<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/">
  <channel>
    <title>Anveesa blog</title>
    <link>https://anveesa.com/blog</link>
    <atom:link href="https://anveesa.com/feed.xml" rel="self" type="application/rss+xml" />
    <description>Release stories, short guides and notes on how we build Anveesa.</description>
    <language>en</language>
    <lastBuildDate>Sun, 15 Nov 2026 00:00:00 GMT</lastBuildDate>
    <item>
      <title><![CDATA[Meet Anveesa Labs, and the browser we're building in it]]></title>
      <link>https://anveesa.com/blog/meet-anveesa-labs</link>
      <guid isPermaLink="true">https://anveesa.com/blog/meet-anveesa-labs</guid>
      <pubDate>Sun, 15 Nov 2026 00:00:00 GMT</pubDate>
      <author>pandhuw58@gmail.com (Pandhu Wibowo)</author>
      <description><![CDATA[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.]]></description>
      <content:encoded><![CDATA[<p>Every tool we ship started as a question we couldn&#39;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.</p>
<p>That leaves out the most interesting part. So we made a place for it: <a href="https://anveesa.com/labs">Anveesa Labs</a>.</p>
<h2>What goes in Labs</h2>
<p>A Labs project is an experiment, not a product. It starts with one question we&#39;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&#39;d rather you read it once it&#39;s worth reading.</p>
<p>Each one tells you four things plainly:</p>
<ol>
<li><strong>Its stage</strong>, and when it started.</li>
<li><strong>What works today.</strong></li>
<li><strong>What doesn&#39;t work yet.</strong> This list matters as much as the first one.</li>
<li><strong>Whether you should rely on it.</strong> 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.</li>
</ol>
<p>If an experiment&#39;s answer is good, it can grow into an Anveesa product with docs and support. If it isn&#39;t, we&#39;ll say so and stop. Both are useful outcomes.</p>
<h2>The first experiment: Broxee</h2>
<p>Broxee is a web browser with its own engine, written in Rust. The question behind it is simple to say and hard to answer:</p>
<blockquote>
<p>How much of a modern browser can you build when the whole app has to stay under 60 MB, on disk and in memory?</p>
</blockquote>
<p>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.</p>
<p>Here is where it stands, from the project&#39;s own measurements:</p>
<ul>
<li>The whole app is <strong>3.7 MB</strong> on disk.</li>
<li>Drawing a Wikipedia article peaks at <strong>15.7 MB</strong> of memory.</li>
<li>Drawing the Hacker News front page peaks at <strong>14.0 MB</strong>.</li>
</ul>
<p>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 <code>rustls</code>. Hand-written cryptography puts people at risk, and that&#39;s not a trade we&#39;ll make for a smaller binary.</p>
<h2>What it does today</h2>
<p>Broxee already works as a daily browser for a lot of the web:</p>
<ul>
<li>Tabs sleep to disk and wake without the network, back where you left them. Sleeping tabs use no page memory, however many you have.</li>
<li>Profiles keep their own cookies, history and tabs, so a login at work never reaches your personal profile. Third-party cookies are blocked.</li>
<li>Split view shows two or four pages side by side, each pane in its own profile.</li>
<li>One keyboard palette (⌘K) reaches every tab, bookmark, history entry and command.</li>
<li>Pages get flex, grid, floats, tables, web fonts, gradients, shadows, 2D transforms, forms, and JavaScript with promises, async functions and <code>fetch</code>.</li>
</ul>
<h2>What it doesn&#39;t do yet</h2>
<ul>
<li>The window runs on macOS only. On Linux, Broxee renders pages to images from the command line.</li>
<li>Some sites use scripts or CSS the engine skips, and they look wrong.</li>
<li>The builds aren&#39;t signed.</li>
</ul>
<p>The code is public at <a href="https://github.com/PandhuWibowo/broxee">github.com/PandhuWibowo/broxee</a>. It doesn&#39;t have a license file yet, so for now it&#39;s readable but not formally open source; we&#39;ll fix that before asking anyone to build on it.</p>
