<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Constantin Alex</title>
    <description>The latest articles on DEV Community by Constantin Alex (@constantin_alex_05c9c24b7).</description>
    <link>https://dev.to/constantin_alex_05c9c24b7</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4173475%2F4145fc68-593c-4f23-a0f2-1a8027584705.png</url>
      <title>DEV Community: Constantin Alex</title>
      <link>https://dev.to/constantin_alex_05c9c24b7</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/constantin_alex_05c9c24b7"/>
    <language>en</language>
    <item>
      <title>OpenKeyword is live: 389k keywords, 5 markets, and what I cut to ship it</title>
      <dc:creator>Constantin Alex</dc:creator>
      <pubDate>Thu, 19 Nov 2026 23:08:51 +0000</pubDate>
      <link>https://dev.to/constantin_alex_05c9c24b7/openkeyword-is-live-389k-keywords-5-markets-and-what-i-cut-to-ship-it-24eh</link>
      <guid>https://dev.to/constantin_alex_05c9c24b7/openkeyword-is-live-389k-keywords-5-markets-and-what-i-cut-to-ship-it-24eh</guid>
      <description>&lt;p&gt;&lt;a href="https://openkeyword.org" rel="noopener noreferrer"&gt;OpenKeyword&lt;/a&gt; opened today. It's a free keyword research tool: you type a keyword and get a 0 to 100 popularity score, the search intent, a 12-month trend and up to 50 related ideas. No account for 30 lookups a day.&lt;/p&gt;

&lt;p&gt;I wrote earlier about seeding 305k keywords. This is the launch-day&lt;br&gt;
follow-up.&lt;/p&gt;
&lt;h3&gt;
  
  
  What shipped
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;About 389,000 scored keywords in the US, UK, Germany, France and Spain. The German, French and Spanish sets are smaller (20 niches each instead of 42) and cost about $3.50 per market.&lt;/li&gt;
&lt;li&gt;The website in 8 languages, with a market picker. Picking a country changes the data, and the language setting changes only the interface.&lt;/li&gt;
&lt;li&gt;Optional accounts with email-code login and no passwords. An account raises the limit from 30 lookups a day to 200.&lt;/li&gt;
&lt;li&gt;The Chrome extension, approved by Google yesterday. It's contributor-only: it shares the words you search after you opt in, and shows you nothing while you browse.&lt;/li&gt;
&lt;/ul&gt;
&lt;h3&gt;
  
  
  What I cut
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;Exact search volumes. The raw Google Ads numbers stay on my home server. The public score is a relative 0 to 100 rank against the other keywords in the same market.&lt;/li&gt;
&lt;li&gt;Intent outside English. The intent labels (how-to, best, vs, question) come from English patterns, so most German, French and Spanish keywords say "Other" for now.&lt;/li&gt;
&lt;li&gt;Canada, Australia and Romania. They're in the market picker as "coming soon" because there's no data yet, and I'd rather show that than an empty result.&lt;/li&gt;
&lt;/ul&gt;
&lt;h3&gt;
  
  
  The last-hour bugs
&lt;/h3&gt;

&lt;p&gt;I did a final check an hour before deploying. Every test and build passed: 124 API tests, 59 web, 77 home server, 71 extension. Then I opened the site at phone width and found two pages that scrolled sideways.&lt;br&gt;
The first was a tooltip. The "Estimate ?" badge on the score card opens a 293px box anchored to the badge's left edge. On a 375px phone the badge sits 165px from the left, so the box ended at 458px. It was hidden with &lt;code&gt;visibility: hidden&lt;/code&gt;, but hidden elements still count for scroll width, so the whole page could be dragged sideways. On small screens the tooltip now anchors to the full label row instead of the badge.&lt;/p&gt;

&lt;p&gt;The second was a CSS grid classic:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight css"&gt;&lt;code&gt;&lt;span class="nc"&gt;.layout&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="nl"&gt;grid-template-columns&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;1fr&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;code&gt;1fr&lt;/code&gt; means &lt;code&gt;minmax(auto, 1fr)&lt;/code&gt;, and &lt;code&gt;auto&lt;/code&gt; won't shrink below the content's minimum width. A three-column table with &lt;code&gt;min-width: 130px&lt;/code&gt; cells wanted 390px, so the column grew to 390px, even though the table already sat inside its own &lt;code&gt;overflow-x: auto&lt;/code&gt; wrapper. The fix:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight css"&gt;&lt;code&gt;&lt;span class="nc"&gt;.layout&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="nl"&gt;grid-template-columns&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="nf"&gt;minmax&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mi"&gt;1fr&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Neither bug shows up in tests. I only found them by loading every page at 375px and comparing &lt;code&gt;scrollWidth&lt;/code&gt; to &lt;code&gt;clientWidth&lt;/code&gt;, which I'll keep as a pre-deploy script.&lt;/p&gt;

