· 7 min read

Why I shut down Vortex

Last year I built a Chrome extension that worked on about 90% of the websites I threw at it, and broke on the handful I needed most.

I had like 80+ tabs open while researching something and had zero idea how any of them connected to each other. Tab managers don't really solve this, they just add more UI on top of the same list. Browsing is spatial but tabs are linear, and nobody thinks in a flat row of 90 rectangles.

So I built Vortex. Every website becomes a node on an infinite canvas, and when you click a link inside a node it opens as a connected child node, so your research trail maps itself out as you go. Think Figma's prototype view, but for actual browsing.

Vortex

I used it daily for weeks. It had a layer panel like Figma, a minimap for getting around, sticky notes on any node and color coding for different threads, and it worked well.

Manifest V3 made iframes of real sites close to impossible

The whole product is rendering live websites inside nodes on a canvas. When you click a link inside one of those sites, instead of navigating away, it spawns a new connected node.

To do that you have to embed third-party websites in iframes inside your extension, and Manifest V3 made that nearly impossible for any real website. It was a few things stacking on top of each other.

Vortex with working sites

CSP headers block the embed silently

Most websites ship a Content-Security-Policy with frame-ancestors 'self' or frame-ancestors 'none', which tells the browser not to let anyone embed them in an iframe. Google Docs, Notion, GitHub, Linear and every other tool you'd actually want on a research canvas set this. The browser refuses to load them and you get a blank frame, and there's no error you can catch either.

Older sites use X-Frame-Options instead

Sites that don't use CSP still set X-Frame-Options: DENY or SAMEORIGIN. Same result, the browser blocks the embed before your code even runs.

V3 took away the workaround

In V2 you could use the webRequest API to intercept responses and strip the CSP and X-Frame-Options headers before the browser enforced them. It was hacky but it worked. V3 removed webRequest for response modification and moved everything to declarativeNetRequest, which is static rules you declare upfront. You can't change response headers based on context anymore, and Chrome reviews those rules and doesn't like you stripping security headers.

COEP blocks you one layer down

Then there's the newer stuff. Cross-Origin-Embedder-Policy: require-corp means the embedded site has to explicitly opt in to being embedded, and almost no site does that for a random extension. So even if you somehow got past CSP, COEP would block you at another layer.

Why I didn't use a proxy

I could've gone the proxy route and run every website through a server that strips the security headers before serving it to the extension. Some tools do this. But it means proxying every site the user visits through your server, which is a privacy problem, and it's expensive since every page load hits your infra. It still breaks sites that rely on same-origin cookies, WebSockets or service workers, and you end up responsible for all of that traffic.

Asking for all_urls scares people off

The other option was requesting <all_urls> with full host access. Chrome Web Store reviewers flag that immediately and review times go from days to weeks. Users also get the "this extension can read and change all your data on all websites" warning, and I don't think anyone clicks allow on that for a canvas tool.

Layer panel and minimap

It broke on the apps people research with

Around 90% of the sites I tried worked fine: blogs, docs, articles, marketing pages. The ones that broke were dashboards and apps like Google Docs, Notion, GitHub and Linear, which have the strictest headers and are also the ones you open most during deep research. I couldn't see a way to ship a research tool that didn't work on them.

The browser was the wrong place for the canvas

Having a spatial canvas to think on while you research is a real problem worth solving. 80 tabs is still a mess, and the context disappears when you close the browser. People wanted this too, the response when it went viral was crazy.

But rendering third-party websites inside a canvas from a browser extension is always going to be a losing fight. Security policies are getting stricter, and the web keeps moving towards more isolation. I was literally trying to break web security to make my product work, and you can't build on that for long.

I also realised I'd been solving the browsing problem, and what I actually cared about was the thinking: seeing how ideas connect. The websites were just where the ideas came from.

When I looked at how I actually used Vortex, the sticky notes I attached to nodes were worth more than the rendered websites. Laying things out as "this connects to that" was the useful part, and I didn't need a live copy of the page for that.

Sticky notes and color coding

Moving the good parts into SideNotes

I'm taking the parts that worked and putting them into SideNotes.

SideNotes is my desktop app. It's Electron, local first, with Obsidian-style linking and no cloud, account or login. It already had the note graph and a block editor, but it was missing the spatial canvas.

SideNotes editor

So the infinite canvas is now a first class view. Instead of embedding live websites, which is what killed Vortex, the nodes on the canvas are notes. You can drop in a URL and it saves as a link card with the title, favicon and OG image, but it doesn't render the site.

It's the same spatial way of working, and none of the CSP, Manifest V3, Web Store review or proxy problems apply because nothing is being embedded.

The canvas

It's built with React Flow. You drag notes from the sidebar onto the canvas, connect them and group them. It saves as .canvas JSON, which is Obsidian-compatible, so you can open the same files in Obsidian if you want.

The Figma shortcuts came over too: ⌘K to add, ⌘F to search, ⌘D to duplicate. You can select multiple nodes, group them, collapse groups and color code threads, so all the spatial stuff that made Vortex feel good is there without the iframe hell.

Everything else in SideNotes

There's also a block editor with slash commands, drag handles and code blocks with syntax highlighting, and a connections graph (Sigma + WebGL) that shows every wikilink in the vault. It has daily notes, mood tracking, tables, Mermaid diagrams, six themes in light and dark, and PDF and HTML export. Everything is plain markdown files on disk, with no database, sync or account.


Vortex failed because I picked the wrong surface. The browser shows other people's apps, and you can't turn it into a canvas without fighting every security policy on the internet. Those policies exist for good reason na. With SideNotes I own every layer of the stack, so the canvas just works.

Vortex on GitHub / SideNotes on GitHub

Comments

Loading comments
Leave a commentno sign in
0 / 1000