Offline Semantic File Search: Find Files by Content

How do I find a file when I forgot the name but remember what's in it?

Why Spotlight and Windows Search fall short

The built-in search on both platforms is organized around the one thing you have forgotten: the name. Spotlight and Windows Search index filenames, file types, and metadata, and they do full-text matching on documents that carry a text layer. That is fast when you know a distinctive word that is literally written in the file. It falls apart the moment your memory is about meaning rather than text: a photo has no words to match, a scanned receipt is just pixels, and a report you think of as the pricing deck was saved as Q2-final-v3.pdf. Recent-files lists and date filters help only when the thing you want is also the thing you touched last.

What semantic search matches instead

Semantic search indexes what a file is about, not the characters in its name. It reads each document and looks at each image, turns that into a numeric fingerprint of the meaning, and stores it so you can query it later. When you type a description, it compares the meaning of your words against those fingerprints and ranks the closest ones first. That is why you can find a photo by describing the scene in it, or a contract by its topic, without knowing a single exact word that appears in the file. Dotient is one such tool, built to run that entire loop on your own computer, which is the part the next sections get into.

Find files by describing content instead of filenames

Search photos and documents by content

The reason describing a file works is that Dotient indexes two kinds of things the same way. A PDF, a spreadsheet, or a note becomes a fingerprint of its text and topic. A photo, a screenshot, or a design export becomes a fingerprint of what is actually visible in it. Both land in one searchable index, so a single description can pull back a document and an image side by side. Ask for the invoice with the blue logo from last spring and the actual invoice can surface next to a screenshot of the billing email, because both were indexed by what they contain rather than by whether the word invoice happens to be in the filename. That combined coverage, text and visual in one query, is the capability most built-in and single-purpose tools do not offer.

Natural-language queries, not keywords

Because the match is on meaning, you do not have to guess the exact term the file uses. A keyword search for receipt misses a file that only ever says invoice or statement. A semantic query understands those are close in meaning and returns them anyway. You can describe a scene (a whiteboard covered in sticky notes), a subject (the contract with the penalty clause), or a look (the dark screenshot with a red error), and get ranked results without tags, folders, or a naming convention you kept up for years. The point is not that you type more, it is that you type what you actually remember.

How Dotient indexes and searches your files on-device

An on-device model, with no cloud upload

The engine is a quantized vision-and-text model that embeds your files into fingerprints right on your computer. On the visual side Dotient uses embeddings from a SigLIP image model, so a photo is understood by its contents rather than its filename. Search itself is a hybrid: Dotient runs classic BM25 keyword matching and semantic matching at the same time, so an exact term you do remember is never silently lost while the meaning-based match handles everything you can only describe. All of it, the indexing and the querying, stays local. Your files never leave the device, which is the whole reason a person with sensitive documents can point this at their real archive instead of a folder of samples.

A graph view, Cubbies, and Signals

Search is the front door, but Dotient also gives you ways to work with the results. A node graph plots your whole archive with edges drawn between similar files, and it auto-detects clusters, so you can see the shape of what you have rather than scrolling a flat list. You can group related files into named collections called Cubbies to keep an archive organized. And when a search almost gets it right, Signals let you fine-tune the ranking toward what you meant, nudging the results instead of retyping the query five different ways. These are the features that turn a one-shot search box into something you can actually live in.

Where semantic search helps, and where it misses

Accuracy depends on the model

Being honest about the limits is the point of a buy-once directory, so here it is. A semantic match is a similarity score, not a lookup, which means the file you want is usually near the top but not always the first row, and a very generic description (the document about the project) gives the model little to lock onto. Visual search is bounded by what the vision model can recognize, and a scanned page with no embedded text is pixels until something reads the characters. Dotient softens each of these: the parallel BM25 keyword pass catches exact terms the semantic side might rank low, and Signals let you correct a near-miss by hand. What it does not do is promise the top result is always the right one, and you should not expect that from any tool in this class.

When plain filename search is still faster

Semantic search is not a replacement for everything. If you already know the exact filename, the built-in search is instant and you should just use it. If your files are meticulously named and foldered and you never lose one, the payoff here is small. And if what you need is bulk renaming or auto-sorting rather than finding a specific file you have in mind, that is a different job with different tools, some of which show up in the comparison below. Dotient earns its place precisely in the gap those leave: the messy, half-remembered archive where the name is the one thing you do not have.

Who Dotient is for, and who it is not

Best fit: big private archives you search by memory

The clearest fit is anyone whose files pile up faster than they get named well:

• A researcher or writer sitting on years of PDFs and notes, who remembers the argument in a paper but never its filename.

• A designer or photographer with thousands of screenshots and images, who wants to find one by describing what is in the frame.

• A developer or student whose downloads and exports accumulate into a flat heap that Spotlight only helps with when they already know the name.

• Anyone privacy-conscious or often offline, who will not upload personal or client files to a cloud service just to make them searchable.

Weaker fit: tidy libraries and pure organizers

If you keep a disciplined naming scheme and a clean folder tree, the built-in search already covers you and the value here is thin. If your goal is to auto-sort a downloads folder or batch-rename hundreds of files rather than find a specific one, an organizer or a renamer is the better tool, and the comparison below names a few. Dotient is a find-the-file-you-mean tool first, and it is happiest against a large archive you navigate by memory rather than by structure.

How Dotient compares to other offline file-search tools

A one-time license, with an optional add-on

Dotient is a one-time purchase, and it is worth being precise about the tiers so the no-subscription claim is honest. The $5 Standard license is a one-time buy that works fully offline on its own, with three license keys and one AI model. A $15 Lifetime license, also one-time, adds new AI models monthly, Adobe file-format support, priority support, and all future updates. There is one recurring option, a Plus add-on at $1 per month, which layers the same extra models and format support onto a Standard license, but it is optional and the app is fully usable without it. So the accurate framing is not no subscription anywhere, it is a working one-time app with an optional subscription on top, and the base search runs offline either way.

Where a rival is the better pick

A spotlight is only worth trusting if it says when something else fits you better, so here are the honest calls. LocalSynapse is free and open-source and searches document contents across a dozen-plus formats offline, and if you only ever search text and want to pay nothing, it is the obvious start, though it does no image search. File Brain runs offline with built-in OCR that indexes scanned PDFs, so if your archive is mostly scanned paper it reads pages Dotient cannot, though its visual search sits behind a paid tier. Kalycs is a lifetime purchase but needs your own OpenAI key, which breaks the fully-offline promise. Talyx is an on-device organizer that sorts and classifies rather than finding one file you already have in mind. And Mylio consolidates photos across your devices but is a subscription with no describe-the-contents search. Dotient's specific edge is the combination: photos and documents together, described in plain language, fully offline, at the lowest confirmed price, on both macOS and Windows.

Frequently asked questions

Sources

1. 1

   Apple Support: Search with Spotlight on Mac

   Apple’s own reference showing Spotlight is organized around filenames, types, and metadata rather than a description of what a file contains.

2. 2

   Microsoft Support: Search for anything, anywhere in Windows

   Confirms Windows Search’s filename-and-index model, so the contrast with describing a file’s contents is verifiable rather than asserted.

3. 3

   Google Machine Learning: Embeddings

   A plain-language explainer of the vector embeddings that let a tool match a description to a file by meaning, and why the results are ranked by similarity rather than exact.