<h2>Following along</h2>
<p>Broxee is on the <a href="https://anveesa.com/open-source">open-source page</a> 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.</p>
<p>If you try Broxee and a page breaks, that&#39;s exactly the kind of report we want. <a href="https://github.com/PandhuWibowo/broxee/issues">Open an issue</a> with the address, or <a href="mailto:pandhuw58@gmail.com?subject=Anveesa%20Labs">write to us</a>.</p>
]]></content:encoded>
      <category>labs</category>
      <category>broxee</category>
      <category>open-source</category>
    </item>
    <item>
      <title><![CDATA[What it takes to keep a web browser under 60 MB]]></title>
      <link>https://anveesa.com/blog/keeping-a-browser-under-60-mb</link>
      <guid isPermaLink="true">https://anveesa.com/blog/keeping-a-browser-under-60-mb</guid>
      <pubDate>Sun, 15 Nov 2026 00:00:00 GMT</pubDate>
      <author>pandhuw58@gmail.com (Pandhu Wibowo)</author>
      <description><![CDATA[Notes from Broxee's budget log, on making DOM nodes smaller, skipping CSS rules a page can't use, and finding memory the allocator was holding on to.]]></description>
      <content:encoded><![CDATA[<p><a href="https://anveesa.com/labs#broxee">Broxee</a> is a browser we&#39;re building in Labs with one rule: the app has to stay under 60 MB, on disk and in memory. The rule is enforced by a script, not by good intentions. It builds the release binary, generates a stress page of 20,000 rows, lays it out and paints it at 2x (a Retina frame), and fails if the binary or the peak memory reaches 60 MB.</p>
<p>Every change that touches parsing, layout or painting gets a line in the project&#39;s budget log, with its numbers. Most of what follows comes straight from that log. These are the changes that moved the numbers most.</p>
<h2>The DOM was twice as big as it needed to be</h2>
<p>The first DOM was the obvious one: each element held a <code>Vec</code> of children and owned its strings. On the 2.6 MB stress page it peaked at <strong>55.8 MB</strong>, before any styling or layout. That alone would have used up the whole budget.</p>
<p>The next version changed four things at once:</p>
<ul>
<li>Children became a linked list instead of a vector per node.</li>
<li>Nodes refer to each other by <code>u32</code> ids instead of pointers.</li>
<li>Tag and attribute names are interned, so <code>div</code> is stored once, not 20,000 times.</li>
<li>The tokenizer hands out borrowed slices of the source instead of copying strings.</li>
</ul>
<p>Peak memory on the same page dropped to <strong>27.0 MB</strong>. A later pass shrank each node to 28 bytes and moved text and attribute values into shared buffers. The lesson repeated through the whole project: the first data structure you reach for is usually the one that costs the most.</p>
<h2>Most of a site&#39;s CSS doesn&#39;t apply to the page you&#39;re on</h2>
<p>Sites ship one stylesheet for all their pages. GitHub&#39;s home page loads 30,309 CSS rules. Parsing and matching every one of them is a lot of work and a lot of memory for rules that can never match.</p>
<p>So before Broxee parses a stylesheet, it collects every tag, id and class that actually appears in the document. A rule whose selectors all need a name the page doesn&#39;t have is skipped without parsing its declarations. On GitHub, <strong>28,170 of the 30,309 rules</strong> are skipped. Memory for that page fell from 43.7 MB to <strong>25.3 MB</strong> before painting, and from 60.2 MB to <strong>40.5 MB</strong> painted at 2x.</p>
<p>We checked that this changes nothing you can see: computed styles on GitHub, Wikipedia, Hacker News and MDN were compared node by node with and without the shortcut, and they were identical.</p>
<h2>The allocator was keeping memory we&#39;d already freed</h2>
