You don't remember the thing.
You remember pieces of it.
A colour. A Tuesday. A name you're half sure about. The project it belonged to. None of those is the thing. Together, they're enough.
Reconstruction is Shelf's answer to the way human memory actually returns: in fragments, out of order, warm before it's exact.
01 · Pieces, not things
Memory doesn't return whole.
Nobody recalls a document. They recall that it was long, that it had a table near the end, that they read it on a train, that it came from someone at work, and that it changed their mind about something.
Every one of those is a fragment. Search asks you to convert them into a keyword and throw the rest away. Shelf asks you to keep them.
02 · The fragments
Eight ways in.
Vague recall“Something about pricing psychology”
Meaning, not wording
Matched semantically
Visual fragment“Dark chart, one red line”
Appearance and read text
Previews you recognise
Approximate time“Around the launch, so late spring”
Capture date kept
Ranges, not exact days
Project context“It was for the pricing rework”
The work it sat inside
People“Ana sent it, I think”
Names read from content
Early capability
Source“A browser tab, not a file”
App or site recorded at capture
Related memories“It was near the competitor screenshots”
Neighbours in time and project
Semantic relationships“Same idea, different words”
Meaning links things that share no keyword
Give Shelf three fragments and it can usually give you back the moment.
03 · Assembly
How a memory gets rebuilt.
Reconstruction is iterative on purpose. Each fragment doesn't filter the list down to nothing; it makes some things warmer.
Fragment oneA meaning“pricing psychology” finds things that discuss the idea, including ones that never use the phrase.
Fragment twoA periodLate spring reweights rather than deletes. You keep the near misses in view.
Fragment threeA projectThe pricing rework pulls a cluster forward, with everything saved around it.
RecognitionYou see itNot because you described it perfectly, but because the field narrowed to things you'd recognise.
The bonusThe rest of the momentThe note, the screenshot and the article from that week come back with it.
A half-remembered documentLate springApproximate timePricing reworkProjectAnaPerson read from contentBrowserSource appCompetitor screenshotsSaved the same weekNote: “the anchor matters”Semantic match
Each fragment is a way in. The memory is where they overlap.
04 · Not search
Reconstruction is not
a better search box.
The difference isn't quality. It's the direction of the question.
Two different questions, not two speeds of the same one.
Aspect
Ordinary search
Memory reconstruction
What you supply
The right keyword
Whatever you happen to remember
What it matches
Words in the text
Meaning, appearance, time, source, project
Failure mode
Zero results, no next move
A warmer list you can narrow again
What you get back
A single file
The thing and the moment around it
Assumes
You named it well once
You lived it once
What you supply
Ordinary search
The right keyword
Memory reconstruction
Whatever you happen to remember
What it matches
Ordinary search
Words in the text
Memory reconstruction
Meaning, appearance, time, source, project
Failure mode
Ordinary search
Zero results, no next move
Memory reconstruction
A warmer list you can narrow again
What you get back
Ordinary search
A single file
Memory reconstruction
The thing and the moment around it
Assumes
Ordinary search
You named it well once
Memory reconstruction
You lived it once
Two different questions, not two speeds of the same one.
05 · Today
What's real, and what's still forming.
Working now: semantic matching across text, notes, page content, and text read from images and PDFs; time and source filters; previews built for recognition; and the related items surfaced around anything you open.
Still developing: entity understanding for people, companies and meetings, and richer visual similarity. We'd rather show you the seam than claim a finished machine.
Next thought
Fragments only connect because the library is connected.
Shelf V3 is the current build. Shelf does not use automatic updates, so download the latest build here and install it over your current version.
This is intentional. Shelf is built around a local-first, privacy-focused philosophy. An automatic updater requires additional background processes, network connections, update infrastructure, and another system with permission to modify the app. We prefer to keep Shelf’s network footprint and background activity as small as possible.
Manual updates keep the process explicit: Shelf only checks the internet when a feature actually needs it, and you decide when a new version is downloaded and installed.