FileMorrow

Organize your Downloads with on-device Apple Intelligence—guided by your own rules.
Private, reversible, and made for macOS by Nabeegh.

FileMorrow keeps fresh files in Today, Yesterday, and Last 7 Days views. Only older loose files become eligible for category folders. Downloaded folders are never moved or reorganized, and every organization batch can be undone.

	Format mode	Smart Content mode
Default	Yes	Opt-in
Classification	Known extension and system file type	Filename, local extracted evidence, then Apple Foundation Models when needed
Apple Intelligence required	No	Yes, for unresolved content
Predictability	Highest	More meaningful subject folders, but suggestions can be wrong
Privacy	Local	Local, using Apple's on-device model
Best for	Automatic low-maintenance cleanup	Mixed-subject University, Finance, Medical, Legal, and Work documents

Format mode is the recommended default. Smart Content is useful when subject matters more than extension, but the model is not perfectly accurate. FileMorrow therefore uses deterministic evidence first and leaves uncertain items visible for review.

Highlights

• On-device classification with Apple Foundation Models
• Predictable By File Format mode with no AI queue
• Optional Smart Content mode using metadata, extracted content, and Apple Intelligence
• Fast local evidence scoring before invoking the language model
• Content extraction for PDF, DOCX, PPTX, XLSX, ZIP, text, code, and images
• Local OCR with Vision
• Persistent AI decisions and user corrections
• Quick Look file previews and one-click opening in the file’s default app
• A Teach Organizer workflow for corrections, filename rules, format rules, and reusable on-device model examples
• Separate Awaiting Analysis and Needs Review states
• Analyze-next and continuous Analyze All modes with a Stop control
• Editable, importable, and exportable organization profiles
• User-selectable and fully custom categories
• Confidence-gated organization
• Plan-first approval before any eligible file moves
• Explicit Undo Last Organization, Command-Z, and collision-safe moves
• Progress-aware, cancellable exact duplicate detection with SHA-256 and recoverable Trash cleanup
• Choose which copy of a duplicate group to keep before anything is removed
• Extracted Archives: reclaim space from ZIP files you already unpacked
• Installers: find .dmg and .pkg files for software already installed
• Top-level-only organization: downloaded folders and their contents are never moved
• Read-only recursive duplicate scanning, with explicit confirmation before Trash
• Color-coded Finder icons distinguish FileMorrow-managed category folders from ordinary folders
• Menu-bar companion for status, rescanning, duplicate checks, and reopening the app
• Optional menu-bar-only mode with a Keep FileMorrow in the Dock setting
• Native SwiftUI interface and Settings window
• No analytics, accounts, cloud API keys, or content uploads

Requirements

• macOS 14 Sonoma or later for the app itself
• macOS 26 on an Apple Intelligence-eligible Mac, with Apple Intelligence enabled, for the optional Smart Content mode
• Xcode 26 or later to build from source

Everything except Smart Content works on macOS 14 and later: format and filename rules, organization and undo, duplicate detection, extracted-archive and installer cleanup, teaching, profiles, and file previews. Foundation Models is weak-linked and only ever called on macOS 26, so the app launches normally where the framework does not exist.

The app checks Foundation Models availability on launch under Settings → Compatibility. If Apple Intelligence is unavailable, deterministic rules, file previews, and manual teaching continue to work; only model analysis is disabled with an explanation. Apple documents the runtime availability states in SystemLanguageModel.Availability.

Build

swift build -c release
swift test

Package a standard ad-hoc-signed macOS application:

./Scripts/package-app.sh

The app is written to dist/FileMorrow.app.

Create local ZIP and DMG release artifacts with SHA-256 checksums:

./Scripts/create-release.sh

GitHub release packages are ad-hoc signed and include SHA-256 checksums. They are not notarized because this independent open-source project does not currently have an Apple Developer membership.

Install a GitHub release

1. Download the DMG or ZIP and SHA256SUMS.txt from the same release.
2. Verify the checksum with shasum -a 256 -c SHA256SUMS.txt.
3. Move FileMorrow to Applications and try to open it once.
4. If macOS blocks it, open System Settings → Privacy & Security, scroll to Security, and choose Open Anyway, then authenticate and confirm.

