Grouping tabs by kind of site, without sending them anywhere
27 October 2026 · 5 min read
Grouping tabs by kind of site (code with code, chat with chat, dashboards together) is the most useful way to group them, and the hardest to do automatically. Many tab tools solve it by sending your tab titles and addresses to an AI service to sort. You don't need to: a tab says a lot about itself, and all of it is there in the browser.
What a tab gives away
Four things, from the most certain to the least:
- The site. youtube.com is video and gmail.com is email. That covers most of the tabs people have open, because most of them are on a few hundred well-known sites.
- The shape of the address. Work apps use the same paths wherever they're hosted, so they give themselves away on a company's own domain too:
/-/merge_requests/is a merge request on any GitLab/browse/ABC-123is a Jira issue/owa/is Outlook on the web, and?_task=mailis Roundcube webmail/wp-admin/is a WordPress dashboard/swaggeris API documentation
- The end of the title. Apps sign their tabs: "Grid frequency - Dashboards - Grafana", "ops - Acme Mattermost", "Fotos - Dateien - Nextcloud". The name has to be a whole part of the title, though: "Grafana Labs raises $270M | The Cloud Brief" is a news story, not Grafana. The app maker's own website signs its blog posts the same way, so a page on keycloak.org that ends "- Keycloak" is Keycloak's news, not your login server.
- The rest of the address.
app.,console.anddashboard.at the front of an address almost always mean a web app you're using rather than a page you're reading. An image opened on its own has a title like "photo.jpg (1200×800)", whatever site it's on.
These are worth knowing even if you group tabs by hand. Ctrl/Cmd+Shift+A opens Chrome's tab search, which matches titles, so typing "GitLab" or "Grafana" finds every tab an app signed.
How WithFew does it
WithFew (our extension) groups loose tabs by task, by kind of site, by site when one gets busy, and by time. To know the kind of site, it reads each tab's address and title in the order above, on your device, and stops at the first answer it's sure of. Since version 1.0.9 it knows about 150 work apps by sight, so a company's own GitLab, Jira, Outlook or Grafana is filed with the rest of its kind.
For a site it has never heard of, a small model built into the extension reads the address and title. The model was trained on public web pages and only answers when it's sure. When nothing is sure, the tab goes under Today or Earlier rather than into a group that's probably wrong.
How well it works
We checked it on a month of our own browsing: 150 tabs, labelled by hand, that we never tuned anything on. About three tabs in four landed in a group by kind of site, and four in five of those in the group we'd have chosen ourselves. The rest waited under Today. Well-known sites and work apps are nearly always right. The hard cases are the home pages of small products, where even people disagree: is "ReVideo · Create Breathtaking Videos" a video site or a tool?
Groups you make yourself are left alone, and a tab you move stays where you put it. It's free for 7 days, then from $1.58 a month.