Defuddle now returns Youtube transcripts! Paste a YouTube link into defuddle.md to get a markdown transcript with timestamps, chapters, and pretty good diarization! ...or if you just want to read it, try the new Reader mode in Obsidian Web Clipper powered by Defuddle.

Defuddle and Obsidian Web Clipper are both open source (MIT license)

how does it compare to ?

Give it a try and tell me! I think Defuddle is better in every way: - Runs in any environment (curl, cli, node, browser, self-host, etc) - Produces much cleaner Markdown, including footnotes, LaTeX, callouts, etc - Returns YouTube transcripts - Returns frontmatter with structured variables for author, published date, etc see also

Maybe it's still getting hammered right now, maybe it's where I'm at in learning how to use it, but three things I'm still trying to work out: • Getting the transcript to populate in the Web Clipper dropdown (it does on the defuddled page). • Getting that Reader view • Getting Claude in the Web Clipper

What version of Obsidian Web Clipper do you have? The new features require Obsidian Web Clipper 1.1.0 (available on all browsers except Firefox, still pending approval there) If you want Claude in Web Clipper you need to set up Interpreter:

I use Arc (Chrome) and Firefox and noticed Firefox was lagging behind, but Arc is at 1.1.0. Noted on Intrepreter.

So it works now? You should see a new book icon in the popup, or you can configure a hotkey, see Web Clipper settings

STILL TRYING: Getting the transcript to populate in the Web Clipper dropdown (it does on the defuddled page). ✅ Getting that Reader view IN PROGRESS: Getting Claude in the Web Clipper

What Web Clipper template are you using? Transcripts require using the {{content}} variable in your template. (It's the default template)

Oh whoa, I was doing the extra step of prepending "defuddle.md" to the front of the URL, but I actually can just click the Web Clipper directly on a YouTube page and get everything perfectly. Problem solved ✅

was excited about this but seems like it's not working

try again, I think it's just getting hammered at the moment — you can also run defuddle locally if you want

@johnnydevriese same here do I need an extension installed for it to work?

@johnnydevriese the problem seems unrelated to Defuddle, YouTube says "This video isn't available anymore"

@johnnydevriese idk, I tried another yt video ive just watched and Defuddle gives the same header with no transcript

@johnnydevriese can you try again now?

@johnnydevriese It worked last week, but now I'm getting an error on limits {"error":"Failed to fetch: 429 Too Many Requests"}

@johnnydevriese should be fixed now?

I'm running out of "wow, amazing"s, honestly I want to derive so much from some profound Youtube vids but even Gemini wouldn't go in that much depth. now defuddled md transcripts would work miles better

thank you! 🙏

bro... that's all I need. before, I would open the transcript, scroll down line 1 to the last line (which often didn't work). This saves an absurd amount of time, thanks

:-( tried 7 videos to give a fair chance, same 287b file

Doesnt seem to work for all youtube, does it ? this didnt give me the entire script - This is all I got

it works for me, but I am beginning to suspect there are some differences I need to account for if you're accessing defuddle outside of the US anyway that should only affect the defuddle website, if you use web clipper it should work

That is right, the clipper works. Thanks ! Very useful tool.

can you try again on defuddle.md?

still the website doesnt work

I was surprised to learn that Defuddle already seems to handle X articles (the long form posts) well! 🫡

✅

Great stuff! Seems it works off of a Cloudflare worker? Curious how you work around the YouTube rate limits/IP blocks?

API for defuddle?

see also

Podcasts transcript would also be cool

@lucabernardi what URL? some podcast websites already have transcripts, which should be picked up automatically

I was thinking the ones on like

@lucabernardi Defuddle is an HTML-to-MD tool not a transcription service. It's able to get YouTube transcripts because they're on the page. It can't get transcripts from web pages that don't provide them.

Ah, gotcha! I didn't realize that you were exclusively relying on the innertube's YouTubei/v1 endpoint. Digging more into the project's code, I see why what I was suggesting wouldn't make sense. Thank you!

Does it not work for YouTube shorts?

defuddle is insane. extracting transcripts with diarization is something i need all the time. open source plus obsidian integration is perfect

Reader mode is so good, please add it to the rightclick menu for the extension

will do, but the fastest is the keyboard shortcut, I have it set to Opt+Shift+R

Can reader mode be triggered via other plugins? Would love to make a plugin that automatically shows websites in reader mode based on a user-configurable whitelist. (if you'd consider native functionality: the plugin could show "Always do for this site?")

!!!!!!!!!!!

Love trascripts from YouTube. This is how I do it with echOS:

Sweet!

I do not have an AI interpret section in my Obsidian Clipping plugin. Is this a new feature that will be rolling out in a later release? Browser specific?

This is insane. Keep up the good work! Building something opensource using this. Hopefully people will find it useful!

@obsdmd That’s great @kepano!

nice addition with the transcript support. for anyone needing thousands of transcripts at once i built a youtube transcript scraper on apify that handles full channels in bulk, timestamps included. 154 users running it for RAG pipelines and content research.

this is awesome thanks for sharing with the rest of us.

what is claude doing on the interpretation here?

it's using a prompt variable that summarizes the transcript

Ever have any trouble with the output length? Like token limits?

Interpreter is model-agnostic, any limitations come from them model

Did YouTube transcripts break?

I think this might be what moves me from Readwise to obsidian web clipper.

dude... you are goated fr.

Literally built this last weeks. Will see how it compares.. the race for software is on. 👀

magic

Hey, thanks for the tool! It`s working quite reliably now and it`s become my go-to for summarising yt

what about notebookllm?

already works with notebooklm

I am in California, currently is not working for youtube. thank you for making this tool though.

did you try running it locally? (e.g. via npx)

yes, i tried to run it locally in terminal

Working now for some, but still two common issues I guess: 1. IP geo-bias: A Language IP fails on B Language videos (maybe YouTube returns wrong lang tracks). 2. Language code mismatch: metadata "en-GB" but actual captions are just "en" ? (e.g. Dwarkesh video).

is this still true? have been hitting Youtube rate limits when I tried to use Defuddle on a YouTube video 🤔

Should work, but there are many ways to use Defuddle locally if that happens. Try Obsidian Web Clipper or the Defuddle CLI.

sorry! I'm not that technically savvy. I had asked Claude to create a skill out of defuddle, and that's failing now. But, I'll try to figure how to get it working based on your suggestion. Thank you.

running locally already and apparently the issue was a transient one. working now 🙏🏼