB apps, the timeline engine.
Next
The sandbox, lists and components, video export, our own renderer without Qt.
Later
Yantra Office, games and 3D, phones and headsets, the AI app platform.
Build what comes after the web.
The first builders shape the language. Get in before everyone else.