LocalStash

Guides · Reference practice

How to build a design swipe file that actually gets used

Most swipe files die as a folder of screenshots nobody opens. The file is not the hard part — retrieval is. Here is the capture format, the tag vocabulary, and the twenty-minute weekly loop that keep one alive past the second week.

Updated 16 Nov 2026 9 min read LocalStash guides

A swipe file is where designers keep things worth stealing from: a checkout flow that explains itself, a pricing table that survives four tiers, the way a museum site handles a hundred images without feeling heavy. The habit is old — art directors kept physical folders of clippings long before anyone had a bookmarks bar. The tool changed; the promise did not. Collect what works, then reach for it when you need it.

Almost none of them are still in use a month later. Not because people are undisciplined, but because a screenshot has no retriever. Saved items pile up without a reason attached, and a folder you cannot search by intent is just storage with extra steps.

A swipe file is not a collection. It is a lookup table you build for a version of yourself who has already forgotten the context.

Why swipe files die

Three patterns account for nearly every abandoned file.

The shoebox. Everything gets saved, nothing gets named. After two months there are 340 items and no way to answer “show me three ways to handle a multi-step form” without scrolling. The file has become a museum of taste rather than a set of tools.

The museum. Only polished, pretty work is allowed in. That sounds like curation, but it removes exactly the references you need in the middle of a project: the plain, slightly ugly screen that solves a dense data problem in a way you can actually copy.

The ghost. A screenshot with no source, a link with no note. Six months later the image is unreadable, the link is dead, and the reason it mattered is gone. The entry survives but carries no information.

The test for whether you have a tool or a shoebox is simple. Ask it a design question and time the answer. “Show me two pricing tables with an annual toggle.” “Show me an empty state that is not an illustration.” If the answer takes more than half a minute, the library is not doing its job yet, and no amount of saving will fix that.

Decide what an entry is for before you save it

References do three different jobs, and each one needs different information. Mixing them in one undifferentiated pile is what makes search fail.

JobWhen you need itWhat the entry must carry
Pattern lookupMid-project, solving a specific problemThe mechanism named in one line, plus the element tag
Taste calibrationBefore starting something new, or in a reviewThe source, the author, and your judgment of why it holds up
EvidenceShowing a client or colleague what you meanA live URL, a readable screenshot, and one sentence of framing

Most people only build for the second job, because that is the fun one. They end up with a beautiful wall of inspiration and nothing to open when a form needs to handle eleven fields on a phone. Decide which job you are saving for at the moment of capture; the note practically writes itself once you do.

The four fields that make an entry retrievable

An entry needs four things. Everything else is optional decoration.

  1. Source. The URL, and the studio or person who made it if you know. This is what keeps the reference honest and lets you go back when a screenshot is not enough.
  2. The reusable part. One line naming the mechanism, not the mood. “Sticky summary column that becomes a bottom bar under 768px” is reusable. “Clean layout” is not.
  3. Why it works. One sentence of your own judgment. This is the field that makes the file worth more than a search engine, because it records a decision you already made.
  4. Tags. Two to five, drawn from a closed list you control. Tags are for retrieval, not description.

Weak and strong versions of the same entry, side by side:

FieldWeakStrong
Reusable partNice checkoutThree-step progress bar collapses to a single line on mobile; the collapse, not the styling, is the trick
Why it worksLooks goodKeeps the current step visible without stealing a third of the screen from the form
Tagsinspiration, nice, blue, web, 2026mobile, form, pattern

Notice that the strong version contains no adjectives about beauty. Adjectives do not survive a year; mechanisms do.

A tag vocabulary that stays under forty words

The most common tagging mistake is tagging by what the thing looks like. A year later you will not search for “pastel”. You will search for “empty state” or “onboarding”. Build the vocabulary around the questions you actually ask, and cap it.

Three axes are enough for a personal library:

AxisExamplesWhy
Mediumweb, mobile, print, motion, emailMost patterns are medium-specific once you look closely
Elementnav, form, table, empty-state, onboarding, pricing, typography, color, layout, iconThis is the axis you search on under deadline
Intentpattern, reference, taste, competitor, todoSeparates “copy this” from “admire this” from “try this”

Two rules keep it from rotting. First, if a tag has fewer than three entries after a month, delete it — a tag that matched one item is a note, not a category. Second, never create a tag for a client, a project, or a mood. Those belong in the note or in a collection, not in the vocabulary.

Rule of thumb A tag earns its place if you would type it into a search box while a meeting is waiting. If the answer is “probably not”, it is description, and description goes in the note.

LocalStash suggests tags from the domain, title and note using a small offline rule engine — GitHub becomes code and dev, a Mobbin link becomes inspiration and ui. That is useful for the medium and element axes. The intent axis is yours; no rule engine knows whether you saved something to copy it or to argue against it.

