Stop Losing Video Knowledge Forever: transcribee Builds Your Brain

Stop Losing Video Knowledge Forever: transcribee Builds Your Brain

You just watched a 90-minute Lex Fridman interview with Ilya Sutskever. Mind = blown. You learned three things about AGI alignment that could change your entire project direction.

Forty-eight hours later, it's gone.

Poof. Vanished into the same mental black hole as that productivity hack from last Tuesday's YouTube rabbit hole and the startup advice from YC's latest batch. You remember feeling inspired. You cannot recall a single actionable insight.

Here's the brutal truth: The average developer consumes 5+ hours of video content weekly and retains less than 10% of it. We're building our careers on a foundation of forgotten knowledge. We bookmark, we save to "Watch Later," we tell ourselves we'll take notes. We never do.

What if your Mac could automatically turn every video you watch into a searchable, organized, LLM-ready knowledge library? No manual note-taking. No chaotic folders. No "where did I see that thing about...?"

Meet transcribee — the open-source macOS tool that's making manual transcription and disorganized notes feel as outdated as Subversion. And the best part? It literally builds your knowledge base while you sleep.

What Is transcribee?

transcribee is an open-source macOS video transcriber that transforms passive video consumption into an active, self-organizing knowledge management system. Created by developer Fabio Roma, this tool doesn't just spit out raw text — it architecturally structures your learning into a persistent, queryable library that evolves alongside your interests.

The project emerged from a painfully relatable frustration: Roma consumed massive amounts of video content across YouTube, Instagram Reels, TikTok, and podcasts, but had no systematic way to retain, search, or leverage that knowledge. Existing solutions were either manual (painful), expensive (proprietary SaaS tools), or produced disorganized outputs that became useless at scale.

transcribee solves this through an elegant pipeline: download audio → transcribe with speaker identification → auto-categorize using Claude's intelligence → save to a structured folder hierarchy. The result is a ~/Documents/transcripts/ directory that naturally mirrors your intellectual evolution.

Why it's trending now: The convergence of three forces makes transcribee explosively relevant. First, video has become the dominant knowledge medium for developers (documentation, conference talks, tutorials, podcasts). Second, LLMs have created unprecedented demand for structured text inputs — raw video is useless to ChatGPT, but clean, speaker-labeled transcripts are gold. Third, the "second brain" and personal knowledge management movements have hit mainstream developer consciousness, yet most tools ignore video entirely.

transcribee sits at this intersection, and the GitHub community is responding. It's not just a transcription tool — it's a knowledge infrastructure play for the AI age.

Key Features That Separate transcribee from Basic Transcribers

Let's dissect what makes this tool genuinely powerful, not just convenient.

Multi-Platform Video Ingestion transcribee handles the full spectrum of modern video consumption: YouTube videos and Shorts, Instagram Reels, TikTok clips, and local audio/video files. This matters because developer learning happens everywhere — a TikTok tip about Docker^{↗ Bright Coding Blog} optimization is just as valuable as a two-hour conference keynote. The tool uses yt-dlp under the hood, which means it stays current with platform changes and supports hundreds of sites beyond the officially listed ones.

Speaker Diarization — Know WHO Said WHAT This is where transcribee transcends "dumb" transcription tools. Using ElevenLabs' scribe_v1_experimental model, it identifies different speakers and labels them in the output. For podcast interviews, panel discussions, or multi-person tutorials, this transforms unusable wall-of-text into structured dialogue you can actually query. Imagine asking Claude: "What did the security researcher say about supply chain attacks?" instead of wading through 50 minutes of transcript.

AI-Powered Auto-Categorization Here's the secret sauce. After transcription, transcribee sends the content to Claude (Anthropic's API), which analyzes both the transcript and your existing library structure to determine optimal categorization. It's not keyword matching — it's contextual understanding. Your library organically grows folders like AI-Research/, Startups/, Health/ based on actual content themes, not rigid pre-configuration.

Native LLM Integration Ready Every transcript.txt is formatted specifically for pasting into ChatGPT, Claude, or any LLM. No preprocessing, no cleaning, no "please summarize this messy text." The speaker labels and clean formatting mean you can immediately ask sophisticated questions about the content.

Raw JSON for Power Users Need word-level timestamps? Confidence scores? The --raw flag outputs the complete ElevenLabs API response as transcript-raw.json for applications requiring precision — think automated highlight extraction, precise citation linking, or custom downstream processing pipelines.

Real-World Use Cases Where transcribee Transforms Your Workflow

1. The Research Accelerator

You're investigating retrieval-augmented generation (RAG) architectures. You've got twelve YouTube talks, three podcast episodes, and two conference recordings scattered across platforms. Without transcribee: chaotic bookmarks, no searchability, repeated re-watching. With transcribee: every source becomes a searchable, LLM-queryable document in ~/Documents/transcripts/AI-Research/. You can literally ask your knowledge base: "Compare the chunking strategies mentioned by different speakers."

