Skip to content

Latest commit

 

History

History
2359 lines (2065 loc) · 156 KB

reference.md

File metadata and controls

2359 lines (2065 loc) · 156 KB

Download Engine detailed reference

For a quick introduction and setup, see the README. This guide keeps the detailed feature behavior, limits, troubleshooting and verification notes. Paths and commands are relative to the project root unless noted otherwise.

An original Windows download manager built with Rust and a native egui interface. Browser add-ons use plain JavaScript. There is no TypeScript or Node requirement to run it.

Download Engine's original code is available under the MIT license. Bundled dependencies retain their licenses; see third-party notices.

English, Simplified Chinese, Traditional Chinese, Malay, Japanese, Korean, Vietnamese, Thai, Polish, Russian, Ukrainian, Arabic, German, French, Swedish, Norwegian Bokmål, Swiss standard German, Italian and Romansh are available in the desktop app, offline quick guide, Windows installer and all three extensions. Choose a desktop language in Preferences → Appearance → Language. Each extension has its own selector in Browser options. See Languages for defaults, font support and translation maintenance.

Download Engine was called XDownloader before 2.0, with programs named idm-ex.exe, idm-ex-cli.exe and idm-ex-host.exe. The programs are now de.exe, de-cli.exe and de-host.exe, with de-torrent.exe handling optional torrent downloads since 3.7.0. Upgrades keep existing data and download folders, saved site sign-ins, the Start with Windows entry and browser connections; these keep their internal IDM_EX / idm-ex / XDownloader identifiers.

Install or use the portable version

Public unsigned Windows builds are in dist\releases&lt;version>: DownloadEngine-<version>-UnsignedSetup.exe and DownloadEngine-<version>-Unsigned-windows-x64.zip. These files have no Authenticode publisher signature; Windows may show an unknown-publisher warning. See GitHub release instructions for packaging, source and license materials, and SHA-256 checksums. Setup installs per-user in %LOCALAPPDATA%\Programs\Download Engine without administrator access and adds a Start menu shortcut; a desktop shortcut is optional. Upgrades stay in the existing folder, including %LOCALAPPDATA%\Programs\IDM EX from before 2.0, replace the old programs and shortcuts, and point Start with Windows and connected browsers at de.exe and de-host.exe. The final screen can open Browser setup.

The stateful local unsigned development profile uses dist\IDM-EX-UnsignedDev\de.exe. It preserves private test history and downloads and must not be copied or repackaged for distribution. Clean portable test builds use dist\DownloadEngine-<version>-UnsignedDev-windows-x64.zip; test installers use dist\DownloadEngine-<version>-TestSetup.exe. Signing and browser-store publication remain separate release steps.

Installed copies keep queue history in %LOCALAPPDATA%\Download Engine. Each launch asks Windows for the current Downloads folder, including a location moved to another drive; new files use its Download Engine subfolder. A profile created before 2.0 keeps %LOCALAPPDATA%\IDM_EX and uses the IDM EX subfolder of the current Downloads location until a Download Engine data folder exists. The folder is looked up without creating it; normal startup prepares it and reports errors. An unavailable lookup is not replaced with a guessed user-profile path. Existing downloads and partial files keep their recorded locations and progress. Use Tasks → Exit (or Exit in the tray menu) before an upgrade or uninstall. Uninstall preserves queue history, partial files, and downloads, and removes only browser registrations pointing to the removed copy. Cancelling uninstall leaves those registrations alone. Setup refuses to install over a portable folder. Portable history is not moved or merged automatically.

Upgrade notes by version

Some releases need extra steps beyond the general upgrade steps.

Upgrading to 3.15: Provider core APIs add protected binary/RSA imports, HTTP methods, scoped download credentials, unknown-size and streaming segments, larger manifests and ASS subtitles. Using the new credential types or media features requires profile writer version 9; back up your profile before downgrading. Site-specific protocols remain extension code.

Upgrading to 3.12: Lua extension packages in the app's plugins folder appear for permission review; discovery does not install or enable them. The bundled episode downloader includes editable Lua and generated offline media. See the extension guide and bundled walkthrough. Media providers, available from 3.11, add segments, multiple tracks and muxing. Saving the first media job raises profile compatibility to 7; older builds refuse that profile. Back it up before returning to an older app version.

Upgrading to 3.6: the Scripting page is now Extensions, and your scripts, rules and run history are on its tabs. Installing your first extension upgrades the profile, so older versions then refuse it. Back up the profile before going back to an older version.

Upgrading to 3.0: the Firefox extension now has the identity download-engine@xetro. Finish any pending browser handoffs, remove the old developer-loaded extension, and install the new one, then reconnect it through Tools → Browser setup. Firefox treats it as a different extension, so its settings start fresh; your desktop queues, downloads, and profile stay in place. Reload developer-loaded Chromium extensions after replacing their files. See the browser setup guide.

Start

Open Help → Guide (or press F1) for offline instructions, a topic list beside the article, and copyable Lua and PowerShell examples. The same user guide ships in the portable package and installer. It covers downloads and recovery, browser integration, queues and schedules, automation, CLI usage, settings and diagnostics.

Open Help → Changelog for offline release notes, newest first. The same changelog ships in both packages. Every delivered update gets a new version and release note; see the release workflow.

The Extensions page in the top toolbar can route new downloads by site or file type, apply a speed limit, and start another queue after completion. Rules have enable switches and a text run history. The packaged de-cli.exe provides JSON commands for PowerShell and other scripts, using the same scheduler as the app. See CLI and automation for commands and examples.

Extensions → Scripts adds a built-in Lua editor with optional PowerShell, Save, Run, Stop, timeouts and persistent coloured text logs. Choose manual, startup, interval, download completion or queue completion triggers. The included example uses Lua's xd app API and event for trigger context. PowerShell scripts use Invoke-DownloadEngine and $Event. Scripts run while the app is open or in the tray; only run code you trust. Start with New script → Save script → Run script, then scroll to Run logs. Open Lua guide above the Lua editor or Help → Guide → Lua scripting for a first-run walkthrough, working download and completion-trigger examples, the app API and troubleshooting. Each Lua block has Copy script. The Lua scripting guide ships offline in both packages. Help → Guide also covers recovery and retention.

MCP on the toolbar opens a popup with AI connection controls: Start/Stop, port, View only or Allow changes, optional startup with the app, and a real connection check. Use Copy client config to connect a compatible local AI client over HTTP (default http://127.0.0.1:17842/mcp). MCP starts disabled with view-only access. Setup guide on the MCP popup (or Help → Guide → MCP integration) explains the steps. A terminal server is also available through de-cli.exe mcp; mcp-config prints its connection settings and mcp --read-only restricts it to inspection. CLI flags are independent of the desktop settings. Downloads, queues and saved-script controls use the same scheduler and receipts as the CLI. Closing the popup keeps your current page and pending edits. The server stays running when the popup closes or clients disconnect. See the MCP connection guide and AI usage guide. AI clients can read the latter through get_usage_guide.

For the local development build, open de.exe inside dist\IDM-EX-UnsignedDev (or in the extracted portable package). Choose Add URL, paste a direct HTTP/HTTPS/FTP/FTPS file URL, select a queue, and optionally choose a filename and an absolute destination folder. Select Start download, or Add to queue when files are being held for later. An idle, unscheduled Main queue starts automatically. Cancelling a queue choice in this form leaves the current download view unchanged. FTP filenames come from the URL unless you supply a name. For HTTP/HTTPS, leave Filename blank to use a server-provided name when available. New unnamed downloads read Content-Disposition from the first accepted transfer response, preferring its Unicode filename* form (UTF-8, with legacy ISO-8859-1 support). Missing or unusable metadata keeps the URL-derived name. A malformed extended parameter can fall back to a valid plain filename in the same header; duplicate parameters, multiple headers, or headers over 8 KiB are ignored. Server paths are discarded and the final name uses the same Windows filename sanitization and non-overwriting publication as other downloads. Names you type and browser-supplied names take priority, even if they match the URL name.

Once a transfer starts, its name stays fixed across retries, pause/resume, restarts, and expired-link replacement. Existing history keeps its names. The save folder shown when adding stays fixed too; a later server name can change the displayed category, but does not move the download to another folder. The parser follows Content-Disposition and extended filename parameters.

The native interface follows the Classic HTML mockup. The toolbar offers Add URL, Resume, Pause, Pause all, Remove, Open folder, Clear finished, Scheduler, and Preferences. A file without a resume plan shows Start again instead of Resume. Open folder launches File Explorer directly; there is no extra confirmation dialog. Remove is disabled during an active transfer.

Press Ctrl+N to add a download, or Ctrl+F to focus filename search from any page. While search has focus, Escape clears it; a second Escape releases focus. Shortcuts leave open dialogs and menus alone. × and Escape close Add download and keep its URL, name and folder draft; login fields are cleared.

Use Browse… beside destination fields to choose a Windows folder. Cancelling the picker leaves the draft unchanged. Checksum, authentication, and ZIP preview options are under Advanced in Add download. Use login reveals username and password fields directly in the form. Preferences opens a popup with Appearance, Downloads, Connection, Browser & AI, Antivirus, and Startup & tray categories. Changes to the popup's controls wait for Save; Cancel, ×, or Escape discards pending changes. Themes preview while editing. Reset changes only the current category's draft. Detailed editors retain their own explicit saves; closing them returns to the popup. Table-column edits stay in its pending draft. Dialogs share the same compact frame, icon header, top-right ×, and action buttons. Longer forms scroll while their actions remain visible. Dialogs have no separate Cancel or Close button, except Preferences' explicit Cancel action: the top-right × or Escape dismisses them without saving. Destructive confirmations use a red action. Pending saves must finish before their dialog can close.

Ctrl+click toggles downloads; Shift+click selects a range. With the list focused, use ↑/↓, Home/End, Shift to extend a range, and Ctrl+A to select the visible downloads. Space pauses or resumes the selection; Delete removes it. Toolbar Resume and Pause affect eligible selected files. Remove and Restart ask before discarding saved partial progress; clearing completed history keeps the finished files. For completed downloads, File actions → Delete file… (or Delete files… for a selection) shows the saved paths and asks before permanently deleting the files and their history entries. Files do not go to the Recycle Bin. Missing or locked files keep their history entries and report an error; use Clear from history to forget a file that was already moved or deleted. Search and other text fields keep their normal editing shortcuts. Successful saves use green feedback; errors remain red.

The sidebar filters the selected queue by status and file type, and switches queues. Manage queues shows a card per queue with file counts, progress, Start/Pause, Schedule and View downloads. The ⋯ → Download at once submenu sets the queue's concurrency from 1 to 8; the menu also contains completion actions, rename, clear-finished and empty-queue deletion controls. Tools → Browser setup and Preferences → Browser & AI → Browser integration open the same browser connection dialog. Tools also contains video preview, website grabber, and folder settings. Help → About shows a rounded information card with a purple download badge, the app name, a prominent blue version heading, Created by Xetro, and GitHub (coming soon). It supports both themes and closes with × or Escape. The version follows the full major.minor.patch build metadata. Set GITHUB_URL in src/desktop/about.rs when the repository URL is ready; until then, the placeholder does not open a browser. Browser connection status is absent from the sidebar. Sidebar labels can be switched off in Preferences → Appearance.

The default table shows file name, size, status with progress, speed, time left when space permits, and file actions. Existing saved column choices and custom layouts are retained. View → Table columns → Reset applies the new arrangement after Save. Separate progress, row number, and added-date columns remain available.

Drag a table heading left or right to reorder columns. Drag its right-hand divider to resize; double-click a divider to restore that column's automatic width. Release saves the layout; Escape cancels an in-progress drag. Widths and order survive restarts and are shared across queues. Oversized layouts scroll horizontally while header and rows stay aligned.

Preferences → Table columns also offers visibility, order arrows, numeric widths, Auto, and Reset. Save applies the draft; × leaves the saved layout alone. File name and Actions cannot be hidden. Narrow-window hiding of row number, Added, and Time left remains responsive and does not erase their saved positions or widths.

Click a filename or row to select it and open its details. Download rows have no checkboxes. The details panel shows the source, saved location, resume support, and errors. Close it with × or uncheck View → Show details to make room for the table. Clicking another filename reopens it. File actions, the row menu, and right-clicking a row provide authentication, link recovery, checksums, and other file operations. Move to includes queue destinations and Move up / Move down.

Double-click a download, or choose Download status in its right-click or three-dot menu, to follow it in a separate compact window. It shows percentage, downloaded size, speed, estimated time remaining and the worker's current activity. Its tabs are Status, Speed limit and When done. Expand Show details for resume support, source, destination, and per-connection progress or saved media segments. Unknown totals and unavailable estimates remain blank or marked unknown; media progress counts completed segments.

Manual single downloads (including browser handoffs and reviewed video downloads) open their status automatically once they start. Preferences → Downloads → Download status window turns automatic opening off. Batch, queue, CLI and automation starts stay in the background. Pause keeps the window open and saved progress; close the window afterwards to stop following it. Hide, Escape and the close button only hide the window and downloads continue. Resume / Start immediately starts this file without starting other queued files. A completed file offers Open file (using the existing antivirus review when enabled) and Open folder. Switching queues, filters or pages in the main window does not change an open status window.

Filter the selected queue with All downloads, Downloading, Unfinished, and Completed. Unfinished includes downloading, paused, queued, and failed entries. View → Status filter also offers In progress, Queued, and Failed. Filename search ignores letter case and combines with status and file-type filters. Click an active file type again, or its × chip above the table, to clear that type filter. Clear beside the search box clears only the search. Queue actions apply to the entire selected queue, including files hidden by filters. Clear finished removes completed entries from that queue's history and keeps their saved files. Changing a queue or filter returns the list to the top; progress updates keep your scroll position. An empty search result offers Clear filters to restore all files in that queue. An empty queue offers Add download.

Open file explicitly asks Windows to open a completed local file using its normal association and security prompts. Duplicate and Redownload create new entries in the selected queue, keeping the original file and history. They start from zero with a fresh filename collision check; credentials and partial bytes are never copied. A duplicate of an authenticated item stays paused while you supply its own credentials.

Download Engine defaults to the charcoal Dark theme with mint-green accents, blue selections, and amber paused progress. Existing saved appearance choices are preserved. Preferences → Appearance contains System/Light/Dark theme, compact rows, sidebar labels, details, and table columns. Theme and appearance choices apply after Save; choose System to follow Windows. Short windows use compact rows automatically. Preferences → Startup & tray opens window-close and notification options. The top-right × discards unsaved changes; use Save to apply them. These settings persist across restarts. Browser integration and About also close with ×. Escape continues to dismiss these dialogs. Time left appears only with a known remaining size and measured transfer rate.

The portable package keeps its queue and partial files in data and finished files in downloads, beside the executable. Keep both executables and portable.json together. Without portable.json, development builds use the installed profile folders above. DOWNLOAD_ENGINE_HOME (or IDM_EX_HOME from before 2.0) can override these locations for isolated testing. The queue keeps its SQLite WAL connection alive while the app is open, avoiding repeated checkpoints during progress polling. Saves retain full synchronization; the final store handle releases the database at shutdown. For a manual backup, exit the app, let browser handoffs finish, and copy the whole data folder. Do not copy only queue.sqlite3 while the app is running.