Only override Gatekeeper when the download came from the official FileMorrow GitHub repository and its SHA-256 checksum matches. Apple notes that software from an unidentified developer has not been reviewed by Apple, so users should make this exception only for software they trust. See Apple’s Open Anyway instructions.

Safety model

FileMorrow defaults to By File Format mode: documents, audio, images, videos, spreadsheets, installers, archives, design files, and code are placed by known extensions and system file types. This is fast, predictable, and does not require an analysis queue.

Users can opt into Smart Content mode. It considers filename metadata, extracts a short local evidence sample, and uses Apple's on-device model to suggest subject folders such as University, Finance, or Medical. These are suggestions: the model can make mistakes, so important files should be reviewed. Low-confidence items remain in Needs Review. Organization creates ~/Downloads/<Category> and records every move for undo.

FileMorrow persists each top-level file's first-seen date locally. Moving or undoing a file therefore does not reset the seven-day eligibility window.

The organizer operates only on loose regular files directly inside ~/Downloads. Existing, newly created, and newly downloaded folders are always left in place, and the organizer never moves or modifies their contents. The duplicate finder can read files recursively to compare exact SHA-256 hashes, but it never removes anything without explicit confirmation. The same rule applies to extracted archives and installers: they are found by reading, listed for review, and only the archive or installer file itself is ever moved to Trash.

Menu bar

The menu-bar companion remains available when the main window is closed. It can reopen the app, rescan Downloads, prepare a manual organization plan, check for duplicates, check for already-unpacked archives, check for used installers, open Settings, or quit. Automatic Organization and Launch at Login default to on for new installs and are clearly presented during onboarding. Leaving Automatic Organization enabled grants ongoing consent for hourly organization without repeated prompts. Only loose files older than the configured age (seven days by default) are eligible, uncertain files remain for review, and every completed batch can be undone.

Turn off Settings → Organization → Keep FileMorrow in the Dock for a menu-bar-only experience. The app continues running after its windows close and can be reopened from the menu-bar companion.

Choose Show Welcome Guide from the FileMorrow app menu, menu-bar companion, or Settings to reopen onboarding at any time.

Managed folders

All Downloads also presents files directly inside FileMorrow-managed category folders as a read-only library. Managed folders carry a hidden marker; arbitrary user/downloaded folders are not displayed or organized. Only the explicit duplicate scan reads them to compare hashes. Organized files are labeled and excluded from the seven-day queue so they cannot be moved twice.

Extracted archives

Downloads fill up with ZIP files that were unpacked months ago and never deleted. Extracted Archives finds them and offers to move just the archive to Trash.

An archive is only listed when the evidence is complete:

1. Every file entry in the ZIP exists in the unpacked folder at the same relative path.
2. Every one of those files matches the entry's uncompressed size exactly.
3. Bookkeeping macOS discards during extraction (__MACOSX, .DS_Store, AppleDouble ._ files) is ignored, and an entry that would resolve outside the destination voids the whole archive.

A partial extraction, a renamed folder, or a single edited file is enough to leave an archive alone. Both common shapes are recognised: report.zip unpacked to report/, and a flat archive unpacked into a folder named after it.

Nothing is automatic. Results are listed with the unpacked destination, the verified file count, and the reclaimable size; the user selects what to remove. Each archive is verified against its unpacked folder one final time at the moment of deletion, so an archive whose folder was moved or emptied in the meantime is left in place. Only the .zip file moves to recoverable macOS Trash. The unpacked folders are never touched.

Used installers

Installers finds .dmg and .pkg files in Downloads for software that is already on this Mac. Nothing is mounted, opened, or run.

Two independent kinds of evidence are used:

• A .pkg declares the package identifiers and versions it installs. FileMorrow reads that metadata straight out of the archive and checks it against the install receipts macOS keeps for packages that were actually installed. Every declared identifier must have a receipt at the packaged version or newer. This is an exact match, not a guess.
• A .dmg is matched by name to an app in /Applications or ~/Applications, comparing the version in the filename against the installed bundle's version.

An installer newer than what is installed is never listed. That is an update the user has downloaded but not applied yet, and deleting it would lose the update.

The common case runs the other way: an app that updated itself since the installer was downloaded is now ahead of it, which makes the file on disk clearly stale. Those are shown as "Already updated past this version".

As everywhere else, results are reviewed and selected by hand, only the installer file moves to recoverable Trash, and installed software is never touched.

