NeoDownloader 6

Light, dark, or whatever the system is set to — switched on the fly, with no restart. The accent is one of twelve (teal, blue, violet, pink, red, orange, green and the rest), and the whole interface repaints to match, project cards and progress bars included.

The earlier versions spoke English and nothing else. Version 6 speaks thirteen languages, and the language is chosen inside the program, with no restart. Is yours missing? Write to us — we will translate into any language, if you are ready to check the result in the running interface afterwards.

Looking through what you downloaded, without leaving the project, has become considerably easier. A picture opens full-screen with a strip of thumbnails along the bottom — the rest of the project stays at hand, and paging through it never means leaving full screen. The player is faster: it decodes on the graphics card and carries video up to 8K, where version 5 stopped at 4K. PDF and the text formats are new — txt, md, csv, json — so telling whether a file is the one you wanted no longer needs another program at all.

The program used to be a monolith: the browser, the page parsing and the downloading all lived inside a single file, and a new way of fetching meant a new release of everything. Now the work with a source belongs to a plugin — a separate program with its own runtime and its own dependencies, up to a browser of its own. Four arrive in the box so far: yt-dlp for video, gallery-dl for galleries, a crawler of our own, and one for a plain direct link. Paste an address and they answer which of them can handle it; the project goes to the one that did. Each is added, updated and removed on its own, and a plugin that falls over does not take the program down with it.

yt-dlp is what video is downloaded with these days — the actively maintained successor to youtube-dl, reaching more than fifteen hundred platforms. It takes a video at the best quality on offer, with its sound, subtitles and chapters; it takes whole playlists and channels; and it catches a livestream from its start rather than from the moment you arrived. yt-dlp itself is a command-line tool; here the quality is a list to pick from, and the destination folder and the log belong to the application.

gallery-dl is the same thing for pictures: close to three hundred sites where what matters is an author, an album, a tag or a collection rather than a page. It follows pagination to the end, and it can sign in as you do — by password, by the cookies already in your browser, or through OAuth — so it sees what you see. Files come out named by rules rather than dropped in a heap. Also a command-line tool by nature, also in the box, and also without a single command typed.

The plugin we wrote ourselves, for everything the two engines above do not cover: it walks the site — a gallery, a profile, a board, a forum thread, a page of search results — and collects what it finds. What to look for on a particular site is written in a recipe, a small file beside the plugin rather than code inside the program. Recipes come for 500px, Bing Images, Flickr, Google Images, IMDb, Reddit and others, plus general ones — a forum thread, a gallery, a single page, and a whole site at once.

The window where the program is taught a new site: a recipe for the site you want is assembled on the live page itself — a tree of what the recipe extracts on the left, the page on the right, and the element you want chosen with the mouse. Pointing stands in for knowing markup, so a recipe is within reach of somebody with no experience of it at all. The same window mends a recipe that stopped working after a site changed its layout. Still being written — part of it works, part is under construction, and the application says so in advance.

An address that points straight at a file is downloaded without parsing anything. So is a range like photo[1-50].jpg — the fusker of version 5: one pattern expands into fifty addresses.

The code that handles a site sits in the installation folder in plain Python, readable as it is: open it to see exactly how a site is walked, change it for your own purposes, or take it as the starting point for something else. Keeping it uncompiled was the decision that makes this possible, and it was taken for exactly this reason.

A plugin is a separate Python process, not a class inside ours, and the SDK is MIT precisely so that somebody else’s plugin has the right to link against it. A recipe for a particular site is assembled in the editor out of selectors, and where selectors run out, out of a handler written in Python or JavaScript.

Start a project a second time and it collects only the files that are missing, rather than everything again: what has already been saved is remembered file by file and left alone. Version 5 called this a smart update. Here it is simply how a run behaves, with nothing to switch on.

The earlier versions ran on Windows and nowhere else — that is the system they were written for. Version 6 treats the operating system as a detail: the Windows build is ready and signed, and it is the one on this page. Builds for macOS and Linux are coming soon — the core is cross-platform already, and what remains is packaging and signing.

A running copy will be reachable from outside, and it will look exactly the way it does on your own machine. Manage downloads from a phone, or from a laptop in another room, watch them run, and page through what has already been downloaded — pictures, video, sound, documents.

The program can be driven from a script of your own: create a project, start it, and collect the result — all of it ordinary requests to the core. The interface uses that same API, so from outside you can reach exactly what it reaches. And the core itself will run with no window at all — on a server, say.

Keeping recipes separate from the program is for exactly this: a site breaks, and what reaches you is a file rather than a new version. Today a broken recipe can be repaired on your own machine, but nothing delivers one to you — the channel that would is not built.

Restart a project on a timetable and take only what is new — what version 5 called a smart update, with nobody at the keyboard. The core is made to run without an interface; the scheduler is the next thing to be written.

The program notices that a new version exists and says so. Downloading and installing it is still your click.