The Main queue starts automatically with a limit of three simultaneous files. Each queue can run 1–8 files, with a combined maximum of eight across all queues, and four connections per large file by default, configurable up to 32. Queue history survives restarts. Large ranged downloads now adapt as connections finish: a free connection helps with the largest unfinished range. Downloads begin with four ranges at 8 MiB; smaller files use one. Rebalancing requires at least 4 MiB remaining in a range, keeps each new piece at least 2 MiB, and never exceeds the configured active connections or 256 saved pieces per file. Global and per-file speed limits still apply. Servers that cap total throughput may see little benefit. Actions on a just-paused or completed file wait for its previous worker to finish releasing files before they run. This prevents an immediate Resume, Restart, or Remove from being rejected during the final cleanup step. Exiting the app pauses active downloads; use Resume when ready. A desktop started by the extension runs independently of the browser. Native launches inherit no browser handles, so closing a browser does not leave its pipes held open by the app. An already-running copy owns the queue; launching that copy again restores its window.

Resume requires byte-range support and a strong ETag from the server. Otherwise, the Start again action downloads the entire file. If the remote file changes, use Restart to discard the partial copy and fetch the new version. Saved parts keep their physical identity when split and are assembled in byte order. Pause/resume and replacement-link checks use the same saved layout. Existing queues and fixed-range partial downloads remain readable. Once a file has an adaptive layout, its queue requires this build or newer; older builds reject that layout. Update the desktop and native host together.

Existing destination files are preserved; repeated names receive a numeric suffix. Use a local NTFS folder: safe completion currently requires hard-link support. Allow free disk space for both partial data and the assembled file.

Optional antivirus scanning

Open Preferences → Antivirus → Antivirus scan, choose a scanner, and enable scanning. It is off by default. When enabled, newly completed downloads are sent to the selected scanner. With Defender, ClamAV, and compatible command-line scanners, Open file runs a fresh scan, displays the result, and requires a second explicit Open click after No threats found. Errors, detections, incomplete results, cancellation, and timeouts block that action. Malwarebytes, Kaspersky and Bitdefender use manual result review, as described below. Opening a folder remains available.

The file menu's Antivirus scan / result shows the last report and offers Scan again, including when automatic scanning is off. Completed-file details show the last scan status. Reports include scanner, time, and a SHA-256 of the scanned file; they survive restarts, but historical reports never authorize an Open action. Removing a history entry also removes its scan report.

Available scanners:

  • Microsoft Defender is located automatically in Windows' installed platform directory. It uses a custom file scan with remediation disabled; it does not change Windows antivirus settings. Defender must be available and active. Its exit code 2 is ambiguous (detection, required action, or error), so Download Engine reports Scan needs attention and blocks opening.
  • ClamAV uses your installed clamscan.exe. Update signatures with FreshClam first. Download Engine requires a confirmed single-file scan and enables alerts for encrypted files and exceeded scan limits. It does not remove or quarantine files. Such alerts block opening just like detections.
  • Other antivirus (command line) requires an absolute local scanner .exe, one argument per line with {file} exactly once, and distinct no-threat and detection exit codes from the vendor. Do not put shell quotes around the placeholder. You must verify the scan-only command waits for completion and reports skipped/incomplete scans as errors. The configuration executes the selected program with your user permissions; configure only a trusted AV. Shell interpreters, scripts, and known Malwarebytes GUI/Toolset launchers are rejected. Other products need their own supported command-line interface.
  • Malwarebytes, Kaspersky and Bitdefender (review in their apps) request scans through the installed consumer product's Windows Explorer action: Scan with Malwarebytes, Scan for viruses (Kaspersky), or Scan with Bitdefender. Select the corresponding provider and optionally enable Automatically request antivirus scans. Defender activation is unnecessary. These integrations require an English individual Explorer scan command, a standard Program Files or Program Files (x86) vendor installation, and a running Explorer desktop session. Submenu-only or translated commands that Explorer's automation API does not expose are unsupported and fail closed. The helper verifies the selected vendor's registered local DLL, matches one exact action, and never invokes a default Open action or guesses a shell command. Kaspersky Endpoint policy controls, Bitdefender BEST CLI, and Malwarebytes Breach Remediation/Toolset are separate products and are not automated here. Check setup checks configuration and readable installed files only; it does not run a scan or prove that the engine/signatures/license are ready. Scan requested — review in antivirus acknowledges dispatch, not completion, acceptance by a busy scanner, or a clean verdict. The antivirus controls its normal protection and quarantine behavior. Review the actual report there. Open file requests a fresh scan, requires I reviewed this file's result in the antivirus app, then Open after review. Results are not imported. Historical reports never authorize opening by themselves.

The timeout is configurable from 10 to 3600 seconds (default 300) for the scanner process. For external consumer integrations it limits only the handoff. Requests are sent one at a time; the antivirus manages the scans and may require attention when busy. Cancelling, changing settings, or exiting stops owned scanner/helper work; an already requested external scan continues in its own app and must be cancelled there. A bounded helper asks the running Windows Explorer to invoke the exact selected antivirus action, so the antivirus handoff can outlive the helper. It never falls back to the downloaded file's default Open action. Queue auto-exit is cancelled while owned scan work is outstanding. A verified Open review retains read-only file and parent-folder handles until confirmation, to prevent Windows file or directory replacement between scanning and opening. Network paths and reparse-point folders require a regular local copy for this scan integration. File hashing is cancellable and runs outside the UI thread. External handoffs release their initial handles after dispatch so the antivirus can manage detections. After manual confirmation, Download Engine reacquires protected file and parent handles, rehashes the current bytes against the request, rechecks app state, and retains the handles through Open. That verification is cancellable and runs on the scanner worker rather than the UI thread.

Download Engine does not upload files to an analysis service, change antivirus exclusions, request elevation, or remove Windows' Internet-origin marker. The chosen AV's own settings and cloud behavior still apply. No threats found is that scanner's result, not a guarantee that a file is safe. Disabling the option restores ordinary Windows opening and its security prompts; this setting only controls opening through Download Engine.

Vendor references: Microsoft Defender CLI, ClamAV scanning, ClamAV flags and exit codes, Malwarebytes file scans, Kaspersky context-menu scans, Bitdefender consumer guide, Bitdefender consumer/enterprise CLI distinction.

Categories and automatic save folders

New downloads are sorted into Documents, Programs, Compressed, Music, Video, or Other folders under the normal download folder. The same rules apply to individual URLs, batch URLs, browser link actions, and accepted automatic browser handoffs. Categories use the final extension of the sanitized filename, ignoring letter case: .PDF is Documents, .tar.gz is Compressed, and .pdf.exe is Programs. Images, unrecognized extensions, and files without an extension go to Other. Categories do not identify a file's actual contents or establish that it is safe.

Open Preferences → Downloads → Download folders (also available in the Tools menu) to choose an absolute local folder for each category. A blank field uses the named subfolder under the default download folder. Hover a folder for its full default path, or a category for example file types. Turn off Automatically sort new downloads by category to save new files directly in the normal download folder. Save applies all folder settings together; × discards the draft. These settings persist across app and native-host restarts. Installed defaults are resolved again from Windows on each launch; restart the app after moving the Windows Downloads folder. Portable defaults follow the app's current location. Custom absolute folders stay where you configured them and may need changing on another PC.

In Add download or Add batch, leave Save to blank to use these rules. Entering an absolute local folder overrides automatic sorting for that download or the entire batch. The Add dialog previews the category and destination using the chosen filename. Folders are created when a download starts, and folder errors appear on that download; no fallback silently saves it elsewhere.

Existing queued, paused, failed, and completed downloads keep their stored folders. Changing rules does not move files, rewrite history, or redirect a resume, retry, restart, or already-prepared browser handoff. Old history can be filtered by category immediately. Status and category filters combine within the selected queue; queue controls still apply to the whole queue. Adding a file in the app or reviewing a notification clears the filters so that file is visible.

Tray and notifications

The Windows notification-area icon has Open Download Engine, Pause all downloads, Preferences, and Exit. It may be inside Windows' hidden-icons menu. The tray's Preferences command opens the same popup over the current page. Choose Tasks → Hide to tray to keep downloads running with the window hidden. In Preferences → Startup & tray, enable Close window to the tray to make the window's close button behave the same way. This setting is off by default. If the tray becomes unavailable, the app restores its window and closing exits.

Exit always quits, including while hidden: active transfers settle to Paused, partial files are retained, and pending retry deadlines keep their existing shutdown behavior. Tray Pause all downloads pauses current active, queued, and retrying files across all queues. It does not disable queues or their schedules; newly added files can still start. Use Pause all to stop queue dispatch. The tray command also keeps files paused when a queue pause is still settling; starting that queue or reopening the app cannot undo the individual pause.

Completion and needs-attention notifications are enabled by default, with separate switches in Preferences → Startup & tray. They cover all queues, even while hidden. Temporary errors that will retry stay quiet, and old history does not notify again on startup. Several completions are grouped behind the current notification. Clicking reviews that download (the latest one for a group) in the app; it never executes a file. If a dialog has unsaved input, selection waits until the dialog closes.

Notifications contain filenames, not source URLs or server error details. Old banners are retired before the next batch so stale clicks cannot select another file. Windows notification settings and quiet modes control delivery; accepting a notification does not prove it appeared. Closing or restarting the app clears its transient notifications. No taskbar preference or notification policy is changed. The native integration follows Windows notification-area APIs.

Speed limits

Choose Speed limits at the bottom of the window. Set a total limit for all queues, and optionally a separate limit for the selected unfinished file. Enter KiB/s (1,024 KiB/s = 1 MiB/s); uncheck a limit for unlimited speed. Supported limits are 1 KiB/s to 1 GiB/s. Choose Apply limits to save and apply changes immediately, including to active transfers.

All files and range connections share the total budget. A file's connections share its own budget, and both limits apply together. Replacement-link byte checks also use these budgets. Settings survive pause, retry, restart, and app relaunch; removing a file removes its individual limit. The table shows a file's cap beneath its measured speed, and the footer shows the enabled total cap.

Limits pace downloaded file data with a small burst allowance (up to 100 ms of credit, capped at 1 MiB). HTTP/TLS headers, metadata probes, and operating-system or HTTP buffering can briefly exceed the displayed payload rate; this is not system-wide network shaping. Unlimited remains the default.

Hourly and daily data quotas

Open Preferences → Downloads → Data quotas to set either or both allowances. They are off by default. Enter MiB (1,024 MiB = 1 GiB), then choose Apply quotas. Limits apply across every queue, file, range connection, and replacement-link verification. They can be changed while downloading; changing or disabling a limit preserves the usage already recorded in its current window.

The hourly window repeats every hour and the daily window every 24 hours from the first time quotas are enabled, rather than at local midnight. Settings show the next local reset time and current usage. Both counters track received file data while either limit is enabled; no new usage is counted with both disabled. Counters and window boundaries survive app restarts, and setting the clock back does not grant another allowance. When both limits are reached, downloads wait until both permit more data.

At the limit, resumable files retain their partial data and wait in the queue without consuming a retry attempt. They continue automatically after reset, subject to the queue's running state, schedule, and concurrency settings. Manually paused downloads stay paused. A server without safe resume information requires manual action: raise or disable the quota, then choose Start again. This avoids repeatedly downloading the same prefix each window. A link check interrupted by a quota keeps the original URL and saved bytes; rerun it later.

Usage counts received file payload, including retransferred bytes after retries and replacement-link verification. Deleting a download does not refund usage. Chunks are accounted when their database write is processed. An in-flight chunk may cross a reset or exceed the allowance; this is not an exact ISP meter. Browser traffic, website-grabber HTML scans, metadata probes, HTTP/TLS overhead, and data buffered by the OS or HTTP stack but not delivered to the engine are excluded. Existing speed limits still apply independently.

Recover an expired download link

Select a paused or failed file, choose Update download link, and paste a fresh HTTP/HTTPS file URL. Check link verifies availability and compares every saved byte with the replacement. This uses bandwidth for the saved portion and can take as long as downloading that portion again. You can cancel the check.

If the saved data matches and the server supports verified byte ranges, choose Update and resume. Otherwise, the dialog explains why a restart is needed; Update and restart explicitly discards the partial data. No saved bytes means Update and start. The file keeps its name, folder, download ID, queue, and position. Confirming wakes an idle, unscheduled Main queue. Other paused or full queues keep it waiting; choose Start immediately from the file's menu to run it ahead of that queue. Existing completed files are preserved and filename collisions still receive a suffix.

Until you confirm, checks, failed checks, and cancellation keep the original URL and saved bytes. A check holds this download inactive until you confirm or cancel; other downloads can continue. The app rejects login pages, bad ranges, changing validators, and unsupported links. A file with pending publication must use Resume to recover completion before its link can change. Saved credentials are reused only for a replacement on the same origin (scheme, hostname, and port). A different server is checked anonymously; accepting its link removes the old saved credentials. Browser-login snapshots are reused only on their original server; moving to another origin removes the snapshot.

Matching size and ETag alone cannot identify a file at a different URL (HTTP validator scope). After comparing the saved bytes, Download Engine pins the replacement's strong ETag, size, and final URL for subsequent ranges. This relies on the server honoring its validator and does not establish file authenticity or malware safety.

Browser recovery and connection settings

Browser-assisted recovery is also available in Update download link → Find a fresh link in the browser. Open the original webpage and download the same file with the updated Chrome, Edge, Brave or Firefox extension. It matches the new filename and known size, pauses the browser copy, and asks the desktop to check saved bytes. Only confirming the desktop result updates the existing job and permits cancellation of the browser copy. Cancellation, lost app access or expiry releases it back to the browser. Recovery does not import new cookies, and needs the optional downloads permission; automatic capture can remain off. If matching cannot identify the file, paste its fresh URL manually. See Help → Guide → Downloads & recovery for the full walkthrough.

Preferences → Connection → Download connections saves a default and up to 100 hostname rules. Each file can use 1–32 HTTP range workers; the UI offers 1, 2, 4, 8, 16 and 32. Exact/more-specific rules override broader subdomain rules. For redirects, the lower matching source/final-host limit wins; the default applies when neither host matches. Limits are per file, not aggregate per server. Active files adopt changes after pause/resume, including existing multi-part plans. Small and non-range files use one connection. FTP, HLS and DASH retain their existing transfer behavior.

Downloads that require authentication

Download Engine supports HTTP Basic username/password credentials and bearer tokens for direct file URLs. In Add download → Advanced, check Use login and enter the username and password in the fields that appear. HTTP/HTTPS also offers a Bearer token option. Choose Start download, or Add to queue for a paused queue. The file is created paused while Windows saves the login, then queued only after that succeeds. A failed save stays in the form; Retry login reuses the same paused entry. Cancelling before submission creates no download; cancelling after a failed save leaves that entry paused. Cancelling, changing login method, unchecking Use login, or editing the URL clears the entered credentials.

