Skip to content

Idea

Search isn't memory

Search is one of the great inventions of software. It is also the reason we assume the memory problem is solved.

Search is a lookup: you supply a term, it supplies matches. Recall doesn't start with a term.

01 · The input problem

Search requires the thing you've lost.

To search well you need a word that appears in the thing. But the word is exactly what went missing. You remember a colour, a layout, a week, a person, a purpose. You don't remember that the article used the phrase "retention curve".

So you guess. Three or four guesses later you conclude it isn't there, and you go and find something new instead. The item was never gone; the entry point was.

Every failed search is a vocabulary mismatch, not an absence.

02 · How recall works

Human memory retrieves by resemblance and context.

We reconstruct. A fragment brings back a neighbouring fragment: where you were, what else was happening, what it looked like, what it belonged to. Given enough fragments, the thing appears.

That's why "it was the week we were arguing about pricing" is a better clue than any keyword you could type, and why almost no software accepts it.

Appearance
Dark interface, a table, a hand-drawn diagram, a specific typeface.
Time
Not a date. A period: around the launch, before the trip.
Adjacency
Whatever you saved beside it, that afternoon or that project.
Purpose
The work it belonged to, which is usually the strongest thread.

03 · The honest limit

Semantic search helps. It isn't the whole answer.

Meaning-based search closes part of the gap: describe the idea and you find things that share the idea, even in different words. It's a real improvement and Shelf runs it alongside full text.

But it still expects one description to be enough. Reconstruction is iterative: you narrow with time, then with project, then with what sat next to it. A memory layer has to support the second and third move, not just the first.

Next thought

If recall works by fragments, retrieval should let you assemble them.

See memory reconstruction

Shelf V3 is out now

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.

Download the latest version →Version 3.0 · macOS 14+