Automatic(-ish) summary of saved youtube videos

[j-xella]

When I watch youtube videos, sometimes I save (more like create a note about) them to Obsidian. I use obsidian web clipper with a youtube template.

I want to add a short AI generated summary of the video as the note is being created. These are my goals:

1. I want the summary to be generated at least from the video transcript, or better from the video itself (if the video contains important graphics). In my experience, video descriptions provided by authors are not that good.
2. The process should be as automatic as possible. I don't mind configuring, but when saving a video, ideally I would just like to click "Save", and the description will appear in the note. On the other end of the spectrum, there is copying the video URL, opening ChatGPT/Gemini/etc, writing a prompt to give a summary of the video, then copy/pasting the result into Obsidian note - I want to optimise this as much as possible.
3. I believe Obsidian copilot can be used for this (I am a believer, BTW 😁), but if this can be achieved with other tools, I am open to suggestions. I need the result.
4. If the perfect solution is not possible, any tips that minimize the number of steps are welcome.
5. I am happy for this question to become a feature request, if feasible.

This is what I have tried so far:

---

Copilot web clipper has a native integration with AI. Unfortunately, that AI can only be asked about the HTML content of the page you are trying to save. I tried a template with a prompt like "Give me a summary of youtube video {{url}}", got an error that use of external sources is forbidden.

---

Once there is a note in Obsidian about the youtube video, I tried copilot with prompts like this:

@composer find a youtube link in this note, generate a short summary of that youtube video and append it to the bottom of the note

And got a response:

I understand that the youtubeTranscription tool requires the YouTube URL to be directly in your prompt. However, the URL is present in the note provided in the context: https://www.youtube.com/watch?v=XXXXXX.

Since I cannot directly use the youtubeTranscription tool with a URL from a note, I will first extract the URL from the note. Then, I will use webSearch to find information about the video, which should allow me to generate a summary. Finally, I will append this summary to the note.

Well, currently that plan does not work, if fails during the web search phase.

I can get a summary if I manually paste youtube URL into the prompt, but that means that I can't create a custom prompt for this...

[logancyang]

Thanks for the details! The youtube transcription tool used to be agent callable before, but then we decided to trigger it only when youtube urls are explicitly in the chat input. This was to avoid mistriggers if a note contains a long list of youtube urls and the agent decides to call the youtube tool on every one of them (youtube transcript calls are expensive).

That said, I do see the need to make youtube transcription more agentic as you've made the case. The middle ground could be to let the agent ask for permission if there are 2 or more videos involved. For a single video transcription the mistrigger shouldn't be a huge deal.