For an existing file, pause it or select a failed download, then choose Authentication in its details. Select Username / password, Bearer token, NTLM, or Negotiate (Kerberos / NTLM), enter the values (without the Bearer prefix for tokens), and choose Save and continue. Clear Continue the download after saving to leave it paused. Continuation still respects the queue's schedule and concurrency limit. HTTP 401 and 403 errors include guidance for correcting credentials or an expired link.

Changing or removing credentials when partial data exists requires explicit consent to discard that data and restart. This prevents mixing downloads from different accounts. The filename, folder, download ID, and queue position stay unchanged. The None method removes saved credentials. Credentials cannot be changed during a transfer, link verification, or pending file publication.

Credentials are stored per download in Windows Credential Manager, under opaque IDM EX/Downloads/v1/ entries. The queue records only the entry ID, method, and server origin. Credentials survive pauses and restarts on the same Windows account and computer; they are not part of a portable ZIP or queue backup. If you move the queue to another account or PC, enter credentials again. A different download on the same server never inherits them.

Remote authenticated requests require HTTPS with normal certificate validation. Plain HTTP is accepted only for literal loopback IP addresses. Credentials are attached only to the configured origin; redirects to any other origin, including a different port, are blocked. Use a direct URL for the intended server.

Completion, removal from the queue, switching to None, or accepting a link on a different origin removes the obsolete credential entry. An opaque-ID cleanup journal retries interrupted deletion on the next app start. Uninstall preserves the queue and its credentials for later reinstallation: remove the downloads or choose None first if you want their credentials removed. Older builds do not understand credential bindings; remove authentication before downgrading.

For company servers, NTLM and Negotiate accept DOMAIN\username or user@domain and a password. Negotiate uses Kerberos when available and may fall back to NTLM. Only the explicit account is used; automatic Windows sign-in is disabled. These methods require Windows and Direct connection or a custom HTTP proxy. A custom HTTPS/SOCKS proxy cannot be combined with enterprise server authentication. Kerberos domain availability and service names must be validated in the organization's environment. Local fixtures do not prove domain sign-in.

This does not implement website sign-in forms, interactive OAuth login, or Digest. Proxy authentication is configured separately in Preferences. Automatic browser capture remains anonymous and falls back to the browser for protected files; manual desktop authentication does not grant the extension access to passwords or tokens.

Queues and batches

Choose New queue, enter a name, then select it in the queue picker. New queues wait for Start or a schedule. Single-link and automatic browser handoffs go to Main queue; page-link batches use the queue chosen in their picker. An idle, unscheduled Main queue starts when a new file is added or explicitly resumed. Completed history and individually paused files do not block that automatic start; custom queues, schedules and queues with existing waiting work retain their policy. The tabs and table show the selected queue; completed files stay in its history.

Use a download's right-click or three-dot menu → Start immediately to run that existing file ahead of its queue without starting other waiting files. Saved bytes, filename, destination and queue position are preserved. The override bypasses its queue's pause, schedule and usual file limit, while retaining the global eight-file cap and data quotas. Pause, Pause all, queue pause and app restart cancel it.

Use Add batch to paste one HTTP/HTTPS/FTP/FTPS URL per line (up to 1,000), select the destination folder, and add the entire batch. Blank lines and duplicate links within that batch are skipped. An invalid line prevents the whole batch from being added and leaves the draft available to correct.

In Manage queues, Download at once controls simultaneous files in the selected queue. Reducing it pauses excess transfers and returns them to the waiting list. Pause stops new work and pauses its active files; Start continues them in order. Files paused individually stay paused until you choose Resume or Start again. Pause all clears one-time starts and suspends recurring schedules.

Choose a file, then Move to to reorder it or move it to another queue. Pause an active file first and let cancellation finish. Order and membership are saved separately from transfer progress. More lets you rename a queue or delete an empty queue; the Main queue is kept.

Schedule offers One time, Daily, or Selected days:

  • One time sets a future local start date/time, pausing the queue until then. If the app is closed at that time, the queue starts when reopened. Ambiguous or nonexistent daylight-saving times are rejected. Manual Start/Pause clears it.
  • Daily / Selected days sets a start and pause time in the computer's local time zone. The days refer to the start: Mon 23:00–07:00 runs until Tuesday morning. Equal times are rejected. Saving/enabling inside a window starts it immediately; outside a window the queue waits. Missed windows are skipped.
  • At a window's end, active downloads pause safely and keep their partial files. They return to the waiting list for the next window. Individually paused files stay paused. Recurring transfers also return to the waiting list after an app exit or interruption; they run again only during an enabled window.
  • Manual Start/Pause suspends recurrence and retains the saved settings. Enable schedule (Scheduler or the queue's More menu) restores it. Schedule opens the saved settings for editing. Clear schedule removes either schedule type and keeps the current running/paused state.
  • Schedules survive restarts. Keep Download Engine open (the tray is fine) and the PC awake for on-time transitions. On reopening or waking, it checks the current window instead of replaying missed starts. A recurring window with a nonexistent daylight-saving boundary is skipped; repeated times use the first occurrence for both start and pause. Schedules follow changes to the computer's time zone.

Waking the computer is not implemented. Optional shutdown after queue completion is available through the queue menu. In ordinary queues, recurring windows control when unfinished transfers may run. Servers must support safe resume to reuse partial bytes; non-resumable transfers restart when their queue next runs.

Existing histories migrate into Main queue in their original oldest-first order. The updated database should be used with this version; older executables do not understand all queue/schedule rules and could start files outside their window.

Synchronization queues

Choose New queue → Synchronization queue, then add URLs or move existing completed downloads into it. Check now downloads new entries and checks saved files once. Schedule uses the same one-time, daily, and selected-day windows; each recurring window starts one check, including after an app restart in that window. The app must be open and the PC awake. Manual Check now / Pause suspends the schedule; Enable schedule restores it.

  • Up to date means a strong ETag and the saved SHA-256 still match, or a fresh download has the same bytes as the local copy. Weak ETags and modification dates alone are insufficient; those servers require downloading the file to compare it. Missing or edited local copies are downloaded again safely.
  • Updated means the new file finished downloading and passed its integrity checks. It is published as a separate file, such as file (1).zip; previous copies are never overwritten or automatically removed. Plan for disk space for both versions. The row points to the newest copy. Its ⋮ menu shows the last check and the previous copy's folder.
  • Check failed keeps the previous copy. Open the row for the error; existing retry, pause/resume, authentication, and link-recovery controls apply. A server changing again during a partial replacement can require Restart. Files paused individually remain paused until resumed individually.
  • Browser size/ETag metadata may change for a new version, but its file-type guard is kept. User-entered expected SHA-256 checksums remain enforced. Saved login credentials still expire/are retired on completion; synchronization does not retain or silently renew browser sessions. Protected or expiring links may need fresh authentication or a replacement link.
  • Updated copies use the normal optional antivirus scan and Open-file review. Unchanged checks do not repeat completion notifications or an existing scan. Malwarebytes results still require review in Malwarebytes. Clearing a row removes it from future checks and keeps its saved copies.

Connect a browser

The extensions add Download with Download Engine to the right-click menu for file links, and provide a popup for pasting URLs. The popup also has an Automatic capture switch, on by default after installation. If the desktop app is not connected, file downloads stay in the browser. Reload an already-loaded extension after updating this package.

Capture rules, under the switch, choose which downloads are captured: file types (extensions separated by spaces; * matches any characters, so r0* covers split archives and * alone captures every type), an optional minimum size, and sites never captured from. A plain site name such as example.com also covers its subdomains; *.example.com covers only subdomains. Sites are compared with the download address, its final address after redirects and the page that started it. The type comes from the browser's chosen filename, or from the address until the browser reports one. Downloads outside the rules, or with an unknown size, stay in the browser untouched. New installs start with common archives, installers, disk images, audio, video and PDF; extensions that had capture on before 1.15.0 keep capturing every type. Download with Download Engine and Add a file URL ignore these rules.

When a new direct file download matches the rules, the extension briefly pauses it. Download Engine checks that it can fetch the same file without browser cookies, starts the desktop app, and saves the job in Main queue before the extension cancels the browser copy. Failed checks return the download to the browser. Turning the switch off stops new captures; files already queued in Download Engine continue there. Capture setting changes are applied in order. Revoking download access during a desktop connection check leaves capture off, even if access is granted again.

A recovery journal and a one-minute browser alarm resolve interrupted handoffs. The popup's Recover handoffs button retries immediately. If the browser refuses a resume request, the popup reports it; use Resume in the browser Downloads list. The extension never deletes a downloaded file or clears browser history. If all native replies are lost, it keeps the browser copy and warns that a second copy may have reached Download Engine.

Capture is conservative: it skips private downloads, browser-flagged files, downloads started by other extensions, HTML pages, generated blob/data files, and files whose size or type remains unknown. When a browser has not yet supplied that metadata, the extension rechecks for up to ten seconds while the browser continues downloading. Firefox's paused/interrupted state is handled explicitly for capture, fallback, and recovery. A file that finishes before takeover stays in the browser. Login-only and POST downloads are not supported: a separate unauthenticated GET must match the browser's size and type. This metadata check cannot prove byte-for-byte identity; when available, a strong ETag is also checked when Download Engine starts the transfer.

Download links from the current page

In Chrome, Edge, or Firefox, open a normal HTTP/HTTPS page, click the Download Engine extension, and choose Choose files from this page. Download links from this page is also available when right-clicking the page background. A separate tab lets you:

  1. Search filenames, link text, or URLs; filter by type, source, or reported size; and sort by name, size, type, or page order. Unknown sizes are shown as — and excluded by a minimum-size filter. The picker does not fetch files to measure them.
  2. Select individual links or Select matches; nothing is preselected. Selections remain selected across pages and when hidden by filters, and the count says so. Select matches selects every match across the 100-row pages; Shift+click selects a range in the current filtered order.
  3. Choose a saved Queue and Save to destination, then Add selected files. Automatic folders uses the app's category rules; Default downloads uses its root folder. Named destinations use the folders configured in Preferences. Use Reload destinations after changing queues or folders in the app.

The scan runs only when requested. It reads top-page links and assets, including images/srcset, common lazy-image attributes, audio/video sources, posters, background images on the first 500 elements, and open shadow roots. It resolves relative URLs, removes fragments and duplicates, and rejects embedded credentials and non-HTTP(S) links. Traversal and candidate counts are bounded at 10,000 each; up to 1,000 unique links fit within a byte limit. A notice appears when truncated. It does not crawl pages, inspect frames or closed shadow roots, fetch previews, or transfer cookies. Private windows and browser-protected pages are unsupported. Types use filenames and available media MIME metadata, not downloaded contents; ordinary page links can also appear. HLS/DASH entries open a desktop preview. A direct-file selection is an anonymous GET, so login-only and POST links can fail.

The native host validates and saves the whole batch atomically. A lost reply keeps its original request ID: Retry same batch checks that receipt without adding the files twice, even if their history was removed. A closed pending picker can be reopened with the popup's Recover handoffs button. Once saved, the selected queue controls when its files run; the extension cannot auto-open or execute them.

Unsent scans and pending selections are held in memory for this browser session, including across popup closure or a background-worker restart. Closing an unsent or completed picker clears its scan. Restarting the browser or reloading/disabling the extension clears session data, so resolve uncertain batches first, or check the desktop queue before collecting them again. Saved desktop jobs persist. Browser batch receipts keep IDs and a digest of the request, not another copy of its URLs or filenames.

Chrome, Edge and Brave

Store installation (recommended): install Download Engine Integration from the Chrome Web Store. Edge may ask you to allow extensions from other stores. In Download Engine's Browser setup, choose your browser, paste dedbmjfdofigmfigagjnjileiepfhnio into the extension ID field, then choose Connect browser. Choose Check connection in the extension popup. The store ID differs from the app's suggested unpacked-folder ID; use the store ID for this install. The browser updates store-installed extensions automatically. Store releases can lag behind the desktop release; newer features may require the matching manual package below.

Manual installation:

  1. In Download Engine, open Browser setup and choose Chrome, Edge or Brave.
  2. Choose Copy extensions page, paste it into that browser's address bar, and enable Developer mode.
  3. Choose Open extension folder in Download Engine. On the browser page, use Load unpacked and select that folder.
  4. Choose Connect browser. Download Engine fills in the 32-letter ID the browser gives that folder: Chromium derives an unpacked extension's ID from its folder path (SHA-256 of the path, first 16 bytes as the letters a–p), so the app computes it for its own extensions folders. If you loaded the extension from another folder, paste the ID shown on the browser's extensions page instead. A warning appears when the browser is connected to a different ID than this copy's folder.
  5. In the extension popup, choose Check connection, then send a file link.

These unpacked extensions can only be used in Developer mode. Chrome, Edge and Brave each have their own extension folder, so their extension IDs differ. Registration is scoped to the exact ID you enter, under the current user's registry, with no admin access. The panel shows Registered for this copy after saving the host mapping; only the extension's Check connection verifies a live browser connection. Choose Disconnect to remove this copy's mapping. A mapping owned by another Download Engine copy is preserved: disconnect it in that copy before connecting a new one. Moving a portable package requires disconnecting the old location first.

On Windows, Brave has no native-host registry key of its own. Like other Chromium-based browsers without Google branding, it checks HKCU\Software\Chromium\NativeMessagingHosts before Chrome's key, so Download Engine registers Brave there with its own native-brave.json. Brave's connection is therefore separate from Chrome's. Other browsers that read the Chromium key, such as Chromium itself, find this registration first.

Firefox

  1. Install Download Engine Extension from Firefox Add-ons. It stays installed after a restart, and Firefox keeps it updated.
  2. In Download Engine, open Browser setup, choose Firefox, then choose Connect browser. You can skip the page's first step.
  3. Choose Check connection in the popup.

The Firefox ID is already included; there is no ID to paste.

To test a development build instead, copy the page shown in Browser setup into Firefox's address bar, choose Load Temporary Add-on and select manifest.json from the folder available through Open extension folder. Firefox removes this unsigned development add-on when it restarts. The Chrome/Edge development folders are not store installations either. There is no browser-policy modification or signing bypass in the installer.

If automatic capture switches back off with No such native application com.xetro.downloadengine, the extension is loaded but the desktop connector is not registered. In the desktop app, choose Tools → Browser setup → Firefox → Connect browser, then reopen the extension and try Browser options → Check connection. Enable automatic capture again after it connects. Installing the add-on alone does not register the desktop connector.

When a checkbox asks for optional permission, accepting completes that setting in the background even if Firefox closes the extension popup. Download Engine then returns to the popup in the same focused window and active tab. Video-button setup returns to the expanded Browser options section. Denying permission leaves the feature off. Granted permissions are reused on later visits. Automatic return requires Firefox 149 or newer (or Chrome/Edge 128 or newer); older Firefox versions still save the setting but may require reopening the popup.

Turning either feature off preserves its browser permission. Turning it back on reuses that access and checks it again in the background, without asking for it again. If access was revoked in browser settings, enable the feature again to request it. Connection errors and rejected changes keep the checkbox off.

For troubleshooting, open Diagnostic log directly on the extension popup, below Browser options. You do not need to expand Browser options to find it. It keeps the last 30 setting, connection-check and video-review results plus the latest popup error locally, across popup closings. Select log, then Ctrl+C, copies the selected text; Clear log removes these records. Action names and bounded error messages are logged; web addresses and Windows paths in those messages are redacted. Nothing is sent automatically.

