Links and where they came from

Once Bring in has added thousands of links, cards are the wrong density for them. This page covers two features, both desktop only for now:

the search words that filter by where something came from;
the links arrangement, which shows links one compact row each.

Search words

They work like #tags:

the match is exact, case doesn't matter, and every word must match (AND);
a finished word leaves the search box and becomes a filter chip, and its × removes it;
they filter the list and the map alike.

Word	Matches	Chip
`from:bookmarks`	an item brought in: from a browser's bookmarks, Pocket, or a link kept in a Keep note	from bookmarks
`from:midden`	an item captured in midden itself, on any of your devices	saved in midden
`in:recipes`	an item kept in a bookmark folder of that name, at any depth	in recipes
`in:dev/rust`	an item kept in those folders, in that order	in dev › rust
`site:github.com`	an item whose address, or the page it was captured from, is on that site (subdomains included)	on github.com
`is:unread`	Pocket said unread, no source said read, and it hasn't been looked at in midden since	unread
`is:unseen`	not looked at, or saved again, in 3 months, or as Settings › List says	unseen
`is:twice`	saved more than once: in two places, or twice in one	saved twice
`is:sensitive`	might be sensitive, strong or weak, and not said to be fine (Might be sensitive)	might be sensitive

Pocket isn't named outside Bring in. Once brought in, a Pocket save is a bookmark: it's from:bookmarks, its row says "bookmarked", and the by-source arrangement groups it with them. Bring in still names Pocket, and Settings → Bring in lists each import by its source.
Folder names are matched in lowercase, with spaces written as hyphens: "Long Reads" is in:long-reads. A browser's own top folder ("Bookmarks bar") can be named too, but the rail never shows it.
A word that isn't valid stays as search text, for example from:raindrop, is:great or site:nowhere. Typing from:, in:, site: or is: shows a one-line hint of what goes after it.
Folder data came from someone's export file, so only plain text folder names are used.
"Unread" is only said where it's true. A bookmark is never unread: midden can't know whether someone used it daily. Opening the item, or clicking its title in a link row, counts as looking at it, so it stops being unread.

The links arrangement

links in the list's arrangements shows only links, newest saved first. It stays while searching, with rows ranked by relevance; the other arrangements give way to relevance then.

A row:

The site's letter, on its own colour. midden doesn't fetch site icons, because each fetch would tell that site the link is saved.
The title. It opens the page in a new tab, and counts as opening it here.
Under it:
- the site;
- where it was kept, oldest first: "bookmarked · 2019", "Dev › Rust · 2021", "saved · 2026";
- markers: unread, unseen, saved twice;
- the item's tags and the tags midden associates with it (dashed), as on a card. ✓ makes an association a tag and × dismisses it. There's no "+": new tags are added in the opened item.
On the right, "4 →": how many things it's closely connected to: what the relevance engine finds similar, and what came from the same page. Shared tags and "saved the same day" aren't counted, though the opened item lists them: an import bunches both (a browser set up in one sitting dates its bookmarks together; a broad tag like #later covers hundreds), and the number is there to tell rows apart. So the opened item can list more than the row said. A row asks only after it has been on screen for 250 ms, so a fast scroll asks for nothing.
Clicking anywhere else on the row opens the item.

The rail on the left is hidden in narrow windows. Above the rows are three counts that also work as filters: unread, unseen, saved twice.

Sections: From, Your folders, Sites. Each row in them and each count is a search word: a click adds it, another takes it out.
Counts are over what's listed now, so picking bookmarks shows how many bookmarks are in each folder and on each site.
Your folders leaves out a browser's own top folders. A picked folder opens to show what's inside it.
Folders and sites show 8 at first, then "N more…" up to 60. Past 60, type in: or site:.
Known gap: a site row counts its exact site, but site: also matches subdomains, so the filtered list can be a little longer than the row said.

Addresses inside text

A web address written inside text can be opened from where it's shown. This covers text items (a note you typed, or one brought in from Keep), your notes on an item, and the text read from a picture, on the desktop and the phone.

Shown short: the site, then the end of the path (portal.example.com/…/chest-day). The https://, query and # part are hidden, and hovering shows the whole address.
The site comes from the parsed address, never from the text around it, so https://bank.example@evil.example/ shows as evil.example.
Only http, https and www. addresses become links (a www. address opens as https). javascript:, data:, file: and the rest stay as text.
Punctuation after an address stays text: a closing ., , or ). A bracket that opened inside the address stays part of it.
Opens in a new tab with no referrer, like a link card. Clicking it doesn't open the card.
The stored text never changes, so search, copy and export see what was written.
They're not link items. They don't show in the links arrangement, and site: doesn't match them.