The library is already on your disk. This just searches it.

You have a folder — or six — full of images and clips you saved because you might need them again, and the actual problem was never collecting them. It was that "I know I have one of these somewhere" now takes longer to answer than it would take to find the reference from scratch. A reference manager's whole job is closing that gap without asking you to redo the collecting.

What goes wrong with folders alone

Finder gets you partway. It shows you what is there and lets you rename and tag things, and for a few hundred files that is plenty. The trouble starts around the point where a collection crosses a few thousand items across more than one drive: Spotlight indexes filenames and some metadata but has no idea what a tag means to you, a rating does not exist as a concept, and a search across type:video and rating:>=4 at once is not a sentence Finder understands at all. You end up scrolling, which is not the same skill as searching and does not get faster with practice.

The other failure is quieter. Tag something today, rename the folder next year, and whatever kept track of that tag — a spreadsheet, a naming convention, a second app with its own idea of your files — usually does not follow the rename. The metadata rots while the files sit there fine, and six months later you cannot tell which state you are in until a search comes back empty for something you know you tagged.

Search, as one line rather than a panel

The same string drives the grid, a saved view and the autocompleter. The keys are type ext tag rating is has no in width height added modified, AND across keys and OR inside a comma list, with no parentheses — grouping is what makes a query impossible to draw as chips or autocomplete, so it is left out on purpose. A word the parser does not recognise becomes free text instead of an error, so the grid keeps returning rows while you are still typing.

Save that string as a view and it is re-parsed on open rather than stored as a fixed list of files: rename a tag and the view updates itself, because there is no membership list to repair.

A workspace that is a folder, not a location the app owns

Everything the app writes — the database, the thumbnails, the canvas-resolution proxies — lives inside the folder you pointed it at. There is no application-support directory holding the part that matters and no account anywhere in the chain. Back that folder up with whatever already backs up your drive and the library is backed up. Move it to a new Mac and open it there with no export step.

Registering a second drive is a second, independent workspace. An external drive that comes back online after being unplugged for a month is reconciled by the watcher rather than rediscovered from nothing — filesystem events become the smallest re-scan that makes the index true again, so a rename in Finder is a rename and not a new file with a lost rating.

Ratings, tags and notes that survive being wrong at first

tag:any:portrait,studio          either tag applies
has:notes no:tags                written up but not yet filed
width:>=3000                     will hold up printed large
added:>2026-06                   the last two months

None of this requires deciding on a taxonomy up front. Tag as you go, rate what stands out, write a note when you remember why an image mattered, and the search line reaches all of it later. Getting a tag wrong costs a rename, not a rebuild, because the view that finds it re-parses the tag name rather than a cached list of matches.

What the index actually reads off a file

Registering a folder does more than list filenames. Coii Ref reads dimensions, EXIF and colour off every image it indexes, so width:>=3000 or a colour-based sort work without you having done anything to the files yourself — the data was already inside them, and the first index just makes it searchable. Nothing is written back into the original file: the reading is one-directional, and a reference stays exactly the bytes it was before Coii Ref ever saw it.

Saving a search as a view is a small thing that matters more than it looks. The view stores the query text, not a snapshot of which files matched it on the day you saved it, and re-parses that text every time you open it. Rename a tag from arch to architecture a year later and every view built on it updates itself the next time you open the view — there is no membership list anywhere to go stale, because there never was one to begin with.

At the size where scrolling actually stops working

Nobody needs a reference manager for four hundred images — Finder is faster for that, and this page will not pretend otherwise. The interesting point is somewhere past ten or twenty thousand files, spread across more drives than you can hold in your head, in folders someone else has reorganised since you last looked. That is the case the app is actually built around, and it changes what "search" has to mean: not a filter over what is currently on screen, but a query over an index that was built once and has been kept current since.

Getting there is a first pass you start and walk away from — reading tens of thousands of files takes time regardless of the tool doing it — but it is a pass you do once per folder, not once per session. After it, the watcher is what keeps the index honest: a rename is a rename, a moved subfolder is followed, and three hundred new files dropped in overnight are picked up without a manual re-import. A library that goes stale every time you tidy up in Finder is one you eventually stop trusting, and one you stop trusting is one you stop searching, which defeats the entire point of building it.

Coming from another tool, or from nothing at all

If you already keep references in refern, the only importer Coii Ref has reads that library directly — ratings, tags, notes and canvases, matched by relative path, with your originals left untouched either way. Coming from anything else, including a PureRef scene, an Eagle library or a folder with no system at all, there is no importer to run: you register the folder, let the first index finish, and start tagging and rating from there. Nothing about that requires undoing whatever organisation already exists — folder names, existing tags applied through Finder, and any naming convention you already use all stay exactly as they are, because the app was built to read what is there rather than to ask you to adopt its structure first.

Canvases, when the search turns into a board

Pull a shortlist out of a search and it can go straight onto a canvas — Something.coii-canvas, pretty-printed JSON, saved wherever you put it in your own folders. It diffs in git, syncs through whatever you already sync with, and opens in a text editor. That is not a side feature bolted onto a search tool; it is the same design carried one step further, since a board made this way names files that were never copied anywhere else.

What it costs and what it runs on

$19 once, for three Macs, updates included, nothing that renews. The trial runs 30 days with every feature enabled, no card and no account required to start it. Requirements are macOS 11 Big Sur or later, on Apple Silicon or Intel — two native builds rather than one universal one, same feature set on both.

The app makes two kinds of outbound request in its whole life: the licence check when you press Activate, and one read only for a server's Date header at launch, so a clock wound backward cannot mint free trial days. There is no in-app updater, so there is no update check, and no analytics or crash reporting of any kind.

At day thirty-one of an unpaid trial, one category of action stops — growing the library. Registering a new folder or running an indexing pass by hand needs a licence. Everything that reads keeps working permanently: search, tags, ratings, notes, canvases, thumbnails, and the folder watcher, which means an unpaid copy never starts lying to you about a file you renamed afterward. It only stops taking on more.

What it deliberately does not do

Worth stating before you rely on it for something it was never built for. There is no search by meaning or description — a term either matches a tag, a filename, a note or a piece of metadata you can name, or it does not match at all, and nothing here will surface an untagged image because it looks like what you asked for. There is no browser extension, so saving something from a website is still a manual step of putting the file into a folder first. And there is no relationship graph connecting one image to another — a saved view is the closest equivalent, and it connects by search terms rather than by association. Where any of those three is the actual job, refern has all three and this app has none of them.

Who this is actually for

Someone whose reference habit outgrew scrolling a while ago: a folder or six that has quietly crossed into the thousands, spread across more than one drive, tagged inconsistently if at all, with images and clips both in the mix. If your whole collection still fits on one screen, a reference manager is solving a problem you do not have yet, and Finder is the right tool until it isn't. There is no harm in waiting for that point to actually arrive rather than buying ahead of it — the trial exists precisely so you can test that question against your own folders before deciding either way.

If you are choosing between this and the two most-discussed alternatives, the honest comparisons are worth fifteen minutes each: refern is the more capable app on nearly every axis and runs on three platforms, Eagle copies what you import into its own library rather than indexing it in place, and Finder tags is the free option this page assumes you have already tried. For the boards this search eventually turns into, see the moodboard-focused version of this page. For a wider shelf of what people already use, the offline-first round-up and the no-copy round-up cover the same ground from a comparison shopper's side rather than this one.

Thirty days, every feature, no card and no account: download it and point it at the folder you would be most embarrassed to search in front of someone. The first index is the only part that takes patience; everything after it is a query line.