Check connection displays its progress and result directly below the button. Video-button failures appear in the popup status and diagnostic log, so the error remains available after closing the page or popup.

Optional scripted registration

The packaged Register-Browsers.ps1 remains available for automation. It calls the same validated Rust registration code used by the desktop. Examples:

./Register-Browsers.ps1 -ChromeId YOUR_32_LETTER_EXTENSION_ID
./Register-Browsers.ps1 -EdgeId YOUR_32_LETTER_EXTENSION_ID
./Register-Browsers.ps1 -BraveId YOUR_32_LETTER_EXTENSION_ID
./Register-Browsers.ps1 -Firefox

Add -WhatIf to preview or -Unregister to remove mappings owned by this copy. Run the packaged script beside the executables. The script is optional.

Checksums, queue completion and desktop controls

Add download accepts an optional expected SHA-256 (64 hexadecimal characters). Download Engine calculates SHA-256 for every new download during file assembly. If an expected value does not match, the download needs attention and the assembled file is not published. Restart and replacement-link recovery retain the expected checksum; duplicating a download copies the expectation, never the old result. Completed files offer row menu → SHA-256 checksum to calculate or recheck a saved file, enter a publisher checksum, and Copy SHA-256. This also works for files downloaded by older versions. Manual mismatches preserve the saved file and history, but the app's Open file action is blocked until a successful recheck or an explicit calculation without an expected value. Results describe the bytes when checked; integrity verification is not malware detection.

Use queue More → When queue finishes → Notify / Exit / Shut down computer. Actions start Off, are one-shot, and apply only to this app session. They wait for every remaining file to complete successfully; paused, failed, retrying or mismatched files prevent completion. Removing or moving a watched file cancels the action. Windows notifications follow the system's notification settings. Exit shows a 15-second countdown with Cancel exit, and cancels if any other download is unfinished or a dialog is open. It exits only Download Engine, never the PC.

Shutdown uses a 60-second Cancel shutdown countdown inside the app. It cancels if another download, antivirus scan, script, control command or scheduled website scan is unfinished, or a dialog is open. No OS power event is scheduled during the countdown. At its end Windows is asked to shut down normally, with no forced application termination; unsaved work can block shutdown. This release's shutdown action was reviewed and compiled but deliberately not executed, at the user's request.

Drag an HTTP/HTTPS/FTP/FTPS link (or lines of URL text) into the Windows app to open Add batch for review. Nothing downloads until submitted. Drops received during another dialog wait for it to close; the current draft is kept. The native OLE handler accepts Unicode URL text, rejects embedded credentials and non-web schemes, and does not open or execute dropped files. Its bounded queue rejects additional drops while full. Existing queued work resumes normally after an app restart; completion actions must be enabled again.

The optional floating drop box (View → Floating drop box or the tray menu; off by default) is a 64 px always-on-top, no-activate tool window with its own OLE registration on a dedicated STA thread. It accepts the same drops while the main window is minimized or in the tray: one link opens Add download filled in, several open Add batch, and the main window comes forward. Its on/off state and position (physical screen pixels) are saved in the interface preferences; the position is kept fully on a monitor's work area and follows per-monitor DPI.

Clipboard links (Preferences → Integration & security, on by default) offers HTTP/HTTPS file links you copy in other apps. The copied text must be exactly one link whose last path part ends in an archive, installer, video, audio or document extension (for example ZIP, ISO, EXE, MSI, MP4, MP3, PDF or DOCX). Web pages, plain text, scripts, images, playlists, links with a query-only file name, links with credentials and paragraphs containing a link are ignored. A Windows notification from the tray icon and a card above the status bar show the newest link; Download… or the notification opens Add download with the link filled in, and nothing starts until you add it. While another dialog is open, the card waits for it to close.

The app listens through a Windows clipboard-format listener (no polling) only while the setting is on, and removes it when you turn the setting off or exit. It reads Unicode text up to 8,192 characters, never logs or saves it, and skips copies that password managers mark as private. Copies made while Download Engine is the active app (Copy link, Copy SHA-256, text copied from its fields) and a copy identical to the previous clipboard text are ignored. Like download notifications, the notification's file name and host stay in the Windows notification center until cleared.

Use Preferences → Table columns to show or hide row number, status, progress, size, speed, time left and added date. Save applies changes; × preserves the previous choices, and Reset restores visibility, order and widths. File name and Actions remain visible. Narrow windows still hide extra columns to fit.

Video links in the browser

Click Find videos on this page in the extension, or use its page/video context-menu action. The scan collects recognizable direct video URLs (such as MP4/WebM) from top-frame video elements, their source elements, and video links. Known video MIME types allow sources with opaque URLs. Download Engine opens or comes to the front with a native Video downloads dialog; no browser picker tab opens. Nothing is preselected. Choose direct files, a queue and a saved destination, then click Add selected files. HLS / DASH entries have Choose quality instead of a checkbox. This opens the desktop quality picker and reads the available qualities; downloading still requires Add video to queue. Close either dialog with its top-right × to cancel. An empty scan explains why nothing can be added. The ordinary Choose files from this page action keeps its browser file picker.

This is an explicit, local page scan with temporary activeTab access. It recognizes .m3u8 / .mpd links and HLS / DASH MIME types on video sources, plus manifest URLs in the document's resource-timing entries (useful for some blob/MSE players). Blob URLs themselves and players with attached DRM keys are excluded. It does not inspect iframes, transfer cookies or bypass access controls. The desktop handles video previews and segment downloads after you review them. Open the actual video page and scan again after a source appears. Login-only sources can fail because the desktop performs its own GET request.

For media requested during playback, use Detect media while browsing in the popup. It is on by default once the extension's webRequest and HTTP/HTTPS website access are granted; the extension requests both at installation. Start playback or reload the page after enabling it. The popup shows the active tab's detected audio, video, HLS and DASH entries; its toolbar badge shows their count. Click an entry to review it in Download Engine before any download. Find videos also includes these detected items in its review.

Detection observes successful GET response metadata (URL, MIME type and reported size), including opaque media URLs. It does not read request headers, cookies or response bodies, make preview requests, or queue downloads. Obvious stream segments are excluded. URLs may still contain signed access tokens. Lists stay in browser session storage, capped at 40 items per tab and 16 tabs; entries older than 30 minutes are pruned when the detector next runs. Navigation, tab closure, disabling detection and permission revocation clear the relevant records. Clear empties the current list; later playback requests can add fresh entries. Turning off detection preserves granted access shared by other features; revoke access in browser extension settings if desired. Private windows are excluded.

Detection does not make DRM, live/encrypted streams, POST requests or login-only media supported. The desktop's existing preview checks still apply. The separate browser-login review remains the only path for explicitly approved cookie transfer. YouTube, Bilibili (including Bangumi episodes), TikTok and Twitch video page links use the desktop's built-in readers, even if browser media detection finds zero files. Download video or Find videos on this page opens quality selection in Download Engine. If a built-in reader cannot read the page (for example a region-locked or members-only video), the review shows what the extension found on that page instead, with the reader's error at the top; turning on Detect media while browsing and starting the video first gives it the most to show. Pages covered by one of your own site readers (a Lua script with the Site reader trigger, see the automation reference) open with that reader the same way. Other sites still need exposed direct files or supported playlists. Automatic capture handles browser file downloads; it does not convert a streaming player into a downloadable file. Video buttons are on by default wherever website access has already been granted, including access granted for Detect media while browsing. On YouTube, the button appears only beside the main player on video pages (watch, Shorts, embeds and live), never on channel banners or hover previews; on Bilibili, only beside the main player of a /video/ or /bangumi/play/ page, never on card previews; on Twitch, only beside the player of a /videos/… page, never on live channels or previews; on TikTok, only beside the video a /@user/video/… page is about, not the next video it preloads. Dailymotion's embedded player also gets a button beside its frame. Clicking Download video or Find videos on this page first requests public Dailymotion metadata, without browser cookies, and checks the returned playlist. A readable automatic HLS playlist becomes one video entry; use Choose quality in the desktop app to see its available qualities. The lookup is bounded to eight seconds and never downloads video segments. If metadata is missing, blocked, malformed or returns an unreadable playlist, the review falls back to links and streams captured by the browser. Enable Detect media while browsing, reload the page and start playback for this fallback. The button does not inspect the cross-origin frame's contents. Dailymotion setup fragments and short player previews are excluded. Known playlist URLs for the same video are grouped into one review entry. Up to four browser-captured alternatives are retained even when the metadata playlist is readable in the browser: the desktop may receive a different response. If the desktop receives HTTP 401/403 or a successful response containing an error page or no playlist, it tries those alternatives, preferring a captured CDN rendition. Invalid or unsupported content inside an actual playlist, cancellation and network security failures stop the lookup. A rendition fallback offers the quality the player actually loaded; change playback quality and scan again to capture another quality. Older desktops receive the retained alternatives as separate choices instead of the new grouped message. Update the desktop to 3.19.1 or newer for automatic fallback. Other video IDs, unknown URL layouts and explicitly listed metadata qualities are not silently substituted: they remain separate choices, as do captures beyond the four-alternative limit. Metadata success alone never hides an unmatched captured playlist. If an ad is playing, wait for the main video to start and scan again; the player may not have requested the main video's playlist yet. Refreshed URLs for the same CDN asset and rendition replace the old entry. Logs records each desktop lookup attempt, its result and HTTP status; quality-preview failures are recorded separately. These diagnostics contain no signed URLs, headers or response bodies. A rejection does not establish that login is required or the URL expired. These playlists describe video segments; an unknown playlist size does not mean the video is empty. If a signed link returns an error page, reload the video page and start playback to capture a fresh link. Change playback quality in the player to discover another rendition. A signed Firefox Add-ons installation uses the same detection code as the developer extension; local source changes reach store users only after a new signed update.

For detection problems, open Diagnostic log at the bottom of the extension popup, then Refresh log. The trace shows the metadata request, playlist validation, HTTP status or failure reason, browser fallback counts and desktop acknowledgement. A successful desktop acknowledgement only means the review was received; it does not mean a video was downloaded. Select log, then Ctrl+C, copies the report. Discovery traces keep the last 120 events for the current browser session and exclude URLs, page titles, cookies and tokens. Clear log also clears these traces. Private windows do not perform metadata lookups. Manual URL entry and browser connection/recovery controls are in the popup's collapsible Add a file URL and Browser options sections.

Browser options → Video button on this site is always available on regular web pages in Firefox and Chromium. It starts checked and remembers an explicit off choice for that site's HTTP or HTTPS host, covering its ports. If access is missing, Allow this site requests that optional permission after a click; opening the popup never requests access. Existing pages and future visits with access show buttons automatically unless disabled for that site. Buttons appear just above visible top-frame video players or recognized Dailymotion embeds and open the same desktop review; clicking a button never queues a file. At the viewport edge they move below or beside the player. If no outside space is available (including a full-screen player), the button hides; the extension's Find videos on this page remains available. Buttons avoid covering neighboring players. The setting persists across reloads and restarts. Video buttons and media detection are independent: toggling detection does not change a saved video-button choice. Explicit off choices survive permission revocation and re-granting, as well as reloads and extension restarts. Disabling does not revoke a site permission shared with browser-login support; site permissions can also be removed in the browser's extension settings. No cookie permission is requested for buttons. They are unavailable in private windows, browser-internal pages and embedded frames. The button reads player layout and reacts to DOM changes locally; URLs are collected only on a click. Navigation within the same document (including YouTube watch-page changes) keeps working: where supported, the browser's document identity binds the collector result to the clicked page. A replacement document is rejected before handoff. Closed shadow UI, trusted click checks, sender/frame checks and an extension-controlled handoff keep page scripts from sending direct native download commands.

Files and streams added from a browser page, and downloads the extension hands over, send that page's origin (for example https://example.com/) as the Referer, as browsers do by default. The full page address, cookies and other browser data are never sent, and an HTTPS page sends nothing to a plain-HTTP server. Some video servers refuse requests without it.

Video handoffs save an expiring review, not a download. If the desktop reply is lost, use Browser options → Recover handoffs to retry the same review ID. Pending reviews survive an app restart and expire after 30 minutes. A dismissed or handled review is not reopened by a retry; scan again to start another review. Update both the desktop app and extension when using this handoff.

YouTube video downloads

Use the extension's Download video button or Find videos on this page, or paste a YouTube video link into Add URL / Tools → Preview video. Choose a quality and audio track, preview the selection, then Add video to queue. Loading qualities does not start a download. The filename comes from the video title.

This native Rust implementation ports yt-dlp's anonymous VISIONOS/HLS flow; it does not run Python or a yt-dlp executable. Public, finished videos with a supported HLS response are supported. Availability and qualities depend on YouTube's response; this is not full yt-dlp feature parity. JavaScript challenges, proof-of-origin tokens, sign-in/private/age-restricted videos, DRM, live recording, playlists, subtitles, and direct adaptive-format extraction are not implemented. Unsupported responses show an error instead of downloading the watch page or silently changing quality.

The download saves the video ID and selected video/audio format identifiers. Starting or resuming refreshes expiring links, retaining partial segments only when encoding versions and the segment layout still match. Packed AAC audio is validated, its transport timestamps checked, and merged with the selected video into MP4 using the existing bundled merger. Existing HLS/DASH jobs remain compatible; new YouTube jobs require this updated application. Diagnostics record lookup success/failure and timing without storing titles, URLs, visitor data or signed links.

See the implementation and source map for upstream references, limits, and repeatable verification commands.

Bilibili video downloads

Use the extension's Download video button on a Bilibili video or Bangumi episode page, or paste a link such as https://www.bilibili.com/video/BV… (also av… numbers and ?p= for one part of a multi-part video) or https://www.bilibili.com/bangumi/play/ep… into Add URL / Tools → Preview video. Choose a quality and audio track, preview, then Add video to queue. The filename comes from the video title, with the part number and name for multi-part videos, or from the series title and the episode's name.

The app reads Bilibili's public web API the way its own player does: anonymous visitor IDs from /x/frontend/finger/spi, then WBI-signed /x/web-interface/wbi/view and /x/player/wbi/playurl requests (an MD5 signature over the sorted query with keys Bilibili publishes at /x/web-interface/nav). The reply lists separate MP4 video and audio streams with their initialization and SIDX index ranges; they become an indexed DASH presentation and use the existing segment downloader and bundled merger. Streams from peer-assisted mirrors (PCDN) are avoided when Bilibili offers its own CDN. Requests to Bilibili's video servers carry https://www.bilibili.com/ as the Referer, as its player sends; some mirrors refuse requests without it. No account cookies are sent, so Bilibili offers up to 480p; sign-in qualities, paid, region-locked and DRM videos, live pages and b23.tv short links are not supported.

Bangumi (anime, series and documentary) episodes use the PGC endpoints /pgc/view/web/season and /pgc/player/web/playurl, which list the same DASH streams. A season link (/bangumi/play/ss…) opens its first main episode; trailers and extras listed in a season's sections work by their own ep… links. Episodes that Bilibili shows anonymous visitors only as a short preview (members only), and DRM or region-locked episodes, are refused with a message instead of saving a preview. The request flow follows bilibili-downloader-gui (MIT); no code was copied.

