Reel Notes
devttyac/reel-notes
Assista e resuma qualquer vídeo — baixa e transcreve com Claude
Ver no GitHubPlugin duplo que baixa vídeos de 1.000+ plataformas (YouTube, TikTok, Instagram, etc.) e os converte em notas estruturadas em markdown usando Claude. Divide o fluxo em dois plugins independentes: media-downloader (busca e download) e sumtube (resumo e estruturação).
Instalação
npx convez-ai devttyac/reel-notes1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Resumir palestras e vídeos educacionais em notas estruturadas
- Extrair conteúdo de redes sociais (TikTok, Instagram, YouTube Shorts) para arquivo pessoal
- Documentar reuniões gravadas ou apresentações em markdown organizado
- Processar vídeos de múltiplas plataformas sem interface web
Veredito

Arquitetura bem pensada com dois plugins desacoplados que reduz complexidade de dependências e permite reutilização independente. Ferramenta sólida para fluxo de captura e documentação de vídeos, mas requer ffmpeg + yt-dlp + chaves de API configuradas localmente.
README
A tradução para o português deste README ainda está na fila.
reel-notes
reel-notes — watch and summarise any video with Claude
Plugins
sumtube — summarises a video into one structured plain markdown note. Accepts a local video file path or a direct YouTube URL.
media-downloader — downloads a video from any yt-dlp-supported URL (YouTube, Instagram, TikTok, X, and 1,000+ other sites) to a local file, ready for sumtube to process.
Why two plugins instead of one?
reel-notes splits the workflow deliberately. Each plugin has one job, one dependency surface, and one failure mode:
| Concern | media-downloader | sumtube |
|---|---|---|
| Job | Fetch a video from a URL to local disk | Turn a video (URL or local file) into structured notes |
| External dependencies | yt-dlp, ffmpeg | Anthropic API, Groq API (optional), ffmpeg |
| API keys required | None | SUMTUBE_API_KEY (and optionally GROQ_API_KEY) |
| Risk surface | yt-dlp extractors break as sites change — updated frequently | LLM prompts and model versions evolve independently |
| Cost per run | Free (bandwidth only) | Paid (Anthropic tokens, Groq audio minutes) |
The split buys three concrete things:
- Independent reuse. Want a clean local copy of a TikTok or a Vimeo lecture without paying for a summary? Use media-downloader on its own. Already have a local meeting recording? Skip the downloader and feed it straight into sumtube. Neither plugin assumes the other is installed.
- Isolated failure modes. A yt-dlp extractor breaking (sites change weekly) doesn't disturb the summarisation path. A prompt or Anthropic SDK change doesn't risk the downloader. Diagnosing a failed run is easier when only one moving part can be at fault.
- Independent update cadence. yt-dlp ships near-daily for site coverage; the LLM stack moves on its own schedule. Splitting them means a security or extractor update on one side never blocks an unrelated change on the other, and users can update each plugin without re-validating the full pipeline.
A monolithic plugin would force every user to install both dependency stacks, accept both risk surfaces, and update both code paths even when only one is needed. The two-plugin layout keeps the surface area each user installs proportional to what they actually use.
Two-step workflow
Download, then summarise:
/media-downloader https://www.youtube.com/watch?v=dQw4w9WgXcQ
/sumtube /path/to/downloaded-video.mp4
That's it. The first command fetches the video; the second produces the summary notes.
Installation
Clone the repository, register the local marketplace, then install each plugin:
git clone https://github.com/devttyac/reel-notes.git
cd reel-notes
claude plugin marketplace add .
claude plugin install sumtube@reel-notes
claude plugin install media-downloader@reel-notes
Supported platforms
media-downloader uses yt-dlp under the hood, which supports 1,000+ sites including:
- YouTube (videos, shorts, playlists)
- Local video files passed directly to sumtube
- Non-YouTube URLs: Instagram, TikTok, X (Twitter), Vimeo, and any other yt-dlp-compatible source
Requirements
| Dependency | Required | Purpose |
|---|---|---|
| ffmpeg | Yes | Audio extraction and format conversion |
| yt-dlp | Yes | Video downloading (media-downloader) |
SUMTUBE_API_KEY or ANTHROPIC_API_KEY | Yes | Claude API access for sumtube summarisation. Use SUMTUBE_API_KEY under Claude Code (its sandbox overwrites ANTHROPIC_API_KEY). Either may also live in .env at the plugin root. |
GROQ_API_KEY | No | Enables Whisper transcription fallback for caption-less videos |
Install system dependencies on macOS:
brew install ffmpeg yt-dlp
Why Whisper matters
Captions are a fragile assumption. Whisper (via Groq) closes the gap:
- Most non-YouTube platforms don't expose captions at all. TikTok, X, Instagram, Vimeo, Facebook, Reddit — none surface a transcript through yt-dlp. Without Whisper, sumtube would silently fail on every one.
- A large fraction of YouTube content has no captions either. Newer uploads, shorter clips, non-English videos, and creator-disabled tracks all bypass YouTube's auto-caption pipeline.
- Local recordings (meetings, screen captures, podcasts) have no captions by definition. Whisper is the only path that lets sumtube handle them.
- Graceful degradation. Without
GROQ_API_KEY, captioned videos still work — the Whisper fallback is opt-in, not required. - Cost stays low. Groq's Whisper inference is sub-realtime and inexpensive; only the resulting transcript text reaches Anthropic, never raw audio.
In practice: Whisper is what turns sumtube from a YouTube-caption summariser into a general-purpose video summariser.
Security
No keys in this repository. Never commit API keys or secrets to source control.
Set your credentials as environment variables before running either plugin:
# Under Claude Code (recommended) — its sandbox blocks ANTHROPIC_API_KEY
export SUMTUBE_API_KEY=your_anthropic_api_key_here
# Or, in a standalone shell:
export ANTHROPIC_API_KEY=your_anthropic_api_key_here
export GROQ_API_KEY=your_groq_api_key_here # optional
Alternatively, copy plugins/sumtube/.env.example to plugins/sumtube/.env and fill in your keys (the .env file is gitignored).
Data flow:
- Video audio is sent to Groq only when
GROQ_API_KEYis set and the video has no embedded captions. Groq processes the audio for transcription via Whisper. - The transcript (or embedded captions, if available) is sent to Anthropic (Claude) for summarisation. No raw audio is sent to Anthropic.
- Nothing is stored remotely by reel-notes. Downloaded files stay on your local machine.
License
MIT