342
Across 96 sources you actually read
Your private web memory
Thoth is a browser extension that quietly keeps every page you actually read, then finds it again from whatever you still remember.
10s
Reading threshold
A page is kept once it actually holds you.
0
Requests signed out
Capture, search and notes stay on this device.
1,000
Free memories
Local search and collections. No account.
Ctrl K
Search palette
Find a page from whatever you still remember.
Browser tabs pile up until their titles are crushed into illegibility, one address loading after another. One page is left: a page you read in full, with its address, its headline, its author and the minute you read it. Then that record comes apart. The address goes first, then the time, then the author, then the site, until all that is left of it is Quiet, Kyoto, Dark, alongside fragments of two other things you read: something about switching tasks, and Postgres, rollbacks.
You remember reading it.
Just not where.
What it is
Your browser already keeps two records of everything you read. Neither of them is the one you want three weeks later.
History
14 Mar 202623:42
slowpost.co/features/a-quiet-walk-through-kyoto-after-dark
Where you went, in the order you went there. Exact, complete, and no help at all three weeks later.
History remembers visits.
Bookmarks
A Quiet Walk Through Kyoto After Dark
What you decided, in advance, was worth keeping. Which means it only holds the things you already knew to save.
Bookmarks remember saves.
Thoth
quiet streetskyotoat night
What you actually read, found again from whatever you actually retained.
Thoth remembers reading.
The archive
Everything below is one product doing three things. Start with the one you doubt most.
01 · Finding it again
Two ways of asking for the same page, typed at the same time. The gold is where your words actually landed on it.
One page asked for two ways at once. Given the words that are on it, quiet walk kyoto, lexical search returns A Quiet Walk Through Kyoto After Dark with those three words marked in gold. Given the way you would actually say it, japanese streets with nobody on them, semantic search returns the same page with nothing marked, because none of those words are on it. The second example is the one that matters: task switching cost shares no whole word with Why Your Attention Stays Behind When You Switch Tasks, so lexical search answers that the memory is not here yet, and semantic search finds it anyway.
Looking forthe Kyoto piece, read one night in March
The gold is drawn by the extension wherever a word you typed is literally on the page. When nothing is marked, nothing of what you typed was there, and the page came back anyway.
Lexical search is local, always on, and needs no account. Semantic search is opt-in: it sends your query and bounded page text to Thoth's embedding service to be turned into vectors, which are then stored and ranked on your machine. Your page addresses are never sent.
02 · How something is kept
0:00
After the crowds disappear, the lanes near Gion go silent enough to hear your own steps. The guidebooks send you here at dusk, which is the one hour when the quiet has not arrived yet. Come back at eleven and the city gives you something it will not give anyone in daylight.
Remembered
Saved passage
AutomaticThoth watches how long a page actually holds you, across tab switches, lost focus and idle time. Past a threshold it keeps the page. You do not file anything, and there is no decision to forget to make.
IntentionalSome sentences are the reason you read the thing. Select one and it is kept exactly, as a passage attached to the memory, searchable on its own and weighted above the body of the page.
03 · After a few months
Nothing here was filed by hand. The shape below is what accrues on its own once Thoth has been running long enough to have an opinion about what you read.
342
Memories kept
across 96 sources you actually read
23
This week
9 new sources, steady reading
18
Resurfacing
older pages you came back to
The search above is the product. Try it with the vaguest description you can still put into words, then take a place in line.
What leaves your machine
A tool that remembers what you read has to be exact about this, so here is the whole of it rather than a promise.
Signed out, this is everything
No account. No requests. Nothing to switch off.
And only if you turn it on
Semantic search
Opt-in, and needs an account
Vectors come back and are stored and ranked on your machine.
Encrypted sync
Paid plans, at launch
Included in Pro and Max. Not generally available yet.
Blocked sites are a list you control, and a page on it is never read, never extracted and never stored. Banking, work, anything you would not want an archive of.
Across browsers
Sync moves an encrypted copy of the archive, not your browsing. The service stores ciphertext it cannot read.
Chrome
Built and reviewed
Brave
Built and reviewed
Firefox
Built and reviewed
Thoth Web
Read-only in any browser
Pro covers one browser. Max covers three. Thoth Web reads the same archive in a tab without installing anything.
Plans
Five things are gated and nothing else is. Pro and Max are the same product: the only difference between them is how many browsers they cover, so pick on that and nothing else.
| Capability | Free$0forever, no account | Pro$5.99a month, or $59 a year | Max$9.99a month, or $99 a year |
|---|---|---|---|
| Memories kept | 1,000 | Unlimited | Unlimited |
| Semantic searchAsk for a page the way you remember it | Not included | Included | Included |
| Encrypted backup | Not included | Included | Included |
| Sync across browsers | Not included | Included | Included |
| Browsers coveredThe only line where Pro and Max differ | Not included | One | Three |
| Get started | Start here | Join the waitlist | Join the waitlist |
In every plan, free included
Thoth has not launched, so nothing is charging yet. Paid plans include a trial when it does. Encrypted sync is built and is not generally available.
Before you install it
Thoth watches timing signals the browser already emits and does nothing else until a page passes the reading threshold. Extraction runs once per page, at that moment, and never again for the same route. There is no published benchmark yet, so that is a description of the work it does rather than a promise about milliseconds.
The archive lives in IndexedDB and grows with what you read, mostly text. There is no fixed cap on paid plans and no way to quote an honest average before launch. Free stops at 1,000 memories, which is the only hard number that exists today.
Yes. Blocked sites is a list you control, and a page on it is never read, never extracted and never stored. Nothing is captured for it to forget later.
Signed out, nothing. Turning on semantic search sends page titles, bounded chunks of page text and your queries to the embedding service; page addresses are never part of that. Encrypted sync sends ciphertext. Everything else stays where it was made.
The archive is local, so removing the extension removes it. Export it first if you want to keep it: memories, notes and passages come out in a file you own. Vectors, sessions and diagnostics are not part of the export.
Only for semantic search, which is the one feature that needs the embedding service. Capture, the archive, lexical search, collections, favorites, notes, blocked sites, deletion and export all work signed out.
Six months laterquiet streetskyotoat night
One email when Thoth opens. Nothing else, ever.