Why another YouTube summarizer?

TubeFold is a local Mac app, not another cloud service.

No extra subscription. Use the Claude or ChatGPT plan you already have.
Your notes are files. Every summary is saved as Markdown on your Mac.
Make summaries your own. Edit the prompt and choose the output language.
Add videos in one click. Send the video you're watching straight from Chrome.

How it works

Five simple steps

Paste a link, or click the Chrome extension while watching. TubeFold does the rest and files the note in your library.

YouTube URLpasted in the app, or sent with one click from the browser extension

Transcriptfetched in the video's original spoken language, with the metadata

Prompta plain-text template you can read and edit wraps the transcript

Your CLIclaude or codex runs it on your own subscription. No API key involved.

Markdown notestructured summary + metadata front matter, saved to your library as a file

What a note looks like

Structured, sourced, and yours. Ready for Obsidian, a docs folder, or anywhere plain text lives. The metadata front matter is generated by the pipeline, never invented by the model.

The Craft of Software Design.md

# The Craft of Software Design

## Overview
A 3-minute read of the talk's core argument…

## Key ideas
- The main claims, kept faithful to the speaker…

## Practical takeaways
- What you can actually apply…

## People, products & sources mentioned
- Every reference worth following up…

A real Mac app around it

Chrome extension Send the video you're watching with one click.

PDF exportAny note, saved next to its Markdown file.

Telegraph publishingOne click turns a note into a shareable public article.

Your languageSummaries in any language, independent of the video's.

Small enough to actually read

TubeFold is MIT-licensed: a SwiftUI app supervising an embedded Python backend, with a pipeline you can trace file by file. The prompt is a Markdown template you can edit. A provider is just an executable that takes a prompt file and writes Markdown back. Adding another AI backend, including a local model, is a small shell script. If privacy matters to you, don't take our word for it: read the code.

Star or fork on GitHub Contribute

Questions

Why not just paste the transcript into Claude or ChatGPT?

That works. It's the manual version of exactly this pipeline. TubeFold adds automatic transcript fetching (with original-language selection), a consistent tested prompt, metadata front matter, sensible filenames, a deduplicated library, browser one-click capture, and PDF export. If you do it more than a couple of times a week, the automation pays for itself.

Do I need a paid AI subscription?

You need a working claude or codex CLI signed into your account. Summaries run on your plan's quota. TubeFold shows per-provider token usage in Settings so you can see what it consumes.

Is it really local? Does my transcript leave my Mac?

The app, backend, library, and files are local. The AI inference is not. The transcript goes to Anthropic or OpenAI through your own CLI, the same as pasting it into their chat yourself. TubeFold adds no servers of its own.

Why does it require Claude Code or Codex?

That's how TubeFold uses your existing subscription without ever touching your credentials: the official CLIs handle authentication themselves. It's the trade that makes "no API keys, no second subscription" possible. If you already have either CLI installed, setup is three minutes.

Why trust a new open-source project?

Don't trust us. Read the code. The privacy-relevant code is a handful of small Python files with no web framework. Because your notes are plain Markdown on your disk, the exit cost is zero: if TubeFold disappeared tomorrow, your notes wouldn't.

Watch less. Keep more.

Your next long video can end as a note you'll still have in ten years.

Download for macOS

$ brew install --cask tubefold/tap/tubefold

macOS 26+ Free & open source View on GitHub