Auto-Sort Mac Downloads by Source: The AutoShelf Way

How do I automatically sort my Mac downloads by the website they came from?

The where-from metadata your Mac already records

Every file you download carries a hidden note about where it came from. macOS writes the origin into an extended attribute Apple documents as kMDItemWhereFroms, the same field Spotlight indexes. You can see it yourself: select a download, press ⌘I for Get Info, and the More Info section shows a Where From line. Reading it by hand tells you the origin but automates nothing. That value is what a source rule keys on, which is why sorting by website is possible at all.

One detail matters for writing rules. A downloaded file usually stores two where-from URLs: the direct address of the file on its server, and the web page you clicked the link on. Community reference confirms both entries show up when you inspect the attribute. A well-built rule can match either, so a PDF served from a CDN still files under the site you were actually on when you grabbed it.

Sorting downloads by the website they came from

AutoShelf is a macOS app that watches a folder and moves files based on rules you set, and the download URL is one of the conditions it can read. You point it at ~/Downloads, write a rule that checks the source URL against a domain, and name the destination folder. From then on the app files matching downloads the instant they land, with no drag and no second app to open. The demo above shows the shape of it: four files, four origins, four destinations, decided by where each one came from.

What source-based auto-filing actually does to your Downloads folder

Routing by website vs routing by file type

Most auto-sort advice stops at the extension: send every .pdf to Documents, every .dmg to a junk drawer, every image to Pictures. That breaks down the moment two files share a type but belong in different places. A Stripe invoice and a design brief are both PDFs, yet one belongs in Finance and the other in a project folder. Extension rules cannot tell them apart. Source rules can, because they read the origin instead of the file name, so a PDF from your bank and a PDF from a client site take different routes even though macOS sees them as the same kind of file.

The practical effect is a Downloads folder that stops being a graveyard. Files do not accumulate waiting for a manual tidy, because each one is filed at the moment it arrives. You stop losing the invoice you downloaded last Tuesday under forty screenshots, because the invoice never sat in Downloads in the first place.

Where-from metadata in practice

Source routing is only as good as the metadata behind it, and there is one honest limit. The where-from URL is written when a browser downloads a file, so files you copy from another Mac, pull out of a zip archive, or create locally often carry no origin at all. A source-only rule quietly skips those. The fix is to pair the source condition with a fallback, so a file with no origin still gets caught by its type or the app that produced it. AutoShelf lets you stack exactly that kind of fallback, which the next sections build toward.

Sorting by which app downloaded the file, not just the site

Why per-app routing beats extension-only rules

AutoShelf can route by which app downloaded a file, distinguishing Safari, Chrome, Arc, and Finder. That is a second provenance signal on top of the source URL, and it maps onto how people actually split their lives across browsers. If your job lives in one browser and everything else in another, the browser itself is a reliable proxy for “work” versus “personal,” and a per-app rule turns that habit into filing you never think about.

It is also the rarer capability. Source URL routing exists in a few organizers, but sorting by the app that did the downloading is uncommon: a MacRumors forum thread on the topic notes that no mainstream download manager advertises sorting by which browser or app fetched a file. That is the gap AutoShelf is built to fill, and it is why per-app routing shows up as its own column when the tools are compared later.

Build a multi-condition rule step by step

A worked recipe: file client PDFs by client site

Each condition you add makes the rule stricter, not broader, so one well-built rule can be surgical. Because the same rule can also delete or rename, you can express intentions like “trash any DMG from this site older than a day” without a second pass or a separate tool.

Templates that give you a head start

You do not have to build from an empty rule. AutoShelf ships templates for the common jobs, including a Sort by Source URL template, plus recipes for cleaning Downloads, auto-deleting old DMGs, and tidying the Desktop. A template is a working rule you adjust rather than author, so the first useful sort is a couple of edits away rather than a blank canvas.

Who this fits, and who should look elsewhere

When rules win

Rules win whenever the right destination is predictable from where a file came from. The clearest fits:

• Anyone whose Downloads folder is a dumping ground and who wants files filed the instant they land, not during a monthly cleanup that never happens.

• A freelancer or consultant pulling assets and invoices from a handful of recurring client sites, where the site reliably maps to a project folder.

• A developer who wants DMGs, datasets, and repo downloads routed by origin, and who would rather script the whole thing than click through a GUI.

• Anyone who splits work and personal across two browsers and wants the browser that grabbed a file to decide where it goes.

When content-aware AI wins

Provenance rules route by source URL and downloading app, not by contents. If your sorting logic depends on what is inside a file, which client an invoice is for based on the total, what a screenshot actually shows, or the subject of a document, a content-aware AI organizer that reads PDF text and runs OCR is the better fit. It is also the honest boundary of a rules engine: AutoShelf is fast and deterministic precisely because it does not try to understand the file, only where it came from. Pick the tool by the job you actually have.

Driving the organizer from the terminal or an AI agent

Creating and running rules from the shell

AutoShelf ships a CLI, so rules, watched folders, and actions are all reachable from the terminal. That matters for anyone who already lives in a shell: rules can live in version control, ship in a dotfiles repo, and be recreated on a new Mac with a script instead of a click-through. The watcher runs the same way whether you set it up in the GUI or the command line, so the CLI is a full second front end, not a limited afterthought.

Letting an AI agent maintain your rules over time

The newer piece is the MCP server. Model Context Protocol is the standard local AI assistants use to reach real tools, and AutoShelf exposes its rule management through it. In practice that means you can tell an assistant like Claude or Codex “stop filing newsletters into Work” or “add a rule for invoices from this new client,” and it edits the rules for you while the background watcher keeps sorting. Your organization system becomes something you can adjust in plain language, which is a genuinely different way to maintain rules than opening a preferences window every time your filing needs shift.

How it compares to Hazel, File Arbor, and AI organizers

The honest price picture

On price, AutoShelf comes out ahead of the tools that do the same job. It undercuts Hazel, the category’s mature standard, and adds a free single-rule tier that Hazel does not offer. The bars below line up the rule-based organizers by their one-time cost.

Being fair means naming where AutoShelf does not win. If you only want a one-shot cleanup, a simpler tool like Folder Tidy costs a few dollars and is plenty, because it does not watch folders or read where files came from at all. Subscription AI organizers such as Sortio sort by what is inside a file, which rules cannot do, at an ongoing monthly cost rather than a single payment. And the free route exists: macOS Automator Folder Actions can auto-sort incoming downloads for nothing, though they are commonly limited to file-type rules and do not read website or downloading-app provenance.

Where each tool is genuinely stronger

Hazel earns its price with a long track record and a deep ecosystem; if maturity is what you value most and $42 with no free tier is fine, it is the safe pick. File Arbor is the answer if you need the same rule engine on Windows or Linux as well as macOS, a place where a Mac-only app cannot follow. AI organizers win the moment your sorting depends on file contents rather than origin. AutoShelf’s own ground is narrower and sharper: it is the tool that routes by source and by the app that downloaded a file, scriptable from a CLI, editable by an AI agent through MCP, at the lowest one-time price of the rule-based group, with one rule free to try first.

Frequently asked questions

Sources

1. 1

   Apple Developer: kMDItemWhereFroms metadata attribute

   Apple’s own reference for the metadata key macOS uses to store where a downloaded file came from, the value every source rule reads.

2. 2

   Apple Support: Get info about a file or folder

   Shows how to open Get Info and see a file’s Where From line yourself, proving the origin data exists before an app automates on it.

3. 3

   Apple StackExchange: Finding the URL of a downloaded file

   Community reference showing mdls returns two where-from entries, the direct file URL and the page you clicked it from, which affects how source rules match.