&lt;h3&gt;
  
  
  How it deployed
&lt;/h3&gt;

&lt;p&gt;Both Workers went up with &lt;code&gt;wrangler versions upload&lt;/code&gt; first, so the new code sat on Cloudflare without traffic, and a timer promoted them with &lt;code&gt;wrangler versions deploy &amp;lt;id&amp;gt;@100%&lt;/code&gt; at the planned minute. Rolling back is the same command with the old version id.&lt;/p&gt;

&lt;h3&gt;
  
  
  What I'd like to hear
&lt;/h3&gt;

&lt;p&gt;If you do keyword research for small sites, which market or feature would make you come back? And if you've grown a dataset from user contributions, how did you get the first thousand people to share?&lt;/p&gt;

</description>
      <category>showdev</category>
      <category>webdev</category>
      <category>seo</category>
      <category>privacy</category>
    </item>
    <item>
      <title>3 days before launch, and my free keyword tool isn't finished</title>
      <dc:creator>Constantin Alex</dc:creator>
      <pubDate>Tue, 17 Nov 2026 16:49:51 +0000</pubDate>
      <link>https://dev.to/constantin_alex_05c9c24b7/3-days-before-launch-and-my-free-keyword-tool-isnt-finished-33j5</link>
      <guid>https://dev.to/constantin_alex_05c9c24b7/3-days-before-launch-and-my-free-keyword-tool-isnt-finished-33j5</guid>
      <description>&lt;p&gt;OpenKeyword opens on 20 November. That’s in three days, and I’m not done. Here’s where it stands.&lt;/p&gt;

&lt;p&gt;What OpenKeyword is&lt;/p&gt;

&lt;p&gt;It’s a free keyword research tool. You type a keyword and get a popularity score from 0 to 100, the search intent (how-to, best, vs, question), a 12-month trend and up to 50 related ideas. The interesting part is where the data comes from, which I’ll get to below.&lt;/p&gt;

&lt;p&gt;What’s done&lt;/p&gt;

&lt;p&gt;Security is the part I’m proudest of, because the whole idea falls apart if people don’t trust it. I ran a full audit of the codebase and fixed what it found. Every page and every API response now has a strict Content Security Policy and the usual security headers. IP addresses are never stored: the server keeps a keyed hash for rate limits and deletes it within 48 hours. You log in with an email code, so there are no passwords to leak. CSV exports are guarded against formula injection, which I’d never heard of until the audit flagged it.&lt;/p&gt;

&lt;p&gt;The starting data is in too. 305,000 US and UK keywords are loaded and scored, and smaller German, French and Spanish sets are loading right now. Getting there took two days and four crashes, one of them caused by a rupee sign in a search about frying pans.&lt;/p&gt;

&lt;p&gt;What isn’t done&lt;/p&gt;

&lt;p&gt;The website works, but it doesn’t look finished. I’m going through every page this week, and I keep finding things that break on a phone.&lt;/p&gt;

&lt;p&gt;Then there’s SEO, which is a little embarrassing: I’m building a keyword tool and still had to do keyword research for my own landing page. I looked at competitors, at what ranks for the terms I care about and at the questions people ask around them. Then I wrote beginner guides on the topics I can answer well: what a keyword is, long-tail keywords, finding low-competition keywords and seeing which keywords a site ranks for.&lt;/p&gt;

&lt;p&gt;AEO is newer. More people now ask ChatGPT or Perplexity “what’s a free keyword tool” instead of searching Google, so I checked what those assistants say today. I’m writing the pages so an AI can understand and quote them, with a short answer near the top of each page, plain structure and an llms.txt file. Nobody really knows yet how much of this matters. I’m doing it anyway because it’s cheap.&lt;/p&gt;

&lt;p&gt;After that comes a long tail of small things. Does every button do what it says? Does the login email arrive in Outlook? Is the copy the same in all eight languages? None of it is hard, but there’s a lot of it, and it’s most of what’s left.&lt;/p&gt;

