Podcast Transcript Fetcher

Fetch transcripts from 5 supported podcasts. Tier 2 (RSS+Groq Whisper) is the recommended approach -- fast, free, and the most reliable across all podcasts. Tier 1 free sources are best-effort (limited availability). Tier 3 Taddy API is the premium/commercial option.

Quick Reference

# Get latest episode transcript (auto-detects best method)
python scripts/get_transcript.py "Lenny's Podcast" --latest

# Search by episode title or number
python scripts/get_transcript.py 20vc --episode "Marc Andreessen"
python scripts/get_transcript.py dwarkesh --episode 15

# Force specific method
python scripts/get_transcript.py "cheeky pint" --latest --method whisper
python scripts/get_transcript.py a16z --latest --method taddy

# Save to file
python scripts/get_transcript.py lennys --latest --output transcript.md

# List all supported podcasts
python scripts/get_transcript.py --list-podcasts

Supported Podcasts

Implementation

1. Install Dependencies

# Core (always required)
pip install requests

# Cloud transcription (recommended — fast, free tier)
pip install groq