2. The Competitive Intelligence System

Your startup operates in fintech. Founders and PMs constantly share Instagram Reels and TikTok clips about competitor moves, market trends, regulatory shifts. These vanish into chat histories. transcribee automatically archives and categorizes them into Startups/Competitive-Intelligence/, creating a persistent, searchable competitive database that compounds in value.

3. The Self-Directed Learning Curriculum

You're learning Rust, system design, and MLOps simultaneously. Video tutorials from different creators, at different levels, in different formats. transcribee organizes these into Rust/, System-Design/, MLOps/ folders with full transcripts. Six months later, when you need to revisit that specific explanation of Tokio's runtime, you grep your transcripts instead of re-watching hours of video.

4. The Podcast Knowledge Mine

You consume 10+ hours of tech podcasts weekly. The insights are gold, but ephemeral. transcribee turns every episode into a structured document with speaker identification. That offhand comment about Postgres optimization from the maintainer in episode 347? Searchable forever. Citable in your team's technical discussions.

5. The Meeting Archive (Local Files)

Recorded team meetings, user interviews, conference calls — transcribee processes local MP4/MOV files with the same pipeline. Auto-categorized by project or topic, speaker-labeled for accountability, instantly queryable when someone asks "what did we decide about the API rate limits in March?"

Step-by-Step Installation & Setup Guide

Ready to stop losing knowledge? Here's your complete setup path.

Prerequisites

Ensure you're running macOS Sonoma or later with Node.js 18+ installed. You'll need Homebrew for dependency management.

Step 1: Install System Dependencies

# Install yt-dlp for video/audio downloading
brew install yt-dlp

# Install ffmpeg for media processing
brew install ffmpeg

yt-dlp is the actively maintained fork of youtube-dl, handling platform changes that break lesser tools. ffmpeg extracts audio from local video files with precision.

Step 2: Clone and Install transcribee

# Clone the repository
git clone https://github.com/itsfabioroma/transcribee.git

# Navigate and install Node dependencies
cd transcribee
pnpm install

The project uses pnpm for fast, disk-space-efficient package management. If you lack pnpm, install via npm install -g pnpm.

Step 3: Configure API Keys

# Copy the example environment file
cp .env.example .env

Edit .env with your actual credentials:

# Required: ElevenLabs API key for transcription
# Get yours at: https://elevenlabs.io/
ELEVENLABS_API_KEY=your_elevenlabs_key_here

# Required: Anthropic API key for auto-categorization
# Get yours at: https://anthropic.com/
ANTHROPIC_API_KEY=your_anthropic_key_here

Cost awareness: ElevenLabs' Scribe model is competitively priced for transcription; Anthropic's Claude API costs scale with transcript length for categorization. For heavy usage, monitor both dashboards.

Step 4: Create Shell Alias (Strongly Recommended)

Add to your ~/.zshrc for global accessibility:

# The noglob prefix prevents zsh from interpreting URL special characters
alias transcribee="noglob /path/to/transcribee/transcribe.sh"

Reload your shell: source ~/.zshrc

Step 5: Verify Installation

# Test with a short video
transcribee "https://youtube.com/watch?v=dQw4w9WgXcQ"

Check ~/Documents/transcripts/ for your first auto-categorized entry.

REAL Code Examples: Inside transcribee's Power

Let's examine actual implementation patterns from the repository, with detailed explanations of what happens under the hood.

Example 1: Basic Transcription Commands

The core interface is deliberately simple — one command, multiple source types:

# YouTube video: downloads audio via yt-dlp, transcribes, categorizes
transcribee "https://youtube.com/watch?v=..."

# Instagram Reel: handles platform-specific extraction
transcribee "https://instagram.com/reel/..."

# TikTok clip: resolves short URLs, extracts clean audio
transcribee "https://vt.tiktok.com/..."

# Local audio file: skips download, processes directly
transcribee ~/Downloads/podcast.mp3

# Local video file: extracts audio via ffmpeg first
transcribee ~/Videos/interview.mp4

What's happening: The transcribe.sh script detects source type by URL pattern or file extension. For URLs, it delegates to yt-dlp with audio-only extraction flags. For local files, it uses ffmpeg -i input -vn -acodec copy to strip video streams. This unified pipeline means you never think about format handling — the tool abstracts platform complexity into a single mental model: "point at content, receive knowledge."

Example 2: OpenClaw Skill Integration

For AI agent enthusiasts, transcribee integrates as an OpenClaw skill:

# Install from ClawHub (recommended method)
openclaw skills install transcribee

# Or manual installation for development/customization
git clone https://github.com/itsfabioroma/transcribee.git \
  ~/.openclaw/workspace/skills/transcribee

The power here: Your AI agent gains transcription capabilities. Instead of context-switching to transcribe content, you literally converse with your agent: "Transcribe this video: [URL]" — and it appears in your knowledge base. This is automation at the conversational layer, not just the command line.