The download saves the video ID and part (or the Bangumi episode ID) and the chosen stream keys, never a signed link. Bilibili's links expire after about two hours, so starting or resuming reads fresh links first and keeps partial segments only when the length, byte ranges and SIDX index are unchanged. Live checks: cargo test --lib bilibili::tests::live -- --ignored.

TikTok video downloads

Use the extension's Download video button on a TikTok video page, or paste a link such as https://www.tiktok.com/@user/video/… into Add URL / Tools → Preview video. Choose a quality, then Add video to queue. The filename comes from the video's caption.

The app loads the video page the way a browser's first visit does and reads the video data embedded in it (__UNIVERSAL_DATA_FOR_REHYDRATION__) without running any script. Each quality is one ordinary MP4 file with sound; the app downloads it in resumable 16 MiB byte ranges and saves it unchanged, without FFmpeg. TikTok's video servers only answer with the page's anonymous visitor cookies and a https://www.tiktok.com/ Referer, so the app sends both, and only to tiktok.com servers over HTTPS. The cookies are never written to disk: the page is read again before every start or resume, because links and cookies expire, and completed parts are kept only while the file size and hash are unchanged.

TikTok sometimes answers automated page loads with a JavaScript browser check. The app does not try to pass it; it reports it, and opening TikTok in a browser and trying again a few minutes later usually helps. Private, sign-in-only and photo posts, vm.tiktok.com short links and live videos are not supported. Live checks: cargo test --lib tiktok::tests::live -- --ignored.

Twitch video downloads

Use the extension's Download video button on a Twitch video page, or paste a link such as https://www.twitch.tv/videos/… into Add URL / Tools → Preview video. Past broadcasts, highlights and uploads are supported. Choose a quality, then Add video to queue. The filename is the channel name and the video title.

The app asks Twitch's public GraphQL endpoint (gql.twitch.tv, with the public client ID of Twitch's own web player) for the video's title and an anonymous playback token, as the web player does, and opens the video's HLS master playlist on usher.ttvnw.net with it. The qualities' own playlists and segments need no token, so the download is an ordinary finished HLS download: newer broadcasts are fragmented MP4 and save as .mp4, older uploads are MPEG-TS and save as .ts. The saved download keeps the video page address, never the token, and pausing or resuming does not contact Twitch again. Subscriber-only videos, live channels, clips and sign-in are not supported. Live checks: cargo test --lib twitch::tests::live -- --ignored.

HLS video downloads

Choose Tools → Preview video, paste an HTTP/HTTPS playlist URL, and click Load video qualities. The first supported quality previews automatically, with the site's default audio selected initially. Choose a different quality or Audio track to update the preview immediately. The controls remain visible during previewing; changing either cancels the previous request and clears its preview. A direct media playlist opens its original-quality preview. Review the duration, segment count, output format, estimated size (when the playlist supplies bitrate), filename, queue, and folder, then click Add video to queue. Your filename, queue, and folder choices are kept when changing quality or audio; the filename extension follows the output format. Retry preview retries a failed or cancelled request; Change URL clears the preview and returns to URL entry. Adding is disabled until the current selection has a ready preview. Previewing fetches playlists only, with no queue entries or media downloads. Playback inside the preview is not included.

Chrome, Edge, and Firefox use the same desktop preview. Right-click an exposed .m3u8 link and choose Download with Download Engine, paste it into the extension's URL field, or use Find videos on this page → Choose quality. HLS links pasted in the desktop Add download dialog also open the preview. Opaque playlist URLs can be pasted directly into Preview video. Page scans read exposed URLs; optional media detection can also find supported GET media responses during playback. Blob/MSE-only players may not expose a usable link. Reload development extensions after updating the app and extension files.

Ordinary file downloads reject HLS/DASH manifest responses at the initial probe, fallback GET and range requests, before saving those response bodies. Use Preview video for these links. HTTP authentication and retry errors retain their normal handling even when an error response has a manifest content type.

  • Supports finished, unencrypted HLS playlists (EXT-X-ENDLIST), muxed MPEG-TS segments saved as .ts, and fragmented MP4 with a single initialization segment saved as .mp4. Explicit and implicit byte ranges are supported. Separate EXT-X-MEDIA:TYPE=AUDIO tracks in either of these containers are downloaded with the chosen video quality and merged into one .mp4, without re-encoding. The preview shows the audio name/language and output format. A track labeled "included in video" retains the original media's audio. Codecs must be compatible with MP4; a missing track or merge error fails the download rather than publishing a silent video. Source timestamps determine synchronization, independently of segment counts and media sequence numbers. The bundled FFmpeg helper needs no system installation.
  • Live recording, encryption (including AES-128 and SAMPLE-AES), DRM, alternate video rendition groups, subtitles, packed audio, changing initialization maps, discontinuities/gaps, nested master playlists, and variable substitution are not supported. Unsupported containers returned by a server fail during transfer instead of publishing a file.
  • Downloads use the normal queues, retries, pause/resume, speed limits, payload quotas, SHA-256 verification, antivirus actions and safe file publication. Resume reuses whole completed segments after checking their saved hashes; an interrupted segment starts again. This covers both video and selected audio. Both playlists are rechecked before any media is fetched. The table shows Merging audio and video once all segments are downloaded. Pause cancels the owned merge process; Resume retries merging from verified complete segments. Merging needs temporary disk space for the assembled tracks as well as the output; temporary track copies are cleaned up when the merge stops. Only the successful, hashed output is published and passed to antivirus actions. A changed/expired playlist requires a fresh preview. Restart discards segments for the same reviewed playlist; Duplicate and Update download link open a new preview. Synchronization queues skip completed HLS videos rather than treating manifests as ordinary files.
  • Requests do not inherit browser cookies or file-server credentials. App proxy settings apply. Each destination is DNS-checked and direct routes are pinned; a selected proxy owns final hostname resolution. Custom Lua reader results are an exception: their previews and saved downloads require Direct connection and public addresses at every request, including the root and redirects. CDN requests cannot cross into a different local/special-use origin. An explicitly entered local origin is allowed for local media. Redirects and playlist references cannot downgrade HTTPS to HTTP. No encryption key is fetched.
  • The helper receives only locally assembled track files, forced TS/MP4 input formats, and fixed arguments. Network protocols and external MP4 references are disabled. It writes standard MP4 through an already-open, guarded output handle. The app verifies the pinned helper checksum, protects its file/folders while it runs, and bounds memory (1 GiB), individual allocations (256 MiB), runtime (30 minutes), and output size (1.5 times input bytes plus 8 MiB). Failure leaves complete segments available for retry; no incomplete output is published.
  • Limits: 1 MiB per playlist, 64 qualities/servers, 32 audio choices per group, 10,000 combined video/audio parts, seven days of duration, 256 MiB per segment, five redirects, and bounded network waits. Pending browser previews do not interrupt an open desktop dialog, are limited to ten, and expire after 24 hours. Playlist/resource URLs (including signed query strings) and progress are stored locally with queued jobs; preview-only requests clear their URL when opened. Use the current app for HLS job data.

Playlist handling follows the supported subset of RFC 8216.

Development builds need the pinned helper under tools/ffmpeg. Run ./scripts/Prepare-MediaTools.ps1 to verify its binary, notices and source bundle. Rebuild it using tools/ffmpeg/source/README.md when needed. Packaging and installer builds verify and include the helper, corresponding source and license notices. Release binaries load only their adjacent tools/ffmpeg/ffmpeg.exe; there is no PATH or user-configurable command fallback. See FFmpeg provenance for the version, checksums, source archives, build recipe and license. Keep those materials with any redistributed binary package.

DASH video downloads

Use Tools → Preview video, a .mpd link in the extension, or Choose quality on a DASH entry in the desktop's Video downloads dialog. Select the video resolution/bitrate and audio language/bitrate, then review the duration, output, filename and queue before adding. Loading qualities reads only the MPD. SegmentBase preview additionally reads the chosen tracks' index byte ranges; initialization and audio/video segments are downloaded only after queueing. Preview creates no download. The same flow is used by the on-page video button.

  • Supports static, single-period, unencrypted MP4 video/audio representations. SegmentTemplate supports inherited attributes, $RepresentationID$, $Bandwidth$, $Number$ (including zero padding), $Time$, and bounded SegmentTimeline repeats. SegmentList supports explicit URLs and byte ranges.
  • SegmentBase supports inherited indexRange and initialization ranges in the same MP4 file, flat SIDX versions 0/1, 32/64-bit box sizes and offsets, and presentation offsets. Each index is at most 1 MiB and must receive an exact HTTP 206 range response. Offsets, counts, timestamps and resource bounds are checked before ranges can enter the queue. Only the chosen quality/audio indexes are fetched. Index bytes, resource sizes and supplied strong ETag / Last-Modified validators are checked again before transfer/resume; range responses must retain those validators and resource size during transfer.
  • Chosen tracks are merged into MP4 without re-encoding using the bundled helper. Per-track presentation offsets are applied before merging. A video-only MPD is explicitly labeled as having no audio. MP4-compatible codecs are required.
  • Normal queues, retry, segment-level pause/resume, quotas, speed limits, hash verification, antivirus handling and safe publication apply. The MPD is checked again before transfer/resume; a changed selection requires a new preview. Completed segments are hash-checked before reuse. Old HLS downloads keep their saved format; existing DASH downloads retain version 3. Indexed DASH uses version 4, including the reviewed index identity, so older apps reject it safely.
  • Requests are anonymous and use the same DNS-pinned destination checks as HLS. MPDs are limited to 1 MiB; XML depth, node counts, expanded URLs and segment counts are bounded. DTDs, external entities, XLink and ContentProtection are rejected. MP4 initialization metadata is checked for protected sample entries. FFmpeg receives only guarded local files and fixed arguments, never a manifest, remote URL, cookie, key or page-supplied command.
  • Unsupported: live/dynamic or encrypted DASH, multiple periods, hierarchical or external representation indexes, separate-file SegmentBase initialization, WebM representations, dependent layers, multiple BaseURL alternatives, subtitles, timeline gaps/overlaps, or segments crossing the period's start. Unsupported manifests fail with an explanation; they are not silently saved as videos. Browser login for streams and playback inside the preview remain unavailable.

Timing behavior follows the DASH-IF timing model. For local verification, run cargo test --test hls --locked -- --test-threads=1 and node --test tests/video_button.test.cjs. scripts/video_button_smoke.cjs --firefox uses installed browsers with disposable profiles and fixture-only site permissions; it does not test the browser permission prompt. scripts/grabber_schedule_smoke.cjs --dash captures native quality/preview dialogs with a screenshot-feature build.

Website grabber

The grabber has two modes: Find downloadable files retains the reviewed file discovery workflow below; Offline website saves a browsable static copy.

In Offline website, enter a starting page, destination folder, link depth, page limit and size limit. Save offline copy creates a unique folder; Open offline website opens index.html. Keep the whole folder together. HTML links, CSS imports and URLs, responsive images, icons and fonts are rewritten to local paths, with shared resources using one local file. Missing pages lead to report.html; report.json records the captured files and omissions. Stopping keeps a partial copy when the starting page is available.

This is an anonymous static UTF-8 capture, not a JavaScript/browser-session snapshot. Scripts, forms, SVG animations and embedded pages are disabled. Pages stay on the starting origin; optional CDN assets must resolve to public addresses. Resources use the configured proxy without direct fallback. Limits include depth 5, 200 pages, 2,000 assets, 500 MB total (100 MB by default), 20 MB of HTML/CSS, 2 MB per text resource, 16 MB per asset, 10,000 inspected URLs and ten minutes. The operation runs separately from queues and their bandwidth/quota controls; it does not persist or resume a capture after restart. See Help → Guide for details.

Choose Tools → Website grabber in the desktop app:

  1. Enter the starting HTTP/HTTPS page and comma-separated file extensions, such as pdf, zip, jpg or tar.gz. Presets cover documents, archives, images, audio, and direct video files. Filtering is case-insensitive and recognizes URL filenames and HTML download filename hints; it does not verify contents.
  2. Set Scan depth and Page limit. Depth 0 reads the starting page and its same-origin include-fragment HTML resources, including GitHub release asset lists. Fragments share the page/byte limits and never execute scripts. 1 also reads pages linked directly from it. Depth is limited to 5, with up to 200 pages including fragments (50 by default). GitHub /blob/ file-view links become /raw/ download links; duplicate raw links appear once.
  3. Choose Scan website. Stop scan keeps partial results for review; closing the dialog cancels the scan and discards its preview.
  4. Search the preview and select individual files or Select visible. Nothing is selected automatically. Filtering preserves hidden selections; Clear selection clears all of them. Choose a queue and a saved destination, then Add selected to queue. A running queue can start the selected files immediately, and an idle, unscheduled Main queue wakes automatically. Other paused queues wait. Folder rules, retries, antivirus settings, and the normal file-opening protections apply to these downloads.

Saved grabber projects

In Tools → Website grabber, expand Saved projects, enter a Project name, and choose Save project. A project remembers the starting URL, file types, scan depth, page limit, queue, and destination folder choice. You can choose the queue and folder before scanning, or save them from the preview.

  • Pick a project from the saved-project list to restore its settings, then choose Scan website. Scan again reruns the current settings from a preview. Every run fetches fresh results with nothing selected; you must review and choose Add selected to queue each time.
  • Use Save changes to update settings or rename the current project. Enter a different name and use Save as new to keep a separate copy. New scan in the list starts an unsaved scan with default settings.
  • Delete project asks for confirmation and removes the saved project, its schedule, pending discoveries, and discovery history. Existing downloads are preserved. The currently displayed manual preview can still be used.
  • Opening a project or choosing New scan replaces the current unsaved settings and preview. Project controls are disabled while a scan is running.
  • If the saved queue was removed, choose another queue before saving or adding files. Folder choices use the current destination rules from Settings rather than a frozen folder path.

Projects persist locally in the app's database across restarts. Up to 100 projects can be saved with unique names of 1–80 characters. The starting URL, including its query string, is saved; browser credentials and preview selections are not. Manual scan results stay in memory. Invalid URLs or scan limits are rejected. Changes to a newer saved version cannot be overwritten by a stale editor. Saving or opening a project does not scan, create downloads, or start a queue.

Scheduled grabber projects

Save a project, then choose Schedule scans, or open Scheduler → Website scans → Manage scan schedule. Choose One time or Repeating, a first scan date and local time, and, for repeating scans, an interval of 1–720 hours (24 hours is the default). Save any edited project settings before choosing Save scan schedule. Clear scan schedule cancels an active scheduled scan and prevents future runs while keeping pending files.