&lt;p&gt;The Chrome extension is built and its store listing is written. Store review takes a few days, so it might go live a little after the website.&lt;/p&gt;

&lt;p&gt;Where the data comes from&lt;/p&gt;

&lt;p&gt;Keyword tools are expensive. Ahrefs Lite is $129 a month and Semrush Pro is about $140, and thousands of bloggers, niche site builders and small shops each pay for roughly the same data.&lt;/p&gt;

&lt;p&gt;That data is built from something we all produce anyway: the searches we make every day. If you install the OpenKeyword extension and turn sharing on (it’s off by default), it shares the words you type into search engines. It doesn’t share the pages you visit or anything about who you are.&lt;/p&gt;

&lt;p&gt;I wrote the privacy rules before the code:&lt;/p&gt;

&lt;p&gt;Health, adult, legal and money searches are dropped on your own computer and never sent.&lt;br&gt;
The server never records “you searched X”. Each keyword gets a one-way token, so your searches can’t be linked together.&lt;br&gt;
A keyword stays hidden until 5 different people have searched it.&lt;br&gt;
Every search waits in a local queue for 30 minutes, where you can see and edit it, before it goes out in a batch.&lt;/p&gt;

&lt;p&gt;The 305k keywords I loaded make the tool useful on day one. After that it grows from sharing: every search that passes the 5-person rule adds a phrase a real person typed, free for whoever needs it next. One person sharing barely moves it, but a few thousand people can build something none of us would pay $129 a month for on our own.&lt;/p&gt;

&lt;p&gt;Launch&lt;/p&gt;

&lt;p&gt;OpenKeyword opens on 20 November at &lt;a href="https://openkeyword.org" rel="noopener noreferrer"&gt;openkeyword.org&lt;/a&gt;. It won’t be perfect on day one, and I’d rather fix things while real people use it than spend another month polishing it alone.&lt;/p&gt;

&lt;p&gt;If you’ve launched something that gets better as more people use it, how did you get your first hundred users? And if you try OpenKeyword and something’s broken, please tell me.&lt;/p&gt;

</description>
      <category>buildinpublic</category>
      <category>showdev</category>
      <category>webdev</category>
      <category>seo</category>
    </item>
    <item>
      <title>How I built a keyword research tool that can't see anyone's search history</title>
      <dc:creator>Constantin Alex</dc:creator>
      <pubDate>Fri, 13 Nov 2026 09:44:19 +0000</pubDate>
      <link>https://dev.to/constantin_alex_05c9c24b7/how-i-built-a-keyword-research-tool-that-cant-see-anyones-search-history-9ko</link>
      <guid>https://dev.to/constantin_alex_05c9c24b7/how-i-built-a-keyword-research-tool-that-cant-see-anyones-search-history-9ko</guid>
      <description>&lt;p&gt;I'm a 4th year CS student. Every time I had an idea for a blog post or a side project, I wanted to answer one question first: does anyone actually search for this?&lt;/p&gt;

&lt;p&gt;The tools that answer it well cost money. Ahrefs Lite is $129 a month. Semrush Pro is about $140. And what bothered me more than the price was this: thousands of bloggers, niche site builders and small shops each pay, separately, for roughly the same data.&lt;/p&gt;

&lt;p&gt;So I started building OpenKeyword, around one idea: if enough of us share the searches we already make, we can build that data together, and nobody has to pay for it.&lt;/p&gt;

&lt;p&gt;That idea only works if sharing is safe. This post is about how I built it so the system forgets who shared what. It opens on 20 November at openkeyword.org, and the extension is &lt;a href="https://github.com/CincaAlex/openkeyword-extension" rel="noopener noreferrer"&gt;open source on GitHub&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;What the tool does&lt;br&gt;
You type a keyword and get:&lt;/p&gt;

&lt;p&gt;a popularity score from 0 to 100&lt;br&gt;
the search intent: how-to, best, vs, question, or other&lt;br&gt;
a 12-month trend&lt;br&gt;
up to 50 related ideas, sorted by score&lt;br&gt;
It's free.&lt;/p&gt;

&lt;p&gt;The keyword list grows from a Chrome extension. If you turn sharing on, it shares the words you search on search engines. That's where the privacy work is.&lt;/p&gt;