<p>This one took the longest to find. After visiting GitHub and then example.com, the macOS <code>heap</code> tool counted only <strong>6.8 MB</strong> of live allocations, yet the process held <strong>19 to 25 MB</strong> more in small-allocation regions. Those were blocks Broxee had freed, which macOS malloc kept in its per-CPU caches.</p>
<p>Asking the allocator to give pages back (<code>malloc_zone_pressure_relief</code>) changed nothing. Starting the process with <code>MallocSpaceEfficient=1</code> did. For the same sequence, idle memory went from <strong>43 to 29 MB</strong> and the peak from <strong>59 to 48 MB</strong>, with parse, cascade, layout and paint times unchanged across three runs.</p>
<p>libmalloc reads that variable only when a process starts, so Broxee restarts itself once with it set, which takes about 10 ms. It isn&#39;t documented in <code>man malloc</code>. If a future macOS ignores it, memory goes back to the old numbers and nothing else breaks, which is the kind of failure we can live with.</p>
<h2>Tabs that don&#39;t live in memory</h2>
<p>A browser that keeps every tab loaded can&#39;t stay under 60 MB. In Broxee only the visible tab is a live page. The others sleep: the HTML and stylesheets exactly as they arrived (usually still compressed) and the compressed images go to a private file on disk. GitHub&#39;s sleeping tab takes 2.5 MB, a Wikipedia article 1.2 MB, MDN 57 KB.</p>
<p>Selecting a tab rebuilds it from that file without the network, at its old scroll position, in 5 to 45 ms. And because sleeping tabs share stylesheets, scripts, fonts and images, six tabs (three Wikipedia articles, two GitHub repositories and example.com) take <strong>1.7 MB</strong> on disk instead of 2.9 MB.</p>
<h2>JavaScript, parsed only when it runs</h2>
<p>Broxee&#39;s JavaScript interpreter is written in the repository too. Many sites ship far more script than they run on a given page, so the most recent change parses a function&#39;s body only the first time it&#39;s called. On a site with 868 KB of Next.js script, peak memory from the command line dropped from <strong>40.3 to 27.0 MB</strong>, and in a window from 63 to 52 MB.</p>
<h2>Where it stands</h2>
<p>As of today the binary is <strong>3.7 MB</strong>, Wikipedia peaks at 15.7 MB, and Hacker News at 14.0 MB. Four panes side by side, showing GitHub, Wikipedia, Hacker News and example.com, peak at 47 MB.</p>
<p>Most of these tricks aren&#39;t new. Production engines bucket CSS rules, filter selectors with bloom filters and share styles between siblings too. What a hard budget changes is that you can&#39;t postpone them. Every feature has to fit, so every feature gets measured the day it lands.</p>
<p>The full budget log, with every change and its numbers, is in the <a href="https://github.com/PandhuWibowo/broxee#budget-log">Broxee README</a>. If you know a trick we haven&#39;t tried, <a href="https://github.com/PandhuWibowo/broxee/issues">open an issue</a>.</p>
]]></content:encoded>
      <category>labs</category>
      <category>broxee</category>
      <category>rust</category>
      <category>performance</category>
    </item>
    <item>
      <title><![CDATA[anveesa.com now speaks Korean, Japanese, Chinese and Arabic]]></title>
      <link>https://anveesa.com/blog/anveesa-in-five-more-languages</link>
      <guid isPermaLink="true">https://anveesa.com/blog/anveesa-in-five-more-languages</guid>
      <pubDate>Sun, 15 Nov 2026 00:00:00 GMT</pubDate>
      <author>pandhuw58@gmail.com (Pandhu Wibowo)</author>
      <description><![CDATA[The site is now in five more languages, including right-to-left Arabic. What's translated, what's still English, and how it works.]]></description>
      <content:encoded><![CDATA[<p>Not everyone who runs our tools reads English first. Starting today, anveesa.com is also in five more languages:</p>
<ul>
<li><strong>한국어</strong> at <a href="https://anveesa.com/ko">/ko</a></li>
<li><strong>日本語</strong> at <a href="https://anveesa.com/ja">/ja</a></li>