  • Download Engine must be open (including in the tray) and the PC awake. After a missed time, sleep, or restart, one scan runs when the app can resume. Missed repeats are coalesced; subsequent scans keep the original fixed hourly interval. This is an elapsed-time interval, so its local clock time can shift with daylight saving or a time-zone change. Ambiguous/nonexistent first local times are rejected.
  • Scheduled scans run one at a time and use the saved filters and bounds. They collect links only: they never enqueue files or start download queues. The existing anonymous, same-origin crawl restrictions also apply here.
  • The first scheduled scan treats all matching URLs as new. Later scans show URLs not previously found by this project's scheduled scans, ignoring URL fragments but preserving query strings. This does not compare file contents or deduplicate against unrelated/manual downloads. Manual reruns still show all matches and do not change scheduled discovery history.
  • A banner shows new files awaiting review. Open Scheduler → Website scans → Review new files, or Review new files inside the saved project. Nothing is selected automatically. Select files and choose Add selected to queue; Ignore selected dismisses those URLs without downloading them. Both actions keep the URLs remembered. Unselected files and later discoveries stay pending.
  • Closing a scheduled preview keeps its files across restarts; selections are intentionally cleared. Refresh preview loads discoveries that arrived while the dialog was open. Adding selected files and removing those pending entries commit atomically, so a failed add preserves the preview and a stale submission cannot enqueue duplicates. Choose another queue if the saved one was deleted. Normal folder rules and antivirus/open-file protections apply.
  • Partial results and scan failures are shown with the last scan status. Failed scans keep older discoveries; repeating projects try again at the next interval. A cancelled/interrupted scan does not establish a discovery baseline. Project edits, rescheduling and deletion invalidate in-flight results from older settings.

Each project retains up to 1,000 pending files / 700,000 bytes of preview data and 10,000 URL fingerprints. Full previews wait for review or report that additional links were left for another scan; those links are not marked seen. Reaching the history limit stops the schedule. Reset discovery history → Forget reviewed links allows previously reviewed URLs to appear again, preserves pending files, and does not automatically enable a stopped schedule. A project copy starts with no schedule, pending files, or history.

Pending URLs, suggested filenames, and their source pages are saved in the local app database until added, ignored, or the project is deleted. They can include query strings such as expiring links; review those links before downloading. Discovery history retains SHA-256 URL fingerprints rather than a second copy of the URLs. Browser cookies and download credentials are never used by these scans.

Discovery reads static UTF-8 HTML with a separate anonymous HTTP client. The starting hostname is resolved once and pinned for the scan; mixed public/local answers and unusable destinations are rejected. App proxy settings apply; a selected proxy is trusted to resolve the final destination. Direct IP or localhost scans remain supported, but one scan cannot cross between public and local networks. It follows page links only on the starting origin (same scheme, host, and port), including redirects. If a site redirects to another origin, enter that final page URL and scan again. Matching file links on other origins, such as a CDN, can appear in the preview but are not followed during discovery. File URLs are deduplicated without fragments; query strings are preserved. Relative URLs, HTML entities, base URLs, and direct image/audio/video sources are recognized. Page links with no filename extension or a common HTML/server-page extension can be followed. Matching HTML pages are both previewed and followed within the depth limit; unknown binary responses are skipped without reading their bodies. The scan does not execute scripts or programs, submit forms, enter frames, or send browser cookies or saved download credentials.

Scanning is sequential and paced, with a 15-second timeout per page, a 3-minute scan limit, 2 MiB per HTML page, 20 MiB of HTML per scan, and bounded preview storage. Each page contributes at most 10,000 inspected links; the preview holds up to 1,000 files. Limits, failed pages, and partial results are reported. Queue insertion reuses the atomic batch receipt so repeating a submission cannot duplicate that batch. Manual previews stay in memory; scheduled discoveries are persisted for review. Neither creates jobs until you confirm your selection.

File discovery does not render JavaScript, scan login-only pages, discover opaque file endpoints without a matching filename hint, or download HLS/DASH. The separate Offline website mode saves bounded static copies as described above. Help → Guide includes a GitHub example for each mode.

Proxy settings

Open Preferences → Connection → Proxy and choose System proxy (the existing default), Direct connection, or Custom proxy. Custom addresses support http://, https://, socks5:// (local DNS), and socks5h:// (proxy DNS). One custom endpoint handles all HTTP and HTTPS file destinations, including browser automatic-capture checks, download retries/resume, synchronization, replacement-link checks, ZIP previews, HLS/DASH previews and segments, YouTube lookups/refreshes, and manual/scheduled grabber scans. System/environment bypass lists do not override a custom route, and failures never fall back to direct access. Pause and resume an active download to apply changed settings; a running transfer retains its route snapshot. FTP/FTPS continue to connect directly. PAC editing, custom bypass lists and separate proxies per protocol are not implemented.

For a company proxy, choose NTLM or Negotiate (Kerberos / NTLM) and supply an explicit account. Enterprise proxy authentication requires Windows and an http://host:port endpoint; HTTPS destinations use CONNECT tunnels with normal certificate validation. Basic proxy authentication retains the HTTP, HTTPS and SOCKS options. Enterprise requests require Direct or Custom, because automatic System proxy discovery cannot guarantee an explicit proxy account. When no file-server login is supplied, use a destination hostname rather than localhost or an IP address through an enterprise proxy. Likewise, an unauthenticated proxy for an enterprise file login must use a hostname. These restrictions prevent WinHTTP's IP-address exception to its automatic-login policy from using the current Windows account. Supplied accounts are set before the first handshake. Standard requests retain System/environment routing. Each enterprise request owns a bounded WinHTTP session and authentication exchange. Cookies and automatic Windows credentials are disabled; redirects are validated by the caller. Proxy credentials never become destination headers. See Microsoft's authentication contract and handle concurrency contract.

Choose Enter login to save a proxy username/password in Windows Credential Manager. SQLite contains only the proxy endpoint, mode and an opaque credential identifier. Proxy credentials are separate from download/server credentials. Existing passwords are never loaded into the dialog; use Keep saved login, enter a replacement, or No login. Changing the endpoint requires a new login or removal. Saving System/Direct mode removes the custom endpoint and saved login. Replacements/removals are journaled so interrupted cleanup retries at app startup. Missing or mismatched credentials stop the connection rather than silently downloading anonymously or directly.

HTTPS proxies validate their certificate and hostname. Basic HTTP and SOCKS5 proxy logins are unencrypted even when the destination file uses HTTPS. NTLM/Negotiate uses a challenge response, but does not encrypt the proxy connection itself. Remote HTTP/SOCKS proxy logins require explicitly allowing that trusted connection. Prefer HTTPS proxies or a trusted local proxy. Literal loopback proxies are allowed for local VPN/proxy applications. An HTTPS destination still has its own certificate validation, and proxy credentials are sent to the proxy rather than inside its connection to the destination. OS/HTTP libraries may retain temporary copies of credentials in memory; Windows vault protection is scoped to your user. See the HTTP client's proxy contract.

ZIP contents preview

In Add download, paste an HTTP/HTTPS link and choose Preview ZIP contents. You can also use Preview ZIP contents in a download's row menu. The preview shows searchable file/folder names, expanded and packed sizes, password-protected entry markers and unusual-path indicators. Back to download preserves the URL, filename, folder and checksum draft; nothing is added to the queue until you use the normal download action. Existing protected downloads reuse that download's saved authentication. For a new protected link, save Authentication with Continue the download after saving off, then preview from its row menu.

The preview reads at most a 65,633-byte tail, a 56-byte ZIP64 record if needed, and 4 MiB of central-directory metadata. Small ZIP files may fit entirely in the tail read. The server must honor HTTP ranges and return consistent ranges, size, location and validators. A directory outside the tail requires a strong ETag or valid Last-Modified value; changed/missing validators stop the preview. Servers that ignore ranges do not trigger a whole-file download fallback. Closing or cancelling stops the request, and the complete operation has a 60-second deadline.

Standard single-volume ZIP and ZIP64 version-1 directory records are supported, with UTF-8, CP437 and CRC-checked Unicode path names. Limits are 10,000 entries, 4 MiB of directory data and 4,096 bytes per filename. Split archives, encrypted or compressed directories, ZIP64 record extensions and unsupported layouts are rejected. Encrypted entries can be listed when their directory is readable. Nothing is decompressed, extracted, opened or scanned. Reported sizes and names are archive metadata, not a safety verdict; antivirus checks still apply to the normal downloaded file. This does not download individual archive members. Format reference: PKWARE ZIP specification.

Verification uses loopback HTTP/HTTPS/SOCKS fixtures, a ZIP written by .NET, malformed/range-change fixtures and native dialog interaction tests. Run cargo test --lib proxy:: --offline --locked -- --test-threads=1 and cargo test --test zip_preview --offline --locked -- --test-threads=1. No external proxy credentials, production servers or OS certificate changes are used by these tests.

First-version boundaries

  • HTTP GET downloads, including the bounded HLS / DASH VOD workflows above. Explicit browser-login snapshots apply to direct files; POST downloads, encrypted/live streams and DRM players remain unsupported. Direct video files exposed in the current page can be collected on request.
  • The extension has contextMenus, nativeMessaging, storage, alarms, activeTab, and scripting permissions. Temporary page access is used only for an explicit page-link or video scan. On-page buttons optionally register a persistent content script on individually enabled sites, or on HTTP/HTTPS pages while media detection and its granted access are active; there is no default broad host access. The optional downloads permission is requested only for automatic capture. Optional cookies are requested only by the explicit browser-login review. Browser login and standalone per-site video buttons request individual site access; optional media detection requests webRequest and broad HTTP/HTTPS host access to recognize playback served from other hosts such as CDNs.
  • Temporary failures retry automatically; permanent failures and exhausted retries stay visible with an error. Resume starts a fresh retry budget and uses saved bytes only when the server supports safe resume. See automatic retries below.
  • New unnamed downloads accept sanitized server-provided filenames when available; explicit names, browser-supplied names, and existing history retain their names. Folder selection stays fixed at add time.
  • Signed download URLs are retained locally in the queue database, so treat the data folder as private and clear queue entries you no longer need.
  • Portable and per-user release builds support Authenticode signing and timestamping, or explicitly unsigned GitHub releases. Browser-store publication is separate.

Automatic retries

Temporary connection failures, timeouts, interrupted bodies, and HTTP 408, 429, 500, 502, 503, or 504 receive up to three automatic retries, after 5, 10, and 20 seconds. Progress does not reset the budget. A server's longer Retry-After delay is respected, including HTTP dates. Requests to wait more than 24 hours need manual attention; invalid headers use the normal backoff. All range workers share one attempt budget, and the longest server delay is retained.

The compact table shows Retry in ... and the details show the attempt count and last error. Waiting retries release their download slot so other queued files can continue. Queue limits, ordering, and scheduled starts still apply. Pause cancels a file's pending retry; Pause queue holds it until the queue starts again. Remove after pausing cancels it permanently. Waiting times and attempt counts survive app restarts; interrupted active transfers remain paused. Resume, Restart, changing authentication, or accepting a replacement link starts a fresh budget. An older executable does not understand retry timing; avoid downgrading while retries are pending.

Safe range retries recheck the final URL, strong ETag, and size before appending missing bytes. Without a safe resume plan, a retry starts from byte zero. Authentication errors, expired links, certificate/redirect failures, invalid or changed ranges, disk/permission/database failures, and unknown errors do not retry automatically. Saved credentials remain bound to the same download and origin during retries and are removed on completion or deletion as before.

FTP and FTPS files

Use Add download with a file link. Plain ftp://server/path/file.zip uses anonymous login only and sends file contents without encryption. Saved usernames and passwords require ftps://; embedded URL credentials are rejected.

  • ftps://server/path/file.zip defaults to implicit TLS on port 990.
  • ftps://server:21/path/file.zip defaults to explicit AUTH TLS on port 21.
  • For another port, include it in the URL and choose explicit or implicit TLS in Add download. Under Advanced, check Use login and enter a username/password while the job is paused. Authentication is also available for paused or failed FTPS jobs. Bearer tokens and browser login do not apply.

Each saved login belongs to that download and server/port in Windows Credential Manager. Completion or removal retires it; duplication requires a new login. FTPS validates the hostname and trust chain using Windows TLS, requires TLS 1.2 or newer and encrypted data connections, and checks that control and data connections use the same certificate. There is no certificate-bypass option or fallback to plaintext. Private servers need a certificate trusted by Windows. DNS answers that mix public and local or special-use addresses are rejected before connecting. Hostnames resolving entirely to local addresses remain supported.

Pause, Resume, Restart, retries, ordinary queues/schedules, global and per-file speed limits, data quotas, SHA-256 verification, filename collision protection, Windows origin markers and antivirus processing also apply to FTP. Each FTP job uses one data connection. Interrupted/transient transfers use the existing three-retry policy; rejected logins, certificates and permanent server errors require attention. Network waits are cancellable and bounded to 30 seconds; intentional speed-limit waits are outside that deadline.

Resuming checks saved size/time metadata and compares the saved prefix against the remote bytes before appending. This re-downloads the prefix and counts against speed limits and enabled data quotas. FTP timestamps alone cannot guarantee an unchanged file. Supplying the publisher's expected SHA-256 allows REST offset resume when supported; publication still requires that exact hash. Unsupported REST falls back to prefix verification. Changed metadata or saved bytes stop the transfer and ask for Restart.

Paths use UTF-8, binary transfers, and per-directory CWD commands. A normal URL path is relative to the login directory; encode the leading root slash as %2F when needed, for example ftps://server/%2Freleases/file.zip. Encode a literal semicolon as %3B; FTP ;type= suffixes are unsupported. The implementation follows FTP URL paths, passive connections, FTP over TLS and file metadata/restart extensions.

Only passive mode is supported. The data connection uses the actual control peer address, ignoring PASV-advertised hosts; privileged data ports and the control port are rejected. Replies, paths and commands are bounded, control characters are rejected, and error messages omit server text that could echo a password. FTP connections go directly to the server; FTP proxies, SFTP, directory browsing and active mode are not implemented. Browser handoffs, video previews and website grabbing remain HTTP/HTTPS-only. Desktop batch/drop accepts FTP links; batches infer TLS mode from the port as above. Synchronization queues skip completed FTP files; use Redownload to fetch a fresh copy.

cargo test --lib ftp:: --offline --locked -- --test-threads=1 and cargo test --test ftp --offline --locked -- --test-threads=1 exercise loopback FTP and explicit/implicit FTPS, synthetic vault logins and cleanup, TLS refusal, untrusted/incorrect host certificates, control/data identity mismatch, prefix/hash verification, pause/reopen, retries, queues, limits, quotas and malformed replies. The public fixture certificates are trusted only inside test connectors; tests never change Windows certificate stores. node scripts/ftp_smoke.cjs --ui captures both themes using a screenshot-feature build. node scripts/ftp_smoke.cjs checks the unsigned portable app with an isolated database, exact FTP bytes, restart/REST recovery, Windows origin markers and untrusted FTPS rejection. These tests do not establish compatibility with every external FTP server.

Security

New downloads and recovered publications on Windows receive Mark of the Web before completion. This retains the Internet-origin information used by Windows and applications such as Office. The app fails the download if that marker cannot be written. The marker contains no source URL, referrer, or signed token. Already completed files from older builds are not modified automatically. See Microsoft's description of Mark of the Web.