&lt;p&gt;The rules I set before writing code&lt;br&gt;
I wrote these down first and treated them as a contract every part of the system has to follow:&lt;/p&gt;

&lt;p&gt;No search history per person, anywhere. Not in a database, not in a log.&lt;br&gt;
Sensitive topics never leave the device.&lt;br&gt;
IP addresses are never stored. Only a keyed hash, and only for rate limits, for at most 48 hours.&lt;br&gt;
Raw data never leaves the server. The public site only sees the published copy.&lt;br&gt;
The architecture&lt;br&gt;
 Chrome extension&lt;br&gt;
   | local queue, every word waits 30 min, sent in batches of max 50&lt;br&gt;
   v&lt;br&gt;
 Cloudflare Worker  (validate, normalize, HMAC token per keyword)&lt;br&gt;
   v&lt;br&gt;
 Cloudflare Queue&lt;br&gt;
   v  Cloudflare Tunnel, locked with Access service tokens&lt;br&gt;
     server: small API + Postgres 16   (raw data stays here)&lt;br&gt;
   |  hourly sync of public columns only&lt;br&gt;
   v&lt;br&gt;
 Cloudflare D1  &amp;lt;---  website + public API read from here&lt;br&gt;
The website and API run on Cloudflare.&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;On the device: drop first, ask later
The extension does nothing until you turn sharing on. The search-site permissions are optional and only requested at that moment.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;When it sees a search, it normalizes the words and runs them through a filter on your computer. Anything that looks like health, adult, legal, money trouble or a people lookup is dropped right there. A small part of the list:&lt;/p&gt;

&lt;p&gt;const SENSITIVE_TERMS = [&lt;br&gt;
  // health&lt;br&gt;
  "hiv", "cancer", "depression", "pregnan*", "symptom*", "diagnos*",&lt;br&gt;
  // legal / crime&lt;br&gt;
  "arrest*", "lawyer", "divorce", "warrant", "mugshot*",&lt;br&gt;
  // finance distress / identity&lt;br&gt;
  "bankrupt*", "debt", "payday", "ssn", "passport",&lt;br&gt;
  // people lookup&lt;br&gt;
  "who is", "address of", "phone number", "background check",&lt;br&gt;
  // ...&lt;br&gt;
];&lt;br&gt;
Validation also rejects URLs, emails, phone-like numbers, any run of 6+ digits and search operators like site:. A search for a person's phone number shouldn't become a "keyword".&lt;/p&gt;

&lt;p&gt;Then there's timing. If every search were sent the moment it happened, the server would see a live stream of one person's browsing. So every word waits at least 30 minutes on its own clock, in a local queue the user can open and edit. An alarm runs every 5 minutes with 0-5 minutes of random jitter and sends whatever has waited long enough, in batches of up to 50.&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;The Worker: tokens instead of identities
Each install generates 32 random bytes on the client. That install key never gets stored on the server.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;When a batch arrives, the Worker re-validates every keyword (never trust the client) and turns each one into a token:&lt;/p&gt;

&lt;p&gt;/** dedupe token = hex(HMAC-SHA256(CONTRIB_SECRET, install_key|keyword_norm|country|lang)) */&lt;br&gt;
export function dedupeToken(env: Env, installKey: string, keywordNorm: string, country: string, lang: string) {&lt;br&gt;
  return hmacHex(env.CONTRIB_SECRET, &lt;code&gt;${installKey}|${keywordNorm}|${country}|${lang}&lt;/code&gt;);&lt;br&gt;
}&lt;br&gt;
That gives two properties:&lt;/p&gt;

&lt;p&gt;The same person searching the same keyword twice produces the same token, so they count once. You can't push a keyword public by searching it 50 times.&lt;br&gt;
The same person searching two different keywords produces two unrelated tokens. There's no column that ties "running shoes" and "cheap flights rome" to the same person. That's how rule 1 holds.&lt;br&gt;
The contributions table in Postgres is literally (keyword, country, lang, token, first_seen_day). There is no install column to join on.&lt;/p&gt;

