Meet Anveesa Labs, and the browser we're building in it
Labs is where we build things to find out whether they should exist. The first experiment is Broxee, a web browser with its own engine, kept under 60 MB.
Pandhu Wibowo November 15, 2026 3 min read
Every tool we ship started as a question we couldn't stop thinking about. Until now those questions lived in private repositories and half-finished branches, and you only saw the ones that turned into products.
That leaves out the most interesting part. So we made a place for it: Anveesa Labs.
What goes in Labs
A Labs project is an experiment, not a product. It starts with one question we'd rather answer by building than by guessing. Some experiments are open source from the first commit. Others we only describe, because the code is still rough and we'd rather you read it once it's worth reading.
Each one tells you four things plainly:
- Its stage, and when it started.
- What works today.
- What doesn't work yet. This list matters as much as the first one.
- Whether you should rely on it. Usually the answer is no. Labs projects change quickly and can break between versions. Run your team on the products, and use Labs to explore.
If an experiment's answer is good, it can grow into an Anveesa product with docs and support. If it isn't, we'll say so and stop. Both are useful outcomes.
The first experiment: Broxee
Broxee is a web browser with its own engine, written in Rust. The question behind it is simple to say and hard to answer:
How much of a modern browser can you build when the whole app has to stay under 60 MB, on disk and in memory?
Broxee treats 60 MB as a hard rule. A script in the repository builds the app, opens a stress page of 20,000 rows, draws it at Retina size, and fails the build if either the binary or the peak memory reaches the limit.
Here is where it stands, from the project's own measurements:
- The whole app is 3.7 MB on disk.
- Drawing a Wikipedia article peaks at 15.7 MB of memory.
- Drawing the Hacker News front page peaks at 14.0 MB.
Almost everything is written in the repository: the HTML parser, the CSS cascade, layout, painting, image decoders for PNG, JPEG, GIF, WebP and SVG, a font reader with WOFF2 support, and a JavaScript interpreter. The one exception is TLS, which uses rustls. Hand-written cryptography puts people at risk, and that's not a trade we'll make for a smaller binary.
What it does today
Broxee already works as a daily browser for a lot of the web:
- Tabs sleep to disk and wake without the network, back where you left them. Sleeping tabs use no page memory, however many you have.
- Profiles keep their own cookies, history and tabs, so a login at work never reaches your personal profile. Third-party cookies are blocked.
- Split view shows two or four pages side by side, each pane in its own profile.
- One keyboard palette (⌘K) reaches every tab, bookmark, history entry and command.
- Pages get flex, grid, floats, tables, web fonts, gradients, shadows, 2D transforms, forms, and JavaScript with promises, async functions and
fetch.
What it doesn't do yet
- The window runs on macOS only. On Linux, Broxee renders pages to images from the command line.
- Some sites use scripts or CSS the engine skips, and they look wrong.
- The builds aren't signed.
The code is public at github.com/PandhuWibowo/broxee. It doesn't have a license file yet, so for now it's readable but not formally open source; we'll fix that before asking anyone to build on it.
Following along
Broxee is on the open-source page now, with its stars, forks and contributors next to Nias and Vestra. Anyone who sends a change to a public Labs experiment is listed there with everyone else.
If you try Broxee and a page breaks, that's exactly the kind of report we want. Open an issue with the address, or write to us.
- labs
- broxee
- open-source