<li><strong>简体中文</strong> at <a href="https://anveesa.com/zh-cn">/zh-cn</a></li>
<li><strong>繁體中文</strong> at <a href="https://anveesa.com/zh-tw">/zh-tw</a></li>
<li><strong>العربية</strong> at <a href="https://anveesa.com/ar">/ar</a></li>
</ul>
<p>Use the globe button in the navigation bar to switch. It opens the page you&#39;re on in the language you pick, and keeps the section you were reading.</p>
<h2>What&#39;s translated, and what isn&#39;t yet</h2>
<p>This first round covers the parts of the site most people see first: the navigation and footer, the home page with its FAQ, <a href="https://anveesa.com/labs">Labs</a> and the <a href="https://anveesa.com/open-source">open-source page</a>.</p>
<p>Still in English for now: the Timika and use-case pages, the blog, and the product docs. The docs are most of the text on the site, so they&#39;ll take a few more rounds. Those pages still open under each language&#39;s address, so no link breaks, but they show English text and stay out of search results until they&#39;re translated.</p>
<p>Some things will stay in English on purpose:</p>
<ul>
<li><strong>Screenshots, recordings and product mockups</strong>, because they show the real apps, and the apps are in English.</li>
<li><strong>Commands, code and configuration</strong>, because that&#39;s what you type into your terminal.</li>
<li><strong>Product names</strong>, like Vestra, Nias and Timika.</li>
</ul>
<p>These translations are new, and nobody whose first language is Korean, Japanese, Chinese or Arabic has reviewed them yet. If a sentence reads wrong or a technical term isn&#39;t the one you&#39;d use, <a href="mailto:pandhuw58@gmail.com?subject=Translation">tell us</a>. Corrections are a one-line change, and we&#39;d rather fix them than guess.</p>
<h2>Arabic reads right to left</h2>
<p>Arabic isn&#39;t just translated text in the same layout. The whole page mirrors: the navigation starts from the right, lists and borders flip, and arrows point the way you read. Code blocks and product mockups stay left to right inside the Arabic page, since that&#39;s how they appear in the apps and your terminal.</p>
<p>To make that possible, we rewrote the site&#39;s CSS to describe positions by reading direction (start and end) instead of left and right. Arabic letters also join to each other, so letter spacing, which the English headlines use, is switched off for Arabic: spacing would pull the letters apart.</p>
<h2>Fonts for every script</h2>
<p>Our headline and body fonts, Fraunces and Figtree, only cover Latin letters. For the new languages, the Latin parts of a sentence (product names, numbers) keep those fonts, and everything else falls through to a matching typeface:</p>
<ul>
<li>Noto Serif and Noto Sans for Korean, Japanese, and Simplified and Traditional Chinese, so headlines keep a soft serif and body text stays easy to read.</li>
<li>Noto Naskh Arabic for headlines and IBM Plex Sans Arabic for body text. Our code font is IBM Plex Mono, so the Arabic text and the code come from the same type family.</li>
</ul>
<p>A few smaller details: Chinese, Japanese and Korean have no italic, so emphasis keeps its color but isn&#39;t slanted. Korean lines break between words, never inside one.</p>
<h2>For search engines</h2>
<p>Each translated page has its own title and description in its language. Every page links to its other language versions, so a search engine can show Korean readers the Korean page. A translation enters the sitemap only once it&#39;s done.</p>
<h2>What&#39;s next</h2>
<p>The docs come next, product by product, then the blog. If there&#39;s a page you&#39;d like translated first, or a language you&#39;d like us to add, <a href="mailto:pandhuw58@gmail.com?subject=Translation">write to us</a>.</p>
]]></content:encoded>
      <category>news</category>
      <category>website</category>
    </item>
    <item>
      <title><![CDATA[Hello from the people behind Anveesa]]></title>
