AI File Organizer looks at what's actually inside a messy folder (the words in a document, the pixels in a photo, the audio in a clip) and proposes a new structure for it, instead of making you sort by hand. It reads content, not just filenames, and shows you a preview before anything moves.
Every part of that (reading, looking, listening) happens on your device. Nothing is ever uploaded; disconnect your internet after the download finishes and keep working. See the privacy policy. Built and maintained by CloudlessKit, an independent, ad-supported set of browser tools.
There's no one right way to sort a folder. Pick whichever strategy matches what you're looking at.
This is the default for a reason: it treats folders you've already made as ground truth. Any top-level folder with two or more files and a name that isn't auto-generated (not "New Folder", "untitled", or a bare year like "2019") becomes an anchor. Loose files get scored against the anchor's name and its existing files; anything that doesn't match confidently lands in "Unsorted" instead of being forced somewhere wrong. Files already inside your folders are never moved.
This one doesn't start from a fixed list of folder names: it groups similar files first and names the groups afterward. A group of fewer than three files folds into "Other", and two groups about the same thing get merged before naming, so you don't end up with both "Invoices" and "Receipts 2". Naming runs a handful of times per run, not once per file.
By Media Type separates documents, images, audio, and video into their own folders. By Date buckets files into year/month folders based on when they were last modified. By Project looks for files sharing both naming patterns and content, on the idea that "proposal_draft2.docx" and "proposal_budget.xlsx" belong together.
Type your own categories ("Client Work", "Personal Taxes") and files get classified against your labels instead of a built-in list. Matching is deliberately more forgiving here: a short label you typed scores lower than a full description of an invoice, so the bar has to bend or your labels would rarely win.
Five steps, whichever strategy you pick.
Chromium browsers get a native folder picker with write access, so the app can save the structure straight back to disk. Firefox and Safari fall back to a plain upload field and export the result as a ZIP instead.
Choose which file types to include and one of the five strategies above. "Transcribe audio" and "Suggest renames" are opt-in: nothing downloads you didn't ask for.
Before anything else, the app checks for duplicates. Then it reads documents, looks at images, and, if enabled, listens to audio and video, all on your device, staying responsive throughout.
First, a checklist of every category produced, checked by default; uncheck one and its files fall back to "Other" instead of disappearing. Then a side-by-side view of the original folder against the proposed one, switchable between topic, date, or a hybrid.
Files are copied a piece at a time, so even large folders don't bog down your browser, and each copy is size-checked against the source before it's marked done. Originals don't change unless you choose to remove them afterward.
A few things worth knowing about, some automatic and some opt-in.
Exact duplicates are found by grouping files by size first (one with a size nobody else shares is never opened), then comparing what's left by content fingerprint. Near-duplicates (a photo saved twice, a burst of near-identical shots) are caught by visual similarity, held to a strict bar since a false match costs more than a miss, and skipped above 2,000 images to stay responsive. Either way, duplicates move into "Duplicates (review before deleting)": nothing is deleted automatically.
For files named "IMG_4032.jpg" or "document (3).pdf", turn on AI-suggested renames. The app stays suspicious of its own output: chat filler, refusals, and generic words get thrown out in favor of your original name. Suggestions are pre-filled but editable, and nothing renames unless you leave it checked.
Every commit writes a small file into the new folder recording what got copied and where. Reverting only deletes paths resolving strictly inside that folder (an entry pointing anywhere else gets refused), so you can revert a run from a fresh browser session, days later.
Document understanding, image recognition, and, only if enabled, audio and video transcription are each a modest one-time download, cached after the first run. Later visits and folders don't re-download anything. The small model used to name folders and suggest renames downloads the same way, the first time you actually use it.