🫳 grab

Download and archive content from URLs into organized folders with transcripts and AI summaries.

Drop a URL, get everything saved — text, media, transcripts, summaries — organized and synced to Dropbox (or wherever you want).

Supported Sources

Install

brew install yt-dlp ffmpeg openai-whisper

Optionally, set your OpenAI API key for AI summaries and smart folder titles:

export OPENAI_API_KEY="sk-..."

Without it, media downloads and transcription (local Whisper) still work.

Clone and make the script available:

git clone https://github.com/jamesalmeida/grab.git
chmod +x grab/scripts/grab

Or if you're using OpenClaw, just drop the grab folder into your workspace skills/ directory.

Usage

grab <url>
grab --config    # Change save directory
grab --help      # Show help

First Run

On first run, grab asks where to save files:

🫳 grab — First time setup

Where should grabbed files be saved?
  Default: ~/Dropbox/ClawdBox

Save directory (press Enter for default):

Config is stored in ~/.config/grab/config.

Output Structure

Downloads are organized by type, each in its own dated folder:

~/Dropbox/ClawdBox/
  XPosts/
    2026-02-03_embrace-change-you-can-shape-your-life/
      tweet.txt
      video.mp4
      transcript.txt
      summary.txt
  XArticles/
    2026-01-20_the-arctic-smokescreen/
      article.txt
      summary.txt
  Youtube/
    2026-02-03_state-of-ai-in-2026/
      video.mp4
      description.txt
      thumbnail.jpg
      transcript.txt
      summary.txt
  Reddit/
    2026-02-03_maybe-maybe-maybe/
      post.txt
      comments.txt
      video.mp4
      summary.txt

Features

• Auto-transcription — Videos are transcribed locally via Whisper (no API key needed)
• AI summaries — Every piece of content gets an AI-generated summary with key insights
• Smart folder naming — Video folders are renamed based on transcript content (not just the tweet text)
• Image descriptions — Image-only tweets get folder names from AI image analysis
• Configurable save location — First-run setup, change anytime with --config
• Reddit comments — Saves top 20 comments with authors and scores
• YouTube metadata — Description, thumbnail, view/like counts all preserved

How It Works

Tweets with video: Downloads video via yt-dlp → extracts audio → transcribes with Whisper → generates summary → renames folder to descriptive title from content.

Tweets with images: Downloads images → analyzes first image with vision model → renames folder to image description.

X Articles: Detected automatically (exit code 2). Requires OpenClaw agent with browser to extract article content (X blocks headless access).

Reddit posts: Fetches via Reddit JSON API → saves post + top comments → downloads any video/images → transcribes video → generates summary of post + discussion.

YouTube: Fetches metadata → downloads video (1080p max) + thumbnail → transcribes → generates summary.

Requirements

• yt-dlp — media downloads
• ffmpeg — audio extraction and video merging
• curl — API requests
• python3 — JSON parsing and API calls
• whisper (openai-whisper) — local transcription (no API key needed)
• OPENAI_API_KEY — optional, for AI summaries and smart folder titles (GPT-4o-mini). Without it, media downloads and transcription still work.

OpenClaw Skill

This is an OpenClaw / AgentSkills-compatible skill. The SKILL.md contains metadata for automatic discovery and gating.

When used with OpenClaw, the agent handles:

• X Articles (browser-based extraction)
• Reddit fallback (when JSON API is blocked)
• Automatic OPENAI_API_KEY injection from config

License

MIT