Skip to main content

A free, open-source tool that lets AI agents watch videos locally on Apple silicon Macs, returning a transcript and frame sheets tied to what was said.

About Vidscribe

Vidscribe is a free, open-source tool by Samuel Zhang that lets AI agents watch videos locally on Apple silicon Macs. The official repository is github.com.

Give it a local file or any URL yt-dlp handles. It returns a transcript and, for video, overview sheets (small, evenly spaced pictures) with an index.txt mapping each one to what was being said. The agent can then fetch full-size frames at the moments that matter. It runs on-device with MLX: no API keys and no cloud. The default model is Qwen3-ASR, at about 2.5 GB, and downloads on first run. The README says it covers about 30 languages, with a Whisper large option for others.

Install needs an Apple Silicon Mac and Homebrew. Clone the repo and run ./install.sh, then put ~/.local/bin on your PATH. It also installs as an agent skill: npx skills add SamuelZ12/vidscribe, which needs Node.js 22.20+. Without Node, you can copy the skill/ folder into your agent's skills directory.

Limits listed in the README: it transcribes speech only, and does not describe music or other sounds; it has no OCR or word timestamps, so agents read on-screen text from the frames. Outputs stay in /tmp/vidscribe/ for 3 days. The license is MIT; the models, ffmpeg and yt-dlp have their own licenses.

In the launch thread, Zhang also said vidscribe uses ffmpeg under the hood, and that it is local, free, and works with any agent.

Pricing: Free and open source under the MIT license. It runs locally on an Apple silicon Mac. Checked November 17, 2026.

Pricing model: Free

Features

  • Runs locally on Apple silicon Macs with MLX, with no API keys or cloud
  • Accepts a local file or any URL yt-dlp handles
  • Returns a transcript plus overview sheets of evenly spaced frames and an index mapping each frame to what was being said
  • The agent can then fetch full-size frames at the moments that matter
  • Installs as an agent skill with npx skills add SamuelZ12/vidscribe
  • MIT licensed

FAQs

Common questions about Vidscribe

What is Vidscribe?
A free, open-source tool that lets AI agents watch videos locally on Apple silicon Macs. It returns a transcript, overview frame sheets and an index tying each frame to what was said.
Where does it run?
Locally on Apple silicon Macs, with MLX. It does not use a cloud API or API keys.
How do I install it?
Clone https://github.com/SamuelZ12/vidscribe and run ./install.sh, or install the agent skill with npx skills add SamuelZ12/vidscribe (Node.js 22.20+).
Does it caption music or on-screen text?
The README says it transcribes speech only, and does not describe music or other sounds. It has no OCR, so agents read on-screen text from the frames.

Featured