Capture in under ten seconds

If saving takes longer than the moment of interest, you will stop doing it. The whole flow should be four actions:

  1. Paste the URL. Most tools will fill the title from it; if the title is useless, replace it with the mechanism.
  2. Crop a screenshot. The region that contains the pattern, not the whole page. A full-page capture of a long article is unreadable as a thumbnail and useless as a reference.
  3. Write one sentence. The mechanism, in your words. If you cannot write it, you do not yet know why you are saving it — close the tab instead. This is the filter that keeps the file small.
  4. Add two tags. Element plus intent. That is the minimum that will make it findable.

Three small habits make the screenshots more useful later: keep them at 2× so text is legible, hide anything personal before capturing, and use a thin outline or a caption whenever the pattern occupies a small part of the frame. A cropped screenshot of the exact 300 pixels that matter is worth more than a pristine full-page capture you have to squint at.

The twenty-minute weekly review

Capture is cheap; curation is the work. Once a week, ideally on Friday while the context is still fresh, spend twenty minutes on the week’s intake:

  • Five minutes — triage. Open everything you saved this week. Delete anything whose reason you cannot restate. Expect to delete a quarter; that is normal and healthy.
  • Five minutes — group. Sort by element tag. Merge duplicates. If three entries say the same thing about onboarding, keep the clearest one and fold the other two into its note.
  • Five minutes — synthesize. Write five bullets titled “what I keep seeing in this pattern”. This is the step that turns a pile into knowledge, and it is the one almost everyone skips.
  • Five minutes — act. Pick one entry and attach it to something real: a live project, a critique, a note for next sprint. A reference that never touched work is a bookmark.

Friday matters less than the rhythm. The point of a fixed slot is that review stops being a decision. Twenty minutes is long enough to be useful and short enough that you will actually do it; an hour-long monthly session tends to become a browsing session.

What to save, and what to skip

Save these. States you would otherwise forget — empty, error, loading, success. Microcopy that does real work: the sentence that makes a scary action feel reversible. Layouts that solve density rather than decorate it. Motion with a job: a transition that explains where a panel came from. Competitor flows, especially the boring middle steps. Ugly but effective solutions, because they are the ones you can copy without a designer's budget.

Skip these. Full-page screenshots of articles you have not read. Dribbble shots of interfaces that could never be built — they train taste toward the impossible. Anything you describe as “someday”. Anything you cannot name in one sentence. And collections saved for an aesthetic you do not actually work in; they dilute search results and add nothing at review time.

Putting the file to work

In design review. Bring three references, not one. Name the tradeoff each one makes — “this works because it hides the second step, and that is also why it fails when the form has dependencies”. A reference used as evidence is persuasive; a reference used as taste is an argument about personality.

Before estimating. Count the states in two similar flows from your file. Estimates go wrong on the states nobody enumerated: draft, failed, partially saved, permission denied. Fifteen minutes in your own archive is usually more accurate than an hour of discussion.

When explaining to stakeholders. One screenshot with a sentence beats ten adjectives. “More modern” starts a debate; “this is the pattern I mean, here is where it breaks” ends one.

For calibration. Keep a rolling “current five” — the references that best represent where your taste is this month. Replace one when something genuinely better arrives. Comparing this month's five with last month's is a faster read on your own development than any portfolio review.

Advertisement
LocalStash keeps this system in one screen

A URL field, a note, tags that assign themselves, and screenshots that stay on your device.

Open the workspace

Failure modes and their fixes

SymptomWhat is actually wrongFix
Forty tags, all used onceTagging by description instead of intentDelete any tag with fewer than three entries; keep three axes only
Items with no noteCapture without a decisionOne sentence at save time, or do not save
Dead links everywhereScreenshots without sourcesAlways keep the URL; capture the region so the entry survives the site
Never opened after week twoNo retrieval habitTwenty minutes on Friday, starting with triage
Three hundred items, nothing findableThe file is a museumDelete aggressively; a hundred sharp entries beat a thousand vague ones

Questions

How many items before a swipe file becomes useful?

Around twenty, if they are tagged and annotated. Thirty annotated entries are more useful than three hundred screenshots, because the value comes from retrieval, not volume. A file that answers five common questions well is already earning its keep.

Should the file live in a cloud tool or on my machine?

It depends on what you are protecting. A cloud tool gives you sync and a phone client, at the cost of your library sitting on someone else's server. A local-first tool keeps the library in your browser and asks you to move it yourself with export files. The privacy guide covers the tradeoff in detail.

Does automatic tagging make manual tagging obsolete?

No, but it removes the boring half. A rule engine can read a domain and a title and propose medium and element tags with high accuracy. Intent — copy, admire, avoid — is a judgment only you can make, and it is the axis that determines whether the item ever comes back.

What about saving from my phone?

The capture should take two taps: share the URL into the app, add one sentence later. If your tool has no phone path, keep a temporary note on the phone and batch-import it during the weekly review. Half a system that you actually use beats a perfect one you avoid.

Start with four good entries

LocalStash opens with four sample references so you can see the fields, tags and screenshots working before you save anything of your own.

Written and maintained by the LocalStash team. No trackers, no email capture. Workspace · Privacy