Privacy architecture

flowchart LR
    D["Loose files in ~/Downloads"] --> A{"Older than 7 days?"}
    A -- "No" --> F["Fresh views; no move"]
    A -- "Yes" --> R["Format and filename rules"]
    R --> C{"Smart Content enabled<br>and evidence unresolved?"}
    C -- "No" --> Q["Confidence gate"]
    C -- "Yes" --> E["Bounded local extraction<br>PDF, Office, text, Vision OCR"]
    E --> L["Local evidence classifier"]
    L --> M["Apple on-device<br>Foundation Model if needed"]
    M --> Q
    Q -- "Approved" --> O["Top-level managed folder"]
    Q -- "Uncertain" --> V["Visible review queue"]
    O --> U["Undo history"]
    X["All accessible files under Downloads<br>read-only"] --> H["SHA-256 duplicate scan"]
    H --> T["User-selected extras to Trash"]
    Z["ZIP files in Downloads"] --> Y["Entry-by-entry size verification<br>against the unpacked folder"]
    Y --> T
    I["Installers in Downloads"] --> P["Package receipts and<br>installed app versions"]
    P --> T

Loading

There is no server in this path. FileMorrow has no account, analytics SDK, advertising, cloud API key, or file upload code. Extracted evidence is bounded and used locally. Network access is needed only when a user downloads the app or when macOS itself prepares Apple Intelligence.

First launch

Before automatic organization or Launch at Login registration, onboarding explains:

1. Files stay loose for seven days.
2. The organizer never moves downloaded folders or anything inside them.
3. Format mode is the predictable default; Smart Content is optional.
4. Organized batches can be undone.
5. Duplicate cleanup uses exact SHA-256 matching and recoverable Trash.
6. Smart Content availability depends on the Mac and Apple Intelligence setup.

Project structure

• AppState.swift — application state and workflows
• ContentExtractor.swift — PDF, Office, archive, text, and OCR evidence
• AIClassifier.swift — structured on-device classification
• RuleClassifier.swift — deterministic and semantic fast path
• PersistenceStore.swift — decisions and move history
• OrganizerService.swift — collision-safe move and undo
• DuplicateScanner.swift — fingerprinted SHA-256 duplicate detection
• ExtractedArchiveScanner.swift — verification of already-unpacked archives
• InstallerScanner.swift — receipt and installed-app matching for installers
• Views.swift — native macOS interface
• Configuration/default-profile.json — general-purpose format and category catalog

Custom categories

Open Settings → Categories to choose which categories are active, edit their format and keyword guidance, or add completely new categories. Profiles can be imported and exported as JSON, making them easy to share or version. See docs/PROFILES.md for the schema and classification order.

Preview and teach

Select a file to see its native Quick Look preview. Open File launches it in the user’s default app. Teach Organizer… corrects the current file and adds it as an example for the on-device model. A user can optionally provide a reusable filename phrase or explicitly map the file extension to that category. Extension rules are opt-in because broad formats such as PDF and PPTX can cover many different subjects.

Privacy

File contents stay on the Mac. The project does not include telemetry, networking, advertising, or third-party analytics.

Duplicate detection is the one workflow that recursively reads accessible folders inside Downloads. It hashes bytes locally and never reorganizes nested content. Nothing is removed automatically: users review each exact-match group, and confirmed extras go to recoverable macOS Trash. Identical nested project or app files may be intentional, so the full paths must be reviewed before cleanup. The organizer itself remains strictly limited to loose top-level files.

Accuracy and compatibility testing

The suite covers organization and undo, duplicate detection and its refusal paths, archive verification against real ZIP files built during the test run, installer matching against a real package produced by pkgbuild, and profile persistence across relaunches.

SyntheticAccuracyTests generates privacy-safe fake PDFs, presentations, and spreadsheets at test time. It verifies local extraction and subject classification without committing personal documents. Ambiguous filenames and content must remain uncertain. Compatibility copy is covered for available, Apple Intelligence disabled, ineligible-device, model-not-ready, checking, and unknown states.

The packaged-app clean-profile preflight uses a temporary home and empty settings:

./Scripts/clean-profile-smoke-test.sh

See docs/CLEAN_MACHINE_TEST.md for the required second-Mac matrix. A local simulated profile does not replace physical testing with Apple Intelligence both enabled and disabled.

License

MIT