The DEVONthink alternative for people who don't want a library
DEVONthink is the most serious document manager on the Mac and has been for twenty years. The question this page answers is narrower than which one is better: it is whether you want your documents to live somewhere particular.
DEVONthink asks you to bring your documents into a library, and once they are there it gives you sync, versioning, annotation, automation and a great deal else. Loupe reads the folders you already keep things in, adds nothing to them, and never writes. Both read scanned pages. The choice is not about search quality. It is about whether you want to run a library.
Two things most comparison pages get wrong
Worth clearing first, because they are the two claims a new search app is tempted to make about this one, and both are false.
DEVONthink reads scans. It runs proper character recognition over scanned documents and has done for many years. If you have read somewhere that a semantic search tool is the only way to make a scanned invoice findable on a Mac, that was written by someone who had not checked.
DEVONthink has AI, and it can be local. Its recent versions will talk to cloud providers if you ask them to, and they will equally talk to a model running on your own machine instead. So "we are the private one" is not an argument that survives contact with their documentation. The real difference is smaller and easier to state: there is no provider setting in this app, because there is nothing in it that would call one. That is a difference in architecture rather than in configuration, and whether it matters to you is a fair question rather than an obvious one.
The difference that is real
DEVONthink is organised around databases. You import documents into one, or you index a folder into one, and from that point the database is the thing you work in: it is where the tags live, where the annotations attach, what syncs to your other Mac, what the smart rules act on, and what the iOS app opens. That structure is not overhead, it is the product. Everything good about DEVONthink follows from your documents having a place.
Loupe has no place. It reads the folders you already use, keeps everything it learns in one file of its own, and writes nothing back. Nothing is moved, renamed, tidied or deleted, and there is no code path in the app that could. Delete its one index file and your disk is exactly as it was, because it always was.
That is the fork. If you want your filing curated, DEVONthink's structure is the reason to buy it. If what you have is twelve years of Downloads that nobody is ever going to curate, a tool that demands a decision about each file is the wrong shape, and one that reads the mess as it stands is the right one.
| What you're trying to do | DEVONthink | Loupe |
|---|---|---|
| Keep, organise and annotate a document archive | The entire product. Databases, tags, annotation, versioning, smart rules. | Nothing. It reads your folders and leaves them exactly as they were. |
| Find a scanned invoice by the supplier’s name | Works. It runs character recognition over scans you bring into it. | Works, on the folders you already have, with nothing imported. |
| Search files where they already sit, with no setup | Index a folder into a database first, and decide how it is organised. | Tick the folders you keep things in. That is the whole of it. |
| Find the photo of a whiteboard from a workshop | Reads text in an image. A photograph with no writing on it stays a filename. | Describes the scene with a vision model and indexes the description. |
| Find “the lease” when the document says “tenancy agreement” | Good odds. Its similarity engine has been doing this for twenty years. | Yes, by a different route: a meaning index alongside the keyword one. |
| Sync between two Macs and read it on a phone | Encrypted sync and a companion app on iOS. | Neither. One Mac, one index, no accounts. |
| Be certain nothing of yours is uploaded | Achievable — its AI can use a local model — but it is a setting you choose. | Not a setting. No network at runtime once the models are on disk. |
| Run it on an Intel Mac, or an older macOS | Supports macOS versions well before this one. | Apple silicon and macOS 26 or later, full stop. |
| Buy it | Several tiers, perpetual, with a year of updates and paid extensions after. | One product, every feature, monthly or once. |
What each one reads
Both read documents and both read scans. The gap is photographs. DEVONthink will pull text out of an image, which covers a screenshot of a conversation or a photo of a printed page. It does not describe a picture that has no writing in it, so the photograph of the whiteboard, the shelf of books, the serial-number plate on the back of the boiler stay findable only by filename and date.
Loupe hands those to a vision model that writes a plain description — a whiteboard covered in handwriting, three columns with arrows between them — and indexes the description as ordinary text. There is a page on how that works, including what it gets wrong.
Where DEVONthink wins outright
A long archive needs things a search tool does not have: revision histories, databases that record what was deleted, conversion to archival PDF, encrypted sync between machines, annotation that stays attached, automation you can script, a companion app on the phone, and twenty years of evidence that the company will still be there. None of that is on offer here and none of it is planned for this year.
It also runs on macOS versions well before this one. Loupe needs Apple silicon and macOS 26 or later, which with Apple ending Intel support in macOS 27 makes DEVONthink the only one of the two that an older Mac can run at all.
The honest summary
Buy DEVONthink if you want to keep an archive: curated, annotated, synced, and yours for twenty years. Try this if the problem is not the archive but everything outside it, and you want it searchable without first being organised. They overlap on one job out of ten, and plenty of people run both for exactly that reason.
Questions people ask about this
Does DEVONthink do OCR on scanned documents?
Yes, properly, and it has done for years. If somebody tells you a new semantic search app is the only way to make a scan searchable on a Mac, they have not looked at DEVONthink. The difference between the two is not whether scans get read. It is whether the documents have to be moved into a library first.
Does DEVONthink have AI?
Yes, and it can be pointed at a model running on your own machine rather than a cloud provider, so the privacy contrast is narrower than most comparison pages claim. The difference here is that there is no provider setting in Loupe at all, because there is nothing in it that would call one.
So what is actually different?
Where your files live. DEVONthink is built around databases you import into or index from, and it earns that structure back in sync, versioning, annotation and automation. Loupe reads the folders you already keep things in, adds nothing to them, and never writes. If you do not want a library, that is the whole answer.
Can I use both?
Yes. They do not compete for anything — no shared index, no shared database, no daemon fighting another daemon. A common arrangement is DEVONthink for the archive you curate deliberately and this for the twelve years of Downloads you never will.
Which is better for a huge document archive?
DEVONthink, without much argument. Versioning, revision-proof databases, archival PDF conversion, encrypted sync between machines and a companion app on the phone are the things an archive actually needs over a decade, and none of them are on offer here.
Try it on the folders you were never going to file.
No import, no database, no decisions about where anything goes. Seven days, no card, and nothing of yours leaves the machine.
- Apple silicon · macOS 26 or later
- No account
- Nothing leaves your Mac