&lt;p&gt;One subtle leak I had to close: contributors get a small credit count for their contributions. If the credit message and the contribution message arrived together, timing alone could pair them. So the credit goes out as a separate queue message with a random 60-600 second delay.&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;&lt;p&gt;The queue and the tunnel&lt;br&gt;
The Worker puts everything on a Cloudflare Queue. The consumer posts batches to my server through a Cloudflare Tunnel, protected by Cloudflare Access service tokens, so the API isn't open to the internet.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Postgres : the 5-person rule&lt;br&gt;
Every keyword has a status: seeded, candidate, published or hidden. A contributed keyword starts as a candidate and is invisible. An hourly job counts distinct tokens per keyword, and once a keyword reaches 5, it's scored and promoted to published.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Sync back to the edge&lt;br&gt;
Every hour, the home server pushes only public columns (keyword, score, bucket, intent, trend, contributor count) to Cloudflare D1. Postgres has a view called public_keywords, and the sync user can only read that view. The raw signals and the contributions table never leave the house.&lt;/p&gt;&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;The website and API read only from D1, so they're fast and keep working even when my server is offline. Related ideas use SQLite FTS5 in D1 with a prefix match on each word.&lt;/p&gt;

&lt;p&gt;OpenKeyword opens on 20 November. If you want one email when it does, you can pre-register at &lt;a href="https://openkeyword.org" rel="noopener noreferrer"&gt;openkeyword.org&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;I'd love feedback on the design, especially from anyone who can find a hole in the token scheme. That's the part I most want to get right.&lt;/p&gt;

</description>
      <category>showdev</category>
      <category>webdev</category>
      <category>cloudflare</category>
      <category>privacy</category>
    </item>
    <item>
      <title>I deleted my Rust helper and moved zip extraction into the browser. Here's what it cost.</title>
      <dc:creator>Constantin Alex</dc:creator>
      <pubDate>Mon, 19 Oct 2026 18:57:20 +0000</pubDate>
      <link>https://dev.to/constantin_alex_05c9c24b7/i-deleted-my-rust-helper-and-moved-zip-extraction-into-the-browser-heres-what-it-cost-51l8</link>
      <guid>https://dev.to/constantin_alex_05c9c24b7/i-deleted-my-rust-helper-and-moved-zip-extraction-into-the-browser-heres-what-it-cost-51l8</guid>
      <description>&lt;p&gt;A couple of weeks ago I wrote about &lt;a href="https://dev.to[first%20post]"&gt;LinkUnzip&lt;/a&gt;, a Chrome extension that pulls files out of a zip on the internet without downloading the zip. It reads the zip's table of contents with an HTTP Range request, then downloads only the bytes of the files you want and writes them straight into a folder.&lt;/p&gt;

&lt;p&gt;The idea worked. The install didn't.&lt;/p&gt;

&lt;p&gt;To write files into a real folder, version 1.0 needed a small helper program on your PC. I wrote it in Rust and the extension talked to it through Chrome's native messaging. It was fast. It was also a Windows .exe from someone nobody has heard of, unsigned, so Windows SmartScreen warned about it. A few people told me straight up that the helper was a dealbreaker. And of the first five people who uninstalled the extension, four were on Linux or ChromeOS, where the helper didn't exist at all. They installed it, found out it couldn't work, and left.&lt;/p&gt;

&lt;p&gt;So for 1.1.0 I moved the whole engine into the browser. Here's how that went, including the parts where the browser is slower.&lt;/p&gt;

&lt;h2&gt;
  
  
  What the browser already has
&lt;/h2&gt;

&lt;p&gt;It turns out almost everything the Rust helper did has a browser API now:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;code&gt;fetch&lt;/code&gt; with a &lt;code&gt;Range&lt;/code&gt; header gets any slice of the zip.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;DecompressionStream("deflate-raw")&lt;/code&gt; inflates zip data natively. Zip uses raw deflate, without the zlib header, and Chrome supports exactly that.&lt;/li&gt;
&lt;li&gt;CRC-32 is about 40 lines of JavaScript with a lookup table.&lt;/li&gt;
&lt;li&gt;The File System Access API (&lt;code&gt;showDirectoryPicker&lt;/code&gt;, &lt;code&gt;createWritable&lt;/code&gt;) writes into a real folder the user picked.&lt;/li&gt;
&lt;li&gt;An offscreen document gives the engine a page that keeps running after the popup closes.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The core is pretty small. Simplified, one file looks like this:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="c1"&gt;// fetch one file's compressed bytes and inflate them as they arrive&lt;/span&gt;
&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;res&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="nf"&gt;fetch&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;zipUrl&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="na"&gt;headers&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="na"&gt;Range&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;`bytes=&lt;/span&gt;&lt;span class="p"&gt;${&lt;/span&gt;&lt;span class="nx"&gt;start&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="s2"&gt;-&lt;/span&gt;&lt;span class="p"&gt;${&lt;/span&gt;&lt;span class="nx"&gt;end&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="s2"&gt;`&lt;/span&gt; &lt;span class="p"&gt;}&lt;/span&gt; &lt;span class="p"&gt;});&lt;/span&gt;
&lt;span class="k"&gt;if &lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;res&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;status&lt;/span&gt; &lt;span class="o"&gt;!==&lt;/span&gt; &lt;span class="mi"&gt;206&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;throw&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nc"&gt;Error&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="dl"&gt;"&lt;/span&gt;&lt;span class="s2"&gt;the server ignored the Range header&lt;/span&gt;&lt;span class="dl"&gt;"&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;

