bridge architecture with page context downloading would be killer

[tomkabel]

Hi,

I’m a developer and came across Flowpick while researching media sniffing implementations. I found the project really interesting!

For downloading media, I’ve historically relied on yt-dlp, falling back to the Video DownloadHelper extension for unsupported sites. Recently, Video DownloadHelper started pushing premium pricing (the usual enshittification), which made me wonder if it's worth building something custom, better, and faster.

I previously built a student project called Vooglaadija (Estonian for "stream downloader"), which wrapped yt-dlp. However, yt-dlp can be fragile and misses media streams on a lot of sites. That left me debating whether to build a new tool from scratch or contribute to something existing.

When I tested the Flowpick extension, I thought it was very well built, but I ran into an issue with the m3u8 downloader tool on a site that enforces strict Referer headers. After digging into it, I realized the issue was architectural: downloads need to be executed from the page context itself (which seems related to issue #4, if I translated and understood it correctly).

To fix this, you need a sort of "bridge" architecture:

Injected script (page main world)         ← runs natively in the page
  ↓ postMessage("FLOWPICK_FETCH_BRIDGE_INIT", port)
Content script (isolated world)           ← extension context, has chrome.runtime
  ↔ MessagePort (main ↔ content)
  ↔ MessagePort (content ↔ iframe)
  ↓ injects <iframe src="download-host.html">
Download host (iframe, extension origin)  ← extension context, has OPFS
  → StreamDownloadEngine
  → PageFetchClient (bridges fetch through ports)

I built a prototype of this bridge and it worked well, enabling downloads even for source-protected streams.

I'm curious—are you still actively working on this project, or is it currently passive? Have you hit a specific blocker, or would you be interested in collaborating on a v1.1 (or v2) release for Flowpick?

[funcy-boy]

Hi there!

Really appreciate you taking the time to dive so deep into FlowPick's architecture and sharing such a well-thought-out technical breakdown! Your bridge architecture design (leveraging injected scripts, message passing, and extension-origin iframes) is extremely clean and spot-on.

As you correctly guessed, handling streams behind strict headers/anti-leech policies has been a fun engineering challenge. That particular issue with strict Referer constraints has actually already been resolved in our latest updates. Because of limited bandwidth and personal time, my manual site testing coverage hasn't been exhaustive, so running into edge cases like the one you found is totally expected.

To answer your question about the project status: FlowPick is definitely still active! In fact, v1.1.0 has just been officially released, bringing several improvements and stability fixes.

I’d love to keep in touch and chat more about media sniffing, extension architectures, or stream processing. Feel free to open an issue/PR on GitHub or reach out further if you're still interested in collaborating or sharing ideas for future versions.

Thanks again for the great insights and support!

[funcy-boy]

Note: The update for Microsoft Edge is currently still going through their review process; if you are using Edge, you can temporarily switch to Chrome/Firefox, or download the release package to install it manually.

[tomkabel]

Ah nice, I saw you pushed v1.1—my extension auto-updated, which brought me back here to see your messages.

Good to see you and the project are still active! In that case, I’ll likely contribute a few things down the line.

This is a whole new field for me since I don't have any past experience with browser extensions or media sniffing, but researching it has been pretty interesting nonetheless.

[funcy-boy]

Thanks so much for the feedback! It’s great to hear that you’re enjoying the update. I really appreciate your support and interest in contributing. Honestly, I'm quite new to this as well, and I’ve been heavily relying on AI assistance to help build and refine this project. It’s been a fascinating learning curve, and I'm glad to have you along for the journey!😄

[tomkabel]

Yeah, it'd be a pleasure. Your tech stack and setup for the extension look really solid—pretty much what I would’ve picked myself.

And I can definitely confirm there's a ton of agentic development and research involved. Even then, you still have to constantly steer and nudge the LLMs in the right direction when dealing with complex edge cases.

If you're using AI to code, I'd suggest setting up CodeRabbit for the repo. It's a totally free PR reviewer with sensible rate limits. (If you'd rather not set it up repo-wide, no worries—I can just run their CLI on my end whenever I open a PR.)