Example 3: Raw JSON Output for Custom Pipelines

# Enable full ElevenLabs API response with --raw flag
transcribee --raw "https://youtube.com/watch?v=..."

This generates transcript-raw.json alongside standard outputs:

{
  "language_code": "en",
  "language_probability": 0.98,
  "words": [
    {
      "word": "retrieval",
      "start": 120.5,
      "end": 121.2,
      "speaker_id": "speaker_0",
      "confidence": 0.99
    }
    // ... word-level precision for every token
  ]
}

Why this matters: The raw output enables applications transcribee doesn't natively support. Build automated highlight reels by extracting high-confidence segments. Create precise citation links from your notes back to exact timestamps. Feed word-level data into custom NLP pipelines for entity extraction or sentiment analysis. The --raw flag transforms transcribee from an end-user tool into a knowledge infrastructure component.

Example 4: The Shell Alias Pattern

# Add to ~/.zshrc — critical for URL handling
alias transcribee="noglob /path/to/transcribee/transcribe.sh"

The noglob is not optional — it's essential. Without it, zsh interprets ? and & characters in URLs as glob patterns, causing mysterious failures. This single-line configuration eliminates an entire class of support issues and represents thoughtful shell UX design. The alias also means transcribee behaves like a native system command, fitting seamlessly into existing workflows.

Advanced Usage & Best Practices

Optimize Your Categorization Over Time The auto-categorization improves as your library grows. Early transcripts may land in generic folders; as patterns emerge, Claude's contextual understanding sharpens. Don't manually reorganize — let the system learn your taxonomy. If you need intervention, create seed folders in ~/Documents/transcripts/ with representative content.

Batch Processing with GNU Parallel For backlog conversion, pipe URLs through parallel execution:

cat video-urls.txt | parallel -j 2 transcribee {}

This respects API rate limits while maximizing throughput. Adjust -j based on your ElevenLabs tier.

Git-Version Your Knowledge Base Your ~/Documents/transcripts/ directory is plain text — ideal for Git tracking:

cd ~/Documents/transcripts
git init
git add .
git commit -m "Knowledge base: $(date +%Y-%m-%d)"

Suddenly you have temporal versioning, diffable evolution, and backup synchronization.

LLM Prompt Engineering on Transcripts The speaker-labeled format enables sophisticated queries. Instead of "summarize this," try: "What tensions exist between speaker_0's and speaker_1's positions on microservices?" or "Extract all actionable advice from speaker_2 and format as a checklist."

Comparison with Alternatives

The verdict: Choose transcribee if you want automated knowledge management from video sources, value open-source extensibility, and accept cloud API dependencies. Choose Whisper CLI if you need fully offline, privacy-critical processing and don't mind manual organization. Choose Otter/Descript if you need real-time collaboration features and don't mind subscription costs.

FAQ: What Developers Ask About transcribee

Is transcribee free to use? The software is open-source and free. You pay only for API usage: ElevenLabs for transcription and Anthropic for categorization. For moderate personal use, expect $5-15 monthly. Heavy users should monitor both dashboards.

Does it work on Windows or Linux? Currently macOS-only (tested on Sonoma). The core Node.js logic is portable, but the shell scripts and path assumptions are macOS-specific. Community ports are welcome — check the GitHub issues for progress.

How accurate is the speaker diarization? ElevenLabs' scribe_v1_experimental achieves strong results with clear audio and distinct speakers. Overlapping speech, heavy accents, or poor audio quality reduce accuracy. The --raw flag lets you inspect confidence scores per word.

Can I use my own categorization scheme? The AI categorization learns from your existing library structure. For custom schemes, pre-seed folders with representative transcripts. Direct manual override isn't implemented — the philosophy is automated organization, not manual filing.

What happens to my API data? Audio passes to ElevenLabs; transcript text passes to Anthropic for categorization. Both services have standard data handling policies. For sensitive content, consider Whisper CLI's fully local alternative despite organizational trade-offs.

How do I update transcribee? git pull in the repository directory, then pnpm install if dependencies changed. The tool is actively developed — watch releases on GitHub for new features.

Can I transcribe private/enterprise videos? Local file support handles any media you possess. For DRM-protected content, standard limitations apply — transcribee doesn't circumvent content protection.

Conclusion: Your Knowledge Deserves Better Than "Watch Later"

We've accepted a bizarre bargain: we spend thousands of hours consuming the richest knowledge medium ever created, then let it evaporate. Every forgotten insight is a competitive disadvantage, a repeated mistake, a missed connection.

transcribee rejects this status quo. It transforms video from ephemeral entertainment into structured, searchable, compoundable knowledge infrastructure. The self-organizing library isn't a nice-to-have — it's the difference between information consumption and genuine learning accumulation.

The setup takes fifteen minutes. The first transcript appears in seconds. Six months from now, you'll have a ~/Documents/transcripts/ directory that mirrors your intellectual evolution — queryable, LLM-ready, permanently yours.

Stop letting your brain be the bottleneck. Your future self will thank you when you can instantly surface that exact insight from a podcast you half-remember, or ask Claude to synthesize perspectives across twelve video sources you transcribed last quarter.

👉 Get transcribee on GitHub — star the repo, open an issue, or submit your first PR. The hive is growing. bzz bzz 🐝