&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;plain&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nx"&gt;res&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;body&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;pipeThrough&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nc"&gt;DecompressionStream&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="dl"&gt;"&lt;/span&gt;&lt;span class="s2"&gt;deflate-raw&lt;/span&gt;&lt;span class="dl"&gt;"&lt;/span&gt;&lt;span class="p"&gt;));&lt;/span&gt;
&lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="nx"&gt;plain&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;pipeTo&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="k"&gt;await&lt;/span&gt; &lt;span class="nx"&gt;fileHandle&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;createWritable&lt;/span&gt;&lt;span class="p"&gt;());&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The real code does more: it groups neighbouring files into one request, skips each file's local header, counts CRC-32 and size on the way through, and only gives the file its real name once both match. But that's the heart of it, and it's all stuff the browser ships.&lt;/p&gt;

&lt;h2&gt;
  
  
  Where the browser is slower
&lt;/h2&gt;

&lt;p&gt;I measured both versions on the same machine.&lt;/p&gt;

&lt;p&gt;The Blender 4.2 zip is 365.8 MB with 5,518 files. The Rust helper extracted all of it in 15 to 17 seconds. The browser version takes 52 seconds.&lt;/p&gt;

&lt;p&gt;The download isn't the problem. The folder is. Chrome runs file system operations one at a time, and each file needs about four of them: get a handle, create a writable, write, close. At roughly 2 ms each, that's about 7 ms per file no matter how many connections are downloading in parallel. For a zip of big files you don't notice. For a zip of thousands of tiny files you do. In a benchmark with 70,000 tiny files, Chrome took 500 seconds where plain Node took 101.&lt;/p&gt;

&lt;p&gt;I decided that's fine as long as the extension says so. The helper still exists as an optional extra for people who want the speed.&lt;/p&gt;

&lt;h2&gt;
  
  
  Chrome has opinions about where you save
&lt;/h2&gt;

&lt;p&gt;My first test build asked you to pick a folder. Then I tried to pick Downloads and Chrome said no. The folder picker refuses Downloads, Desktop and Documents themselves. You can pick a folder inside them, but not the folder itself. For most people "put it in Downloads" is the whole point, so that was a problem.&lt;/p&gt;

&lt;p&gt;The fix was to use Chrome's own downloads for the default case. Each checked file becomes a Blob and goes through &lt;code&gt;chrome.downloads.download&lt;/code&gt; into &lt;code&gt;Downloads/&amp;lt;zip name&amp;gt;/&lt;/code&gt;. No picker, no permission prompt. Three things I learned the hard way:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;If the Blob has no type, Chrome renames every file to .txt. &lt;code&gt;application/octet-stream&lt;/code&gt; fixes it.&lt;/li&gt;
&lt;li&gt;Chrome holds back some file types (.dll, .ini, .cfg, .manifest and a few more) until you click Keep on each one. The extension now counts them before it starts and warns you.&lt;/li&gt;
&lt;li&gt;It's slow: about 150 ms per file, so 6 or 7 files a second. For a zip with thousands of files the popup now tells you how long it will take and suggests picking a folder instead, which is the fast path (that 7 ms per file).&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  The bug that only showed up for real users
&lt;/h2&gt;

&lt;p&gt;The first build worked in all my tests. Then I tried it as a normal user and got "the engine didn't start".&lt;/p&gt;

&lt;p&gt;Chrome stops an extension's background service worker when it's idle. My offscreen page, the one doing the actual work, outlived the worker and was left holding a dead connection. When the worker woke up again, it saw that the page already existed and waited for it to say hello. It never did.&lt;/p&gt;

