Media and webpages

Open Files → Downloader → Media, galleries & webpages in published v0.7.0. Ordinary file downloads retain their existing aria2 queue and CLI. Direct exchange between ProtoPeek devices has its own Local transfer workspace.

What each option saves

Go inspection reads at most 4 MiB of HTML and returns the first 100 unique media links. It does not execute JavaScript, unlock protected content, render a page or discover every image a browser could load. Original HTML snapshots leave linked assets external. They are not self-contained offline archives. Use ArchiveBox when a rendered webpage capture is needed.

Select positions 1–10000, at most 100 items per job. Each job gets a separate ProtoPeek-<id> output directory. Two jobs can run concurrently; history is capped at 128 jobs. Removing history preserves files. Downloads continue while ProtoPeek is running, including after closing the page. Cancel stops native requests or the external process tree. Retry preserves completed files; native incomplete files restart, while external engines may resume their own partial formats. An app restart leaves interrupted work stopped until explicitly retried.

Normal cancellation and application shutdown ask the OS to stop the owned process group or tree. An abrupt process kill or OS crash can leave external helpers alive; an interrupted job shown after restart does not prove those helpers were stopped.

One-click setup

The install button fetches the pinned upstream distribution, checks its exact byte count and SHA-256 digest, and stores it with license notices and a provenance receipt in the private ProtoPeek media tools directory. No terminal, separate Python setup or administrator prompt is needed for a supported managed build. Installed engines are used only for explicit jobs.

In current source, Extensions → Media opens a specific companion directly in Files. Before a managed job starts, ProtoPeek verifies the selected executable and required managed helpers. To uninstall, stop active jobs, choose Remove in the tool’s setup and confirm. Removal checks the pinned files and notices, and refuses modified or unexpected files. An interrupted installation with verified partial files offers Repair setup. Unknown files require your review. Your downloads, external installations and older versions stay in place.

Current managed builds and approximate compressed download sizes:

Video setup includes missing managed FFmpeg and a JavaScript runtime when a local Node installation is not available. The UI displays their combined download size. The executable bundle is larger on disk after unpacking. No automatic update or idle network check runs. internal/media/tools.json is the authoritative pinned asset manifest. yt-dlp is also available on Windows ARM64 and 32-bit Windows. Linux ARM64 yt-dlp and Deno, plus both macOS architectures, are also included. Other configurations show actual capability and installation help.

The Windows FFmpeg package is Gyan's versioned essentials build, linked from FFmpeg's download page. Its archive license and README identify the build configuration and source revision. yt-dlp standalone executables have aggregate third-party notices in addition to the Unlicense source; Deno and gallery-dl retain their own upstream notices. These are separate upstream distributions, downloaded on request, and are not embedded in ProtoPeek.

FFmpeg enables separate video/audio merging. Without it, video selection is limited to a combined format. Audio keeps the source's native audio format; there is no implicit MP3 conversion. Node must meet yt-dlp's supported runtime version requirements. TLS verification stays enabled. User config files, arbitrary command flags, external yt-dlp plugins and remote components are not loaded.

ArchiveBox has a larger Python/browser/tool stack and does not officially support native Windows. ProtoPeek can invoke a separately installed compatible CLI on Linux/macOS, with no recursive crawl and no Internet Archive submission. On Windows the UI links to upstream Docker/WSL setup; it does not claim to install Docker or enable Windows features. Its collection index reports per-format success/failure; a completed command does not prove every requested format saved.

Architecture and size

Go owns request validation, HTML extraction, streaming, the local queue, installer verification, process cancellation and saved history. It adds no new Go dependency. Media and file views load independently; neither optional Python engine is embedded into the application binary. The new route has an 18 KiB raw JavaScript budget; the aggregate allowance grows by this route and 4 KiB of CSS, while startup and existing file-view limits remain unchanged.

Reimplementing the common path in Go keeps it small. Retaining maintained site extractors avoids pretending that generic HTML parsing covers YouTube signatures, site APIs or frequent platform changes. Speed depends on bandwidth, source limits, format and post-processing; there is no universal speed guarantee.

References: yt-dlp, gallery-dl (active releases on Codeberg), ArchiveBox, Cobalt. No Cobalt extractor code is copied. External executables keep their own upstream licenses: gallery-dl is GPL-2.0, FFmpeg builds have GPL components, ArchiveBox is MIT, and yt-dlp standalone executables contain third-party components beyond its Unlicense source.