Temporary files are created without truncating existing link targets. Windows partial-file handles reject reparse points and extra hard links, and app-owned cache directories reject junctions. Open transfer handles deny concurrent writes or deletion. Unicode direction-control characters are neutralized in new filenames so they cannot disguise a file extension. Opening the destination folder uses the Windows-owned Explorer executable.

HTTPS uses normal certificate validation; HTTPS-to-HTTP redirects and embedded URL credentials are rejected. Browser handoffs, the website grabber and video previews remain HTTP/HTTPS only. Desktop downloads also accept FTP/FTPS, with the protections described below. Password/token fields are masked and their undo history is cleared each frame. App-owned credential buffers are cleared on drop, and vault API calls are serialized; GUI, HTTP, and OS components may still make temporary memory copies. Windows Credential Manager does not isolate secrets from other programs running as your Windows user. Native messages have a 1 MiB limit and a fixed action schema. Registered browser hosts allow only the specified extension IDs, and extension background commands require the extension's own sender identity and URL.

The browser extension loads local scripts/styles only and blocks extension-page web requests, inline scripts, embedded objects, and form navigation through its content security policy. It cannot execute downloaded files. Reload an already-loaded extension after updating this package.

Keep the portable app and its data in a folder writable only by your account. The queue database is not encrypted and can contain signed download URLs. These controls do not sandbox other programs running as your Windows user or guarantee that a downloaded file is trustworthy. Malware scanning is optional and must be enabled separately in Scan settings with a working antivirus. Windows warnings depend on file type and system policy. Registration writes validate exact extension IDs, preserve other copies, and replace manifest files without truncating linked targets. Packaging requires either verified code signing or an explicit unsigned-release choice. Permanent browser-store distribution and independent security review remain separate from publishing a GitHub release.

Before releases, run the regression suite and cargo-audit against the current RustSec database. A clean dependency audit only establishes that no matching published advisories were found at that time.

Remaining IDM feature gaps

Compared with the official IDM feature list, the largest remaining gaps are signed installer/store-extension distribution, live/encrypted streams, broader DASH addressing/container support, and dynamic or authenticated offline websites. Unencrypted HLS / DASH VOD with quality/audio selection and separate-track merging is available. A bounded website grabber with saved projects, scheduled discovery of new links, file-type filters, scan depth, and selection previews is available. Static offline website capture with local pages and assets is also available. Adaptive segmentation is implemented with up to 32 active connections per file. We currently support HTTP/HTTPS and passive FTP/FTPS file downloads, with no SFTP/MMS support or automatic updater. Optional Defender, ClamAV, and configurable CLI antivirus scanning is available; Malwarebytes scan requests can run automatically with results reviewed in its app. Column visibility, resizing/reordering, server filenames, and URL drag-and-drop are available. HTTP/HTTPS proxy settings with saved logins and bounded ZIP contents previews are also available.

Our queues now support ordering, independent limits, one-time starts, and recurring daily/weekday start-stop windows. IDM also has synchronization queues and start/stop schedules. Synchronization queues, queue notifications, app-exit actions and optional computer shutdown are available. Computer wake, restart and sleep actions remain future work.

For the file-download-first scope, the implementation order is:

  1. Installation and browser setup: implemented locally with a per-user installer and in-app registration. Signing and store publication remain release steps; see browser release preparation.
  2. Expired-link recovery: implemented with saved-byte verification and an explicit restart when safe resume is unavailable. Browser-assisted recovery can find a fresh link for that existing job.
  3. Authentication: HTTP Basic and bearer tokens implemented with per-download Windows credential storage, plus opt-in Chrome/Edge/Firefox login snapshots protected by Windows user-scoped DPAPI.
  4. Automatic retries: implemented with persisted bounded backoff.
  5. Speed limits: persistent global and per-file controls, shared across range connections and applied during active transfers and replacement-link checks.
  6. System tray and notifications: implemented with optional close-to-tray, cross-queue completion/failure notices, and safe explicit exit.
  7. Categories and folders: implemented with filename-based filters and saved destination rules shared by app and browser downloads.
  8. Video detection and site grabbing: direct video-file discovery, unencrypted HLS / DASH VOD previews/downloads, optional on-page video buttons, and a bounded HTML website grabber are available; live/encrypted streams, broader DASH compatibility, and JavaScript-rendered crawling remain future work. Saved projects support once-only and repeating scans with persistent new-file previews, plus manual reruns with fresh results.
  9. Antivirus: optional scans, persisted results, and a fresh scan/confirmation before opening through Download Engine. Malwarebytes can receive automatic scan requests and requires manual result review in its app.
  10. Data quotas: optional hourly and daily allowances with durable shared usage, resumable quota holds, automatic reset, and manual-pause preservation.

Category tests cover legacy history preservation, rule validation and persistence, manual and batch overrides, native-message folder restrictions, prepared handoff retries, resume after rule changes, safe publication, and desktop filter/dialog interaction in both themes.

Local diagnostic logs

The desktop app and native browser connector keep separate structured logs: desktop.log and native-host.log in the active data folder's logs directory. An installed copy normally uses %LOCALAPPDATA%\Download Engine\logs (%LOCALAPPDATA%\IDM_EX\logs for a profile from before 2.0); a portable copy uses data\logs beside its executables. DOWNLOAD_ENGINE_HOME overrides this with <DOWNLOAD_ENGINE_HOME>\data\logs. Logs, in the top toolbar, opens the in-app viewer. Open log folder inside that page opens this copy's log files directly in File Explorer when needed.

The viewer combines desktop and browser connector events, newest first, with local timestamps, severity/source filters, and search (Ctrl+F). Events appear as plain, selectable text, colored red for errors, blue for info, and light red for warnings (with darker shades on the light theme). Drag to select, use Ctrl+A/Ctrl+C in the text area, or Copy logs to copy all displayed text. Escape clears the focused log search. It shows up to 2,000 recent events from the retained files; export includes older retained events too. Reads run on a background thread. Auto-refresh checks every two seconds while the Logs page is visible. It pauses while the text area is focused or scrolled away from the top, keeping reading position and text selection stable. Turn it off to hold a snapshot, or use Refresh when ready. Read failures retain the previous snapshot and show an error until a later refresh succeeds. Download filters and selection are preserved when switching to Logs and back.

Each component retains its current file plus three rotated backups, at up to 1 MiB each (8 MiB total, excluding tiny lock files). Logs include UTC timestamps, app version, process/session identifiers, severity, startup/shutdown and panic events, native request actions/outcomes, browser registration results, download start/completion/pause/retry/quota/failure events, and UI/scheduler errors. Job UUIDs and per-session request numbers support correlation. Available OS, SQLite, and HTTP error codes are preserved. There is no automatic upload. Writing runs on a dedicated thread with a bounded 1,024-event queue, so disk contention does not block download workers or the UI. If that buffer fills, an entries_dropped event records the count once writing recovers. Normal exit and export drain pending records with a bounded wait; forced termination can lose buffered events.

Privacy is enforced by the record schema: URLs, download names, local paths, request IDs supplied by a browser, headers, cookies, passwords, request bodies, and raw exception/panic text are never written. Unknown errors retain a category and event context rather than arbitrary text. This also means a log may not contain the full message visible in the UI. Native logs never use stdout, which remains reserved for browser message frames. Logging failures do not stop downloads or browser replies; Logs shows Logging unavailable when the logger cannot write. Errors before the data directory can be resolved still use the existing startup dialog/stderr path. Hard process termination or power loss may leave a partial last record; these are not crash dumps.

Logs → Export diagnostics… asks for a destination folder, writes a uniquely named DownloadEngine-diagnostics-*.json report on a background thread, and shows its saved location in the Logs page. Cancel leaves everything untouched. Export reads only the bounded, known log files, validates each record again, and reports skipped malformed entries. It does not include the download database, settings, browser profiles, or files you downloaded. Existing exports are never overwritten. Browser extension settings diagnostics remain separate under the extension's Diagnostic log; copy that log too for permission/popup bugs.

Help → Crash reports (also available from Logs) lists local failure reports under the active profile's logs/crashes folder. The desktop and native browser host record the first Rust panic in each process directly, independently of the diagnostic log queue. Reports contain version, OS, architecture, component, timestamps and a relative Rust source location where available. Panic payloads, stack/memory dumps, URLs, credentials, personal paths and environment variables are excluded. These are not native exception or signal dumps; they do not diagnose every possible crash.

The desktop owns a session marker while holding its profile lock. Normal exit removes it; the next launch turns a leftover marker into an unclean exit report unless that session already has a panic report. Power loss and forced termination can produce the same marker, so the UI never claims a known cause. Other app processes do not create desktop markers. Startup retains the newest 20 valid reports; malformed or redirected files are skipped, not exported. An unreviewed report prompts once when the app is focused and no other dialog is open. Closing the dialog acknowledges its reports while keeping them available for later review. Exports use exactly the previewed fields, unique filenames and explicit folder selection. No report is automatically uploaded or sent to GitHub.

Build and verify

Rust stable with the Windows MSVC target, Visual Studio C++ build tools, and the Windows SDK (including rc.exe) is required. The supplied Cargo.lock pins the resolved dependencies. build.rs embeds assets/download-engine.ico in the desktop executable; the window and tray use PNG sizes derived from the same master artwork.

cargo fmt --check
cargo test --all-targets --all-features --locked
cargo clippy --all-targets --all-features --locked -- -D warnings
node --test tests/*.test.cjs
$signing = @{
    SignTool = 'C:\absolute\path\to\signtool.exe'
    CertificateThumbprint = '<publisher certificate SHA-1 thumbprint>'
    TimestampUrl = 'https://<publisher-approved RFC3161 service>'
}
./scripts/Package.ps1 @signing
./scripts/Build-Installer.ps1 -SkipBuild @signing

Signed packaging verifies all four executables, Setup, and the generated uninstaller against the configured publisher certificate. For an unsigned public GitHub release, explicitly use -AllowUnsignedRelease instead of the signing parameters; see release instructions. For isolated local smoke artifacts only, use -AllowUnsignedDevelopment; never distribute that output. Development archives and directories include UnsignedDev in the name, and a development installer additionally requires -TestBuild so it cannot be mistaken for a release. For example:

./scripts/Build-Installer.ps1 -TestBuild -AllowUnsignedDevelopment

HLS regressions use local HTTP fixtures for quality selection, encryption rejection, playlist changes, exact TS/fMP4 assembly, byte ranges, pause/resume, retries, quotas, cache tampering, scheduler integration, and the explicit review boundary. Separate-track tests generate local H.264/AAC media with the pinned FFmpeg helper. They verify all four TS/fMP4 combinations, unchanged encoded slices/audio packets, B-frame timestamps and a shared audio/video clock, explicit audio selection, pause/reopen/resume during audio, missing-track failure, collision-safe publishing, and recovery after publication. Helper tests cover checksum rejection, cancellation, timeout, and output limits; shared process-lifetime tests also exercise antivirus child cleanup. HLS, DASH and indexed-MP4 pause tests cancel once at an exact media request, including its byte range. This keeps index/init requests intact and makes the pause point independent of helper verification time or parallel test load. Resume checks verify completed video/audio parts are reused and only the interrupted part is requested twice. With a screenshot-feature build, node scripts/grabber_schedule_smoke.cjs --hls captures the quality picker and preview in both themes. After development packaging, node scripts/hls_packaged_smoke.cjs exercises the actual executables with isolated data: native preview receipts create no job or network traffic, and a seeded reviewed fixture completes with exact bytes, SHA-256, and Windows origin marking. Pass a package directory as its first argument to check another build. For a packaged media check, build the test-only inspector using tools/ffmpeg/source/README.md, copy ffprobe.exe to artifacts/ffmpeg-tools/, and verify it with ./scripts/Prepare-MediaTools.ps1 -TestTools, then run node scripts/hls_merge_packaged_smoke.cjs. It generates TS/fMP4 tracks, exercises the actual packaged queue/merger, and checks both streams, packet counts, presentation/decode timestamps, successful decoding, SHA-256, and Windows origin marking. The inspector is not shipped. These local checks do not establish live-site or registered-browser compatibility.

Quota regressions cover persistent and concurrent accounting, independent reset windows, clock rollback, live settings acknowledgements, safe resume, manual and queue pauses, retry traffic, and link checks that preserve saved bytes. Native previews use python scripts/queue_ui_smoke.py --feature quotas. A packaged, isolated transfer check is available with python scripts/desktop_smoke.py --binary-dir dist/IDM-EX-UnsignedDev --quotas.

Website-grabber regressions use local HTTP fixtures to check depth boundaries, file-type matching, duplicate links, redirects and origin boundaries, malformed or oversized responses, cancellation, preview selection, queue atomicity, and the exact downloaded bytes. Desktop tests cover selection, retry after a removed queue, cancellation on close, stale results, and minimum-size layouts in both themes. Saved-project regressions cover database migration and reopening, invalid settings, duplicate names, stale updates/deletions, concurrent project limits, removed queues, UI save/rename/copy/delete, and reruns that fetch changed page content without reusing a prior selection. Native previews can be captured with a screenshot-feature build and python scripts/queue_ui_smoke.py --feature grabber or --feature grabber-project for the saved-project controls, and --feature grabber-schedule for the schedule form, --feature grabber-discoveries for a persistent preview, or --feature grabber-scheduler for its overview. Scheduled-grabber regressions cover actual local HTML scans, new-link tracking, restart/cancellation recovery, stale results, bounded storage, and atomic preview-to-queue transfers. node scripts/grabber_schedule_smoke.cjs captures the native schedule, preview, and overview in isolated light/dark fixtures, including reopening a pending preview. After packaging, node scripts/grabber_schedule_packaged_smoke.cjs checks the actual portable app against loopback HTML across two launches. It verifies new URL detection and persistent previews with zero automatic downloads, and stops only the fixture desktop processes it starts. It defaults to dist/IDM-EX-UnsignedDev; pass another binary directory as the first argument to verify a different build.

The installer build uses the official Inno Setup compiler. Use -Compiler C:\path\to\ISCC.exe if it is not in a default location. -TestBuild creates a separate installer identity for isolated install/upgrade/ uninstall testing. Installers contain only an explicit list of executables, extension files, the installed-mode marker, the optional registration script, and README; no runtime queue or downloaded files are embedded.

