video-lens GitHub

Open source · for Claude Code, Codex, Cursor and more

Read the video. Watch the part you need.

Paste a YouTube link into your agent. Get a summary and a timestamped outline beside the video, click any line to jump there, and send it as a link anyone can open.

npx skills add kar2phi/video-lens

Example: a shared report

Made by video-lens from a 66-minute interview. Someone said “share” and sent this link.

Open the example report
A video-lens report of Andrej Karpathy’s No Priors interview. Left: the video, a bar of its 13 sections with the fifth playing, and the outline with that section open. Right: the title, the summary and the takeaway. The same report on a phone: the player stays at the top, with Read and Outline tabs under it.

How it works

From link to report in a minute

  1. Install once

    One command; pick all three when asked. The first video asks before installing what’s missing.

  2. Paste a link

    With the command or in your own words. Name a language or a part to focus on.

  3. Read beside the video

    The report opens in your browser and is saved as a file. Ask for changes and it’s rewritten in place.

  4. Share

    Say the word and you get a link anyone can open.

npx skills add kar2phi/video-lens

summarize https://youtu.be/kwSVtQ7dziU

The transcript library isn’t installed yet. Install it into video-lens’s own folder?

yes

✓ Installed. Fetching the transcript…

✓ Report open in your browser

tighten the takeaway

✓ Rewritten in place.

share

✓ https://video-lens.org/s/p02ia3wYih990WIcW3jmmQ/

Features

Four things to ask for

video-lens · what you can sayType it, or say it your way

A report

Summary, key points and the outline beside the video.

/video-lens <url>
the command
summarize this video
or just ask
… in German
any language
… focus on pricing
one part only

A link to send

Unlisted, with the same player beside the text.

share
get the link
unshare
take it down
video-lens.org/s/…

Your library

Every report you made, with thumbnails, summaries and tags, searchable.

/video-lens-gallery
the command
show my gallery
or just ask
The gallery: a search field, tag filters with counts and a grid of saved reports, among them Karpathy on No Priors, 3Blue1Brown on neural networks and Veritasium on entropy, each with its thumbnail, channel, length, summary and tags.

What to watch next

New videos on a topic, from your subscriptions and beyond, ranked with a reason each. It skips what you’ve watched.

/video-lens-recs <topic>
the command
anything good on Rust lately?
or just ask
Recommendations for the AI topic: ranked picks from the last seven days, the first a Stanford lecture at about 23 times its channel’s usual views, each with a reason and buttons to watch, summarize or copy the link.

One-time setup: import your subscriptions (about 5 min) or connect your browser.

Reports come in the video’s language, use its chapters and export to Markdown.

Privacy, in short

Your reports stay on your machine until you say “share”

On your machine

  • Reports are plain HTML files in ~/Downloads/video-lens/reports/. No account, no cloud.
  • Captions come from YouTube; the report is written by the model your agent already uses.
  • Nothing is uploaded unless you ask, and video-lens asks before your first share.

When you share

  • The link is unlisted, not private: there is no index of shared pages.
  • Only the report goes up. Never the transcript or the video description.
  • “unshare” deletes it. The key that can delete it stays on your machine.

This site sets no cookies. It counts visits with Cloudflare Web Analytics, which needs none. Read the full privacy policy

Questions

Why not just ask a chatbot?

They can summarize a video too, as an answer in a chat. video-lens makes a report that sits beside the playing video, with every line one click from its moment. It stays a file on your disk, in a gallery you can search; it is written by the model your agent already runs; and “share” turns it into a link.

What runs where?

video-lens is instructions and a few Python scripts your agent runs. It fetches the captions, your agent’s model writes the report, and the scripts render it into a page on your machine.

What does it cost?

video-lens is free and MIT licensed, with no account or subscription of its own. The report is written by the model your agent runs, local models included, so it costs what a prompt of that length costs you there. A one-hour talk is a long prompt.

Which videos work?

Public YouTube videos with captions in any language, including auto-generated ones. YouTube Shorts are not supported.

What if a video has no captions?

On Apple Silicon, video-lens can transcribe the audio on your machine with Whisper. It asks first, and the report is the same.

Do I need a server or an account?

No. video-lens starts a small local web server to show the report and saves it as a file. Sharing is the only part that uses video-lens.org, and only when you ask for it.

Can I trust the summary?

It is written by a language model from the captions, so treat it like notes from a fast colleague: every key point and section has its time, one click away from what was actually said.

Try it on the next video you were about to watch

npx skills add kar2phi/video-lens