&lt;p&gt;Now a busy page reconnects by itself, an idle leftover page gets a second and a half and is then replaced, and running extractions survive a worker restart. The test suite stops the worker on purpose, both while idle and in the middle of an extraction, so this can't quietly come back.&lt;/p&gt;

&lt;h2&gt;
  
  
  A bonus: preview
&lt;/h2&gt;

&lt;p&gt;Once the engine could pull one file out of a zip, showing that file was a small step. Every file in the list now has an eye icon that opens it in a tab: text and code, CSV as a table, images, PDF, audio and video. Only that one file is downloaded.&lt;/p&gt;

&lt;p&gt;One photo out of the COCO train2017 dataset, an 18.01 GB zip, downloads 219 KB for the preview. For CSV it stops even earlier: the first 1,000 rows of movies.csv from the MovieLens 32M dataset cost 23.3 KB out of a 227.9 MB zip. Checking whether a dataset is the one you want before downloading 18 GB of it is honestly the feature I use most now.&lt;/p&gt;

&lt;h2&gt;
  
  
  What still doesn't work
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Servers that ignore Range requests and send the whole file every time. GitHub's green "Download ZIP" button is one of them. For those you still need the helper, which streams the file once from start to end.&lt;/li&gt;
&lt;li&gt;RAR, 7z and password-protected zips.&lt;/li&gt;
&lt;li&gt;I've only tested it on Windows so far. It should work in Chrome on Mac, Linux and ChromeOS, since nothing in it is Windows-specific any more, but I haven't seen it with my own eyes yet.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Try it
&lt;/h2&gt;

&lt;p&gt;It's free and open source, no account, nothing to install besides the extension:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Chrome Web Store: &lt;a href="https://chromewebstore.google.com/detail/dapeiljhjlaelolngnoagobeojcpkica?utm_source=devto&amp;amp;utm_medium=article" rel="noopener noreferrer"&gt;https://chromewebstore.google.com/detail/dapeiljhjlaelolngnoagobeojcpkica?utm_source=devto&amp;amp;utm_medium=article&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Code: &lt;a href="https://github.com/CincaAlex/LinkUnzip" rel="noopener noreferrer"&gt;https://github.com/CincaAlex/LinkUnzip&lt;/a&gt;
&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;If you're on Mac or Linux I'd really like to know if it works for you. And if you've hit the same Chrome limits with the File System Access API, how did you get around the one-operation-at-a-time thing? I couldn't find a way.&lt;/p&gt;

</description>
      <category>showdev</category>
      <category>javascript</category>
      <category>webdev</category>
      <category>opensource</category>
    </item>
    <item>
      <title>I didn't have room for a zip and its files, so I learned how to skip the zip</title>
      <dc:creator>Constantin Alex</dc:creator>
      <pubDate>Fri, 09 Oct 2026 12:59:33 +0000</pubDate>
      <link>https://dev.to/constantin_alex_05c9c24b7/i-didnt-have-room-for-a-zip-and-its-files-so-i-learned-how-to-skip-the-zip-3gg6</link>
      <guid>https://dev.to/constantin_alex_05c9c24b7/i-didnt-have-room-for-a-zip-and-its-files-so-i-learned-how-to-skip-the-zip-3gg6</guid>
      <description>&lt;p&gt;A while ago I had to unpack a 60 GB zip. The files would have fit on my laptop. The zip plus the files didn't.&lt;/p&gt;

&lt;p&gt;That's the annoying part of "download it, extract it, delete the zip". For a moment you need room for both. Usually that's fine. That day it wasn't, and I got stuck on a question: do I actually need the zip on my disk at all?&lt;/p&gt;

&lt;p&gt;Turns out, not really.&lt;/p&gt;

&lt;h2&gt;
  
  
  The trick
&lt;/h2&gt;

&lt;p&gt;A zip file keeps its table of contents at the very end. It's called the central directory, and it lists every file in the archive, how big it is and where exactly it starts inside the zip.&lt;/p&gt;

&lt;p&gt;The second piece is something most servers already support: HTTP Range requests. You can ask a server for just a slice of a file, like "give me bytes 1000 to 2000". It's the same thing that lets you skip ahead in a video or resume a download.&lt;/p&gt;