      <link>https://anveesa.com/blog/hello-from-anveesa</link>
      <guid isPermaLink="true">https://anveesa.com/blog/hello-from-anveesa</guid>
      <pubDate>Wed, 04 Nov 2026 00:00:00 GMT</pubDate>
      <author>pandhuw58@gmail.com (Pandhu Wibowo)</author>
      <description><![CDATA[Why we started a blog, what we build, and what you can expect to read here.]]></description>
      <content:encoded><![CDATA[<p>Welcome to the Anveesa blog. Until now the only way to follow what we were doing was to read commit messages and release notes. That works if you already live in our repositories, but most people who run our tools don&#39;t, and they shouldn&#39;t have to.</p>
<p>So this is where we&#39;ll write things down in plain words: what changed, why we changed it, and how you can get the most out of it.</p>
<h2>What Anveesa is</h2>
<p>Anveesa is a small family of tools for the infrastructure you already own. Each one installs and runs on its own, on your machine or your server, and none of them needs an account with us.</p>
<ul>
<li><strong><a href="https://anveesa.com/docs/vestra">Vestra</a></strong> puts every storage bucket in one calm interface: GCS, S3, R2, MinIO, Azure Blob and more, as a desktop app or on your own server.</li>
<li><strong><a href="https://anveesa.com/docs/nias">Nias</a></strong> takes you from a SQL query to a dashboard your team can export, share or embed.</li>
<li><strong><a href="https://anveesa.com/docs/aras">Aras</a></strong> reviews every database change and keeps a trail of who ran what.</li>
<li><strong><a href="https://anveesa.com/docs/mikko">Mikko</a></strong> is a SOC platform that sits on top of Wazuh for hunting, correlation and response.</li>
<li><strong><a href="https://anveesa.com/docs/cli">Anveesa CLI</a></strong> gives you one command for every AI model, right in your terminal.</li>
<li><strong><a href="https://anveesa.com/docs/workflow">Workflow</a></strong> is a free infinite canvas for drawing flows with your team.</li>
</ul>
<p>Vestra, Nias, the CLI and Workflow are open source. Aras and Mikko are licensed to teams.</p>
<h2>Why self-hosted</h2>
<p>Your databases, buckets and logs are some of the most sensitive things you have. We think the tools that touch them should run where they live, keep credentials on your side, and keep working when someone else&#39;s service has a bad day.</p>
<p>That choice shapes everything we build:</p>
<ol>
<li><strong>One binary or one container.</strong> If installing a tool takes an afternoon, it won&#39;t get installed.</li>
<li><strong>Your data stays put.</strong> Credentials are encrypted at rest and never leave your machine.</li>
<li><strong>Calm by default.</strong> Show what needs you, and keep everything else one click away.</li>
</ol>
<h2>What you&#39;ll read here</h2>
<p>We&#39;ll keep posts practical. Expect things like:</p>
<ul>
<li>Release notes with the story behind them, not just a changelog.</li>
<li>Short guides, such as connecting Vestra to a private MinIO server or sharing a Nias dashboard safely.</li>
<li>Notes on how we build: the trade-offs, the mistakes and what we learned from them.</li>
</ul>
<p>If you want to try something today, the quickest start is one command:</p>
<pre><code class="language-bash">docker run -d -p 80:80 \
  -e JWT_SECRET=change-me \
  -e ENCRYPTION_KEY=change-me-too \
  -v anveesa-data:/data \
  pandhu612/anveesa-vestra:latest
</code></pre>
<p>Then open <code>http://localhost</code> and create your owner account.</p>
<h2>Say hello</h2>
<p>If there&#39;s something you&#39;d like us to write about, or a tool you wish worked differently, <a href="mailto:pandhuw58@gmail.com">send us an email</a> or open an issue on <a href="https://github.com/PandhuWibowo">GitHub</a>. We read everything.</p>
<p>Thanks for being here.</p>
]]></content:encoded>
      <category>news</category>
      <category>self-hosting</category>
    </item>
  </channel>
</rss>
