mathiaschu

mathiaschu / watch

Public

Give Claude a video input. /watch downloads from YouTube/Instagram/X/Vimeo/any yt-dlp site, extracts frames, and transcribes locally with mlx-whisper — no API key. Fork of bradautomates/claude-video.

claude claude-code claude-skill mlx transcription
13
1
85% credibility
Found May 31, 2026 at 13 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

/watch is a skill for AI assistants that lets you ask questions about any video and get answers based on what the video actually contains. You paste a video URL (YouTube, Instagram, X, Vimeo, TikTok, and thousands more) or point to a local file, ask your question, and the tool downloads the video, extracts key frames, pulls or generates a transcript, and hands everything to the AI. The standout feature is privacy: transcription runs entirely on your machine using local speech recognition, so audio never leaves your computer. No account, no API key, no subscription — just you and your videos.

How It Works

1
💡 You discover a tool that lets AI watch videos

You hear about /watch — a way to ask an AI assistant questions about any video, and it actually sees and hears everything in it.

2
📦 You install the skill in one line

You add it to your AI assistant with a simple command, and run a quick setup that prepares everything on your computer.

3
🎬 You paste a video URL and ask your question

You share a YouTube, Instagram, or any video link along with a question like 'what happens at the 30-second mark?'

4
The tool works its magic automatically
📺
Public videos

The video downloads directly and captions are pulled automatically when available.

🔐
Login-protected videos

If a video needs login (like private Instagram reels), you can share your browser cookies just for that download.

5
🎧 Audio is transcribed privately on your device

If a video has no captions, the audio is converted to text right on your computer — nothing ever leaves your machine.

6
🤖 Your AI assistant sees the entire video

It reads each frame as an image and aligns everything with the transcript, understanding both what's shown and what's said.

✅ You get an answer grounded in what actually happened

No guessing — your assistant answers based on what it actually saw and heard in the video, with timestamps to back it up.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 13 to 13 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is watch?

This is a Python skill for Claude Code that gives the AI "eyes" to watch videos. Paste any URL from YouTube, Instagram, X, Vimeo, or 1800+ other sites, ask a question, and Claude downloads the video, extracts frames, pulls a timestamped transcript, and answers based on what it saw and heard. Local video files work too. The key differentiator: transcription runs entirely on your machine via mlx-whisper (Apple Silicon) or openai-whisper (CPU), with no API key required. Native captions are used when available; audio is transcribed locally as a fallback.

Why is it gaining traction?

The privacy angle is the hook. Audio never leaves your machine, and there's no config file, no API key, no account. For developers who want to analyze video content with Claude without sending data to third parties, this fills a gap. The cookie-borrowing feature for login-gated sources (Instagram, X, private YouTube) is practical for real-world use cases. The auto-scaled frame extraction keeps token costs reasonable across video lengths.

Who should use this?

Developers using Claude Code who need to analyze video content—debugging visual bugs, reviewing tutorial recordings, extracting information from presentations, or working with social media content. Researchers processing video datasets. Anyone who wants Claude to "watch" something without setting up API credentials or paying for cloud transcription.

Verdict

A genuinely useful niche tool with a solid privacy story, but early-stage: only 13 stars and minimal community feedback. The documentation is thorough and the local-only design is well-executed, but test coverage and long-term maintenance are unknowns. The 0.85% credibility score reflects this infancy. Worth trying if you have a specific use case; check the repo activity before committing to it as a workflow staple.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.