The Rust integration tests use isolated temporary directories and local HTTP fixtures, including real TCP transfers and native-host subprocesses. Range-pause fixtures hold the response at 128 KiB until cancellation, so a delayed observer cannot accidentally let the fixture finish first. The pause and link-completion waits keep their eight-second watchdogs, read SQLite off the HTTP fixture's runtime thread, and report unexpected retries or worker/scheduler failures immediately. Regressions exercise both HTTP 503 and actual SQLite write failures, plus WAL lifetime across store clones, concurrent reads, explicit checkpoints, final handle release, and persistence after reopening. Credential leak checks read the live database, WAL and other regular app files. Only queue.sqlite3-shm is excluded: it contains SQLite coordination data, and Windows enforces byte-range locks on it. A regression holds a real write transaction and verifies the scan detects a synthetic marker in the WAL, in the database after checkpointing, and in an unrelated similarly named file. They cover range downloads, cancellation/resume, changed resources, truncated responses, actual retry delays and exhaustion, persisted waits, slot release, queue/user pause, server backoff, and permanent failures, queue recovery, framing, and destination collisions. Replacement-link tests cover all saved segments, cross-URL validator collisions, explicit restart, cancellation, shutdown, stale state, login pages, and preservation of queue order. Queue tests also cover legacy migration, atomic batches, ordering, limits, independent pauses, scheduled starts, and concurrent native-host opens. Security regressions cover link-target preservation, cache junctions, Windows origin markers, and misleading filename controls. Windows authentication tests use only disposable, synthetic credentials and isolated loopback servers. They cover Basic/bearer range transfers, pause/reopen, origin and redirect restrictions, missing/incorrect credentials, cleanup after completion/removal/interruption, concurrent credential writes, guarded restarts, link recovery, and rejecting credentials from the anonymous browser protocol. UI tests check masked fields, cleared undo history, cancellation, save acknowledgement, and minimum-window layout. These tests do not exercise a real site's login or prove all TLS/proxy environments work. Browser-setup tests use disposable HKCU test keys to verify exact permissions, registration repair, missing hosts, conflicting copies, and safe manifest replacement. JavaScript tests exercise extension permissions, sender authentication, content policy, automatic handoff, interrupted recovery, and fallback contracts. These do not by themselves prove installation in each real browser.

Queue regressions hold a worker at the saved-state/cleanup boundary to verify that terminal actions wait for ownership release and still publish notifications, while controls on running downloads continue to require pausing. Link confirmation uses a held HTTP response, and retry ordering uses server-recorded request times, so these checks do not depend on the assertion thread winning a timing race. Limiter regressions include a deterministic lost-wakeup case where file eligibility changes between worker wakeups; the eligible owner is signaled explicitly. Tray tests cover preference migration, failed saves, batching, stale events, terminal scheduler events across queues, and quiet retry/pause transitions. python scripts/tray_smoke.py --binary-dir dist/IDM-EX-UnsignedDev checks native icon registration/removal, hidden downloads, second-launch restore, pause, exit, lock release, and shutdown with an open tray menu. It sends synthetic menu/callback messages only to its fixture windows and simulates TaskbarCreated only on the owned helper window; it does not restart Explorer or prove visible banner delivery. For actual Shell submission and synthetic notification-click routing, run cargo test --bin de native_notifications -- --ignored --test-threads=1. That opt-in test briefly creates its own icon and test notifications. python scripts/queue_ui_smoke.py --preferences captures the Tray & notifications dialog with the screenshot build. Screenshot fixtures do not create tray icons. Use --width 860 --height 600 for the minimum layout, --dark for dark appearance, --compact for compact rows, and --hide-details for the full list. --view queues (or scheduler/settings) captures an application page; --view browsers and --view help open Browser integration and About. The script saves both the original framebuffer and a PNG under a unique artifacts folder, using paused fixture queues and no runnable network downloads.

For an isolated GUI snapshot, build with cargo build --features screenshot, then set DOWNLOAD_ENGINE_HOME to an empty absolute test directory and EFRAME_SCREENSHOT_TO to an absolute .png path before launching target\debug\de.exe. This test-only feature captures the window and exits.

For the Classic Download Engine interface, run node scripts/native_ui_smoke.cjs after that build. It captures the real native renderer in both themes at normal and minimum sizes, plus About, Tray & notifications, Browser integration, queue management, Settings, an empty queue, and a previously saved column layout. All fixture queues remain paused in separate test directories; the report and PNGs are saved under artifacts/download-engine-native-ui. It also covers Add download, validation errors, filtered empty results, and queue controls at the minimum window size in both themes.

Use python scripts/queue_ui_smoke.py --schedule --width 860 --height 600 for the recurring schedule form, or --view scheduler for saved recurring windows. These previews keep every sample download offline. Scheduler tests cover start/stop boundaries, overnight weekdays, daylight-saving gaps/repeats, manual overrides, migration/reopen, and safe worker ownership. The HTTP fixture also checks exact resumed bytes across app exit and automatic cancellation.

Use python scripts/queue_ui_smoke.py --retry with that screenshot build for an isolated compact-table retry preview. python scripts/desktop_smoke.py --binary-dir dist/IDM-EX-UnsignedDev --retry checks real native handoff, a temporary HTTP 503, the observed retry delay, and byte-identical completion in the packaged app.

For a settled dialog capture, set IDM_EX_SCREENSHOT_TO to an absolute .ppm path instead, and launch the screenshot build with --browser-setup. This captures the app's own framebuffer after layout settles and then closes normally; no Windows screen-capture access is required.

Run python scripts/installer_smoke.py after building with -TestBuild to check install, upgrade, uninstall, running-app guards, a missing cleanup helper, and native download flows. It uses a fresh workspace folder, a separate uninstall identity, and isolated queue data, and verifies existing browser mappings are unchanged. The test build does not create shortcuts.

Code layout

  • src/browser_setup.rs: exact per-user browser registration and ownership checks.
  • installer/DownloadEngine.iss: per-user installation, upgrade guards, and safe uninstall.
  • src/engine.rs: HTTP validation, streaming, transfer dispatch, and safe publication.
  • src/engine/segmented.rs: adaptive HTTP range workers and partial-file accounting.
  • src/network_scope.rs: shared destination classification and rejection of DNS answers that mix public and local addresses.
  • src/auth.rs and src/credential_vault.rs: per-download origin binding, credential lifecycle, and Windows vault access.
  • src/download_file.rs: guarded Windows file handles, shared partial-file cleanup, cache protection, and Internet-origin metadata.
  • src/antivirus/: scanner configuration, persisted reports, guarded file review, and bounded child-process lifetime.
  • src/desktop/antivirus.rs: scan settings, results, and explicit opening confirmation.
  • src/store.rs: download records, capture reservations, and durable handoff receipts.
  • src/queue.rs and src/queue_store.rs: named queues, batch validation, ordering, migration, and schedules.
  • src/capture.rs: browser metadata and anonymous HTTP preflight validation.
  • src/scheduler.rs and src/scheduler/: scheduler API, worker settlement, event loop, queue policy, cancellation, and shutdown.
  • src/retry.rs: transient-error classification, bounded backoff, and Retry-After parsing.
  • src/preferences.rs and src/notifications.rs: persisted desktop options and terminal-attempt notification policy.
  • src/tray.rs: Windows notification-area resources, callbacks, and recovery.
  • src/main.rs: native desktop startup and single-instance ownership.
  • src/desktop.rs: shared desktop state, background events, and frame coordination.
  • src/desktop/: table, details, theme, and focused download, queue, authentication, browser, and settings dialogs.
  • src/bin/de-host.rs and src/native.rs: framed browser handoff.
  • extension/: shared JavaScript plus Chromium and Firefox manifests.
  • scripts/: portable/installer builds, optional registration, and isolated smoke tests.
  • analysis/: separate evidence from examining the user's installed IDM. No vendor implementation code is compiled or packaged into this application.

Native messaging follows the public browser interfaces: Chrome , Edge, and Firefox.

The Windows desktop smoke test is available as python scripts/desktop_smoke.py --binary-dir dist/IDM-EX-UnsignedDev. It uses an isolated queue and local HTTP server, checks exact output bytes, exercises native-host cold startup, and closes only its own application windows.

Antivirus integration can be checked without malware or antivirus configuration changes. Compile the harmless protocol fixture (it is never packaged): rustc --edition 2024 scripts/fixtures/antivirus_scanner.rs -o artifacts/antivirus-scanner-fixture.exe. Run python scripts/desktop_smoke.py --binary-dir dist/IDM-EX-UnsignedDev --antivirus-scanner artifacts/antivirus-scanner-fixture.exe. Use --antivirus-mode detected --antivirus-expected detected, --antivirus-mode error --antivirus-expected failed, or --antivirus-mode timeout --antivirus-expected timed_out to exercise failure paths. These tests verify the integration contract, not a vendor's detection quality. --antivirus-defender instead scans only the harmless downloaded fixture with installed Defender; its result depends on that machine's antivirus configuration. The tests use isolated settings/history and paths containing spaces and shell metacharacters.

The popup can be checked in an isolated headless Chrome session with node scripts/extension_popup_smoke.cjs (add --packaged-firefox to check the actual Firefox files in dist; requires Playwright as a development dependency). It applies the extension content policy and checks that injected inline scripts and web fetches are blocked. Its browser API adapter is simulated; this verifies the popup, not installation or native messaging in a real browser profile.

Add --automatic to the desktop smoke command to verify prepare/commit, a real file transfer, and cold desktop startup through the automatic-capture protocol.

The page picker has a separate rendered UI check: node scripts/page_links_ui_smoke.cjs. It verifies filtering, selection, saved folder choices, text-only rendering of untrusted names, narrow/light/dark layouts, and retry controls using a simulated extension adapter.

Installed-browser integration checks

Run python scripts/browser_smoke.py --browser chrome (or edge / firefox) after building a portable package. Chromium checks need Node and Playwright; pass --node <path> if Node is not on PATH. Firefox checks need Mozilla's marionette_driver==3.7.0, installed into artifacts/browser-test-deps (or pass --firefox-client-dir <directory>). Use --binary-dir target/debug to check a development build. These are opt-in Windows integration checks. Run one browser at a time: native-host registration is serialized across all browsers.

The runner uses the installed browser executable in a fresh headless profile, a disposable app copy with a path containing spaces, and local HTTP fixtures. It refuses existing Download Engine native-host registrations, creates a temporary registration, and removes it on exit. Normal browser profiles and the user's queue are not used. The checks cover stock-manifest popup connection and manual handoff, cold desktop startup, duplicate handoff, capture off/on, disconnected host fallback, exact bytes and Internet-origin marking. The desktop must survive browser shutdown and then exit normally when the fixture closes it.

Automatic-capture tests grant downloads access in a test-only manifest while retaining the optional declaration required by Firefox's permission API. The source/release manifests keep that permission optional. Actual permission prompts, store installation/signing, authenticated websites, and right-click context-menu interaction are not covered by these checks. Chromium uses its extension debugging protocol; Firefox uses Marionette temporary add-on installation. Reports and isolated test files remain under artifacts/browser-smoke-*.

Add --page-links to run the current-page batch flow in any of the three browsers. This uses the unmodified release manifest and confirms that page access is denied before invoking the real browser action. The driver triggers that action through CDP/Marionette and invokes the real popup button from a background extension tab, leaving the source page active. It checks link filtering and selection, the native batch receipt, cold desktop startup, a chosen paused queue and folder, duplicate prevention, completed bytes, and Internet-origin marking. It does not automate physical toolbar clicks or right-click menu interaction.

Browser batch unit tests also exercise rollback on job/receipt write failures, concurrent retries, changed-payload rejection, receipt persistence after removed history, and session recovery after interrupted acknowledgements.

Server-filename regressions cover Unicode/legacy encodings, quoted parameters, malformed/duplicate headers, redirects, range and single-response downloads, empty files, error metadata, explicit/browser names, old database records, resume/retry/restart, safe leaf paths, collisions, and Windows origin markers. Column tests exercise migration, rejected writes, pointer resizing/reordering, cancel/drop-outside behavior, file selection, Settings reset/save/cancel, and responsive sizing. Use python scripts/queue_ui_smoke.py --custom-columns to render an isolated customized layout, or --feature columns --width 860 --height 600 to render its settings at the minimum size.

Adaptive range verification

The regular Rust suite covers connection handoff, pause/resume, missing new part files after a saved split, stale progress checkpoints, the four-connection cap with many saved parts, validators, transient retries, checksum failures, failed layout commits, unexpected part files, and replacement-link verification in byte order. Fixed-range legacy records remain readable; versioned layouts and invalid partitions are checked independently.

Run the optional 32 MiB loopback timing fixture separately from other tests:

cargo test --release --locked --test adaptive uniform_and_uneven_benchmark -- --ignored --nocapture

It reports three complete-file timings for uniform connections and three for an uneven connection, including assembly and SHA-256 verification. Every sample checks exact output bytes and the observed connection cap. Timing is reported, not used as a pass/fail threshold; these local fixtures are not an internet speed guarantee.

The packaged desktop and native-host path can be checked with an isolated profile:

python scripts/desktop_smoke.py --binary-dir dist/IDM-EX-UnsignedDev --adaptive

This verifies an adaptive transfer, exact bytes and SHA-256, Internet-origin marking, normal shutdown, and native-host cold startup without personal queue data.

Downloads using your browser login

In Chrome, Edge, Brave or Firefox, sign in to the website normally and keep that tab open. Right-click a direct file link and choose Download using browser login…. Alternatively, open the extension popup on that site, paste the file URL, select Use this site’s login (review first), and choose Download with Download Engine.

The review shows the exact file URL and server. Choose Allow site login & download, then approve the browser's optional cookies and site permission. The review itself reads no cookies. The link must use HTTPS and the same origin (scheme, hostname and port) as the original site tab. Literal loopback HTTP addresses are allowed for local testing. Browser host permissions cover a hostname across ports, but Download Engine restricts the snapshot to the exact origin.

Only matching, unexpired, unpartitioned cookies are copied, including HttpOnly cookies. Firefox uses the original tab's cookie store/container. Private tabs, partitioned cookies and Firefox first-party-isolated sessions are unsupported; the extension does not broaden access to find another session. Automatic capture, page-link batches and video scanning remain anonymous. No browser passwords, profile databases, OAuth storage, arbitrary headers or POST bodies are read.

Each download receives its own snapshot, encrypted with Windows user-scoped DPAPI before it reaches SQLite. Job metadata contains only an opaque binding and origin. Encryption and the job/receipt commit in one database transaction; retries cannot overwrite a running download's login or create duplicate jobs. Cookies are never written to extension storage or diagnostic messages. Managed Rust secret buffers and native message frames are wiped on drop; the browser's JavaScript runtime does not provide guaranteed memory erasure.

A snapshot can resume after pausing or restarting on the same Windows account. It is usable for at most 24 hours, with each cookie also respecting its own expiry. Signing out or closing the browser does not erase a snapshot already sent to Download Engine. Completion, removing the download, Authentication → None, or accepting a replacement link on another origin removes the saved ciphertext. Removing the review's site permission stops future cookie reads, not existing desktop snapshots. Encrypted remnants may remain in SQLite journals or backups; these are not plaintext cookies. Backups of the data folder can contain ciphertext, while distribution archives exclude all runtime data.

Every HTTP redirect and range request recomputes cookies using domain, path, secure and expiry rules. A different origin or HTTPS downgrade is blocked before contacting that destination. Response Set-Cookie headers do not modify the saved snapshot. HTTP 401/403 and HTML login pages produce an actionable failure instead of a completed file. Session expiry never triggers automatic login or refresh.

If login expires, sign in again and send a fresh download from the browser. Refreshing a snapshot in place, cross-server login redirects, login-page HTML downloads, device-bound sessions and sites requiring POST requests are not supported. The extension limits each snapshot to 128 cookies and a 32 KiB Cookie header. Password/token authentication remains available separately in the app. DPAPI protects data at rest under your Windows account; it does not isolate secrets from other programs already running as that same account.

After updating, reload the unpacked Chrome/Edge extension or temporary Firefox add-on to load the new code. Signing and store publication remain separate work.