Turn podcasts into readable content. Automatically.
You subscribe to 20 podcasts. Each episode is 60-90 minutes. That's 30+ hours of content per week. You'll never catch up by listening.
thestill downloads your podcasts, transcribes them with speaker identification, cleans up the messy speech-to-text output, and generates summaries. You get a morning briefing of everything that dropped overnight.
# Add some podcasts
thestill add "https://lexfridman.com/feed/podcast/"
thestill add "https://www.youtube.com/@hubermanlab"
# Generate your morning briefing
thestill digest
# Or run the web UI
thestill serverThat's it. New episodes are discovered, downloaded, transcribed, cleaned, and summarized. You read instead of listen.
Even better - talk to your podcasts through Claude Desktop:
- "Generate 3 LinkedIn posts from the latest Lenny's Podcast episode"
- "What themes keep coming up in the last 10 Prof G episodes?"
- "Find every mention of 'product-market fit' across all my podcasts"
Your podcasts become a searchable, queryable knowledge base.
Six atomic steps, mix and match as you like:
Discover → Download → Downsample → Transcribe → Clean → Summarize
Each step is independent. Failed at transcription? Fix it and continue - no need to re-download. Want cloud transcription but local LLM? Go for it.
Transcription - Pick your engine:
- Whisper (local, free, private)
- Google Cloud Speech-to-Text (fast, accurate)
- ElevenLabs Scribe (great quality)
- Dalston (self-hosted)
LLM Processing - Pick your brain:
- OpenAI GPT
- Anthropic Claude
- Google Gemini
- Mistral
- Ollama (local, free)
Sources - Works with:
- RSS feeds
- Apple Podcasts
- YouTube channels & playlists
# Install
git clone https://github.com/ssarunic/thestill.git
cd thestill
pip install -e .
# Configure (edit .env with your API keys)
cp .env.example .env
# Add a podcast and process it
thestill add "https://example.com/podcast/rss"
thestill digest- Web UI - React dashboard for managing podcasts, viewing transcripts, monitoring queue
- Speaker Diarization - Know who said what in multi-person conversations
- Morning Briefing - Daily digest of new episodes with summaries
- MCP Server - Natural language access to your podcast library via Claude Desktop
- Multi-user Auth - Google OAuth for hosted deployments
- Failure Handling - Automatic retries, dead letter queue for manual review
| Guide | Description |
|---|---|
| Configuration | Environment variables and settings |
| Transcription Providers | Setup guides for each provider |
| Web Server | API endpoints and webhooks |
| MCP Usage | Claude Desktop integration |
| Transcript Cleaning | LLM-based cleanup options |
| Logging | Structured logging setup |
# Podcast management
thestill add <url> # Add podcast (RSS, Apple, YouTube)
thestill list # List podcasts
thestill remove <id> # Remove podcast
thestill status # System stats
# Processing pipeline
thestill refresh # Discover new episodes
thestill download # Download audio
thestill downsample # Convert to 16kHz WAV
thestill transcribe # Transcribe to JSON
thestill clean-transcript # Clean with LLM
thestill summarize # Generate summaries
# Batch processing
thestill digest # Full pipeline + morning briefing
thestill digest --dry-run # Preview what would be processed
# Web server
thestill server # Start on localhost:8000All commands support --podcast-id, --max-episodes, and --dry-run.
data/
├── original_audio/ # Downloaded MP3/M4A files
├── downsampled_audio/ # 16kHz WAV for transcription
├── raw_transcripts/ # JSON with timestamps & speakers
├── clean_transcripts/ # Cleaned Markdown
├── summaries/ # Episode analysis
├── digests/ # Morning briefings
└── podcasts.db # SQLite database
pip install -e ".[dev]" # Install dev dependencies
make test # Run tests
make check # Lint + typecheck + testSee CLAUDE.md for architecture details and docs/code-guidelines.md for contribution guidelines.
Apache 2.0