&lt;p&gt;Put those together and you get this:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Ask the server how big the zip is.&lt;/li&gt;
&lt;li&gt;Download only the end of it and read the table of contents.&lt;/li&gt;
&lt;li&gt;Now you know every file and where it lives, so download just the bytes of the files you want.&lt;/li&gt;
&lt;li&gt;Unpack each one as it arrives and write it straight into a folder.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;The zip itself never gets saved anywhere.&lt;/p&gt;

&lt;h2&gt;
  
  
  The numbers
&lt;/h2&gt;

&lt;p&gt;This is the part that still feels a bit like cheating to me.&lt;/p&gt;

&lt;p&gt;COCO train2017 is a popular machine learning dataset: an 18.01 GB zip with 118,287 photos. Getting one photo out of it downloaded 12.2 MB in total, and 11.9 MB of that was just the table of contents. It took between 3.7 and 6.8 seconds over four runs.&lt;/p&gt;

&lt;p&gt;For a full archive: the Blender 4.2 zip is 365.8 MB with 5,518 files. Extracting all of it wrote 925.6 MB of files with 0 B of zip on disk, in 15 to 22 seconds on my home connection. The normal way, you'd need 1.26 GB free for the zip and the files together.&lt;/p&gt;

&lt;p&gt;To be clear, for a whole archive it downloads about the same amount as a normal download. It's not faster magic. You save the space the zip would take, and when you only need part of an archive, you only download that part.&lt;/p&gt;

&lt;h2&gt;
  
  
  From a script to an extension
&lt;/h2&gt;

&lt;p&gt;It started as a small command line tool in Rust, just for me. You gave it a link and a folder, and it did the rest.&lt;/p&gt;

&lt;p&gt;Then I got tired of copying links into a terminal and wanted it on a right-click. So it became a Chrome extension. The catch: an extension can't really write a whole folder tree to your disk on its own. So there's a small Windows helper program, and the extension talks to it through Chrome's native messaging. The extension shows the list and the buttons, the helper does the downloading and writing.&lt;/p&gt;

&lt;p&gt;Now it's: right-click a zip link, see what's inside and how much space it needs, pick the files, done.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Frgmooajfed0klsteuurj.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Frgmooajfed0klsteuurj.png" alt=" " width="800" height="500"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Things that broke along the way
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;GitHub doesn't play along.&lt;/strong&gt; The "Download ZIP" button on GitHub doesn't support Range requests at all, so the whole trick falls apart there. I ended up writing a second mode that reads the zip once from start to finish and unpacks files as they go by. No preview, but still no zip on disk.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Zips can be nasty.&lt;/strong&gt; A zip can contain a file called CON or NUL.txt, which Windows simply refuses to create. Worse, it can contain paths like ../../something that try to write outside the folder you picked. All of that has to be checked before a single byte gets written.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Every file gets checked.&lt;/strong&gt; Each file is written under a temporary name first, its checksum and size are compared with what the zip says, and only then does it get its real name. If the connection drops halfway, you don't end up with a half-written file that looks fine.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The Chrome Web Store review.&lt;/strong&gt; An extension that talks to a native program and can ask for cookies (for downloads that need your login) gets a careful review. Explaining every permission clearly took a while, but it went public this week.&lt;/p&gt;

&lt;h2&gt;
  
  
  The honest limits
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Windows only for now&lt;/li&gt;
&lt;li&gt;It needs the helper installed (per user, no admin rights)&lt;/li&gt;
&lt;li&gt;No RAR or 7z yet, and no password-protected zips&lt;/li&gt;
&lt;li&gt;The installer isn't code-signed yet, so Windows SmartScreen shows a warning the first time&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Try it or read the code
&lt;/h2&gt;

&lt;p&gt;It's free and open source, no account and no tracking. It's called LinkUnzip.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Code: &lt;a href="https://github.com/CincaAlex/LinkUnzip" rel="noopener noreferrer"&gt;https://github.com/CincaAlex/LinkUnzip&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Website: &lt;a href="https://linkunzip.app" rel="noopener noreferrer"&gt;https://linkunzip.app&lt;/a&gt;
&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;I'm curious whether other people run into this too, or if it's just me. And if you know a zip link it chokes on, I'd really like to hear about it.&lt;/p&gt;

</description>
      <category>rust</category>
      <category>showdev</category>
      <category>javascript</category>
      <category>webdev</category>
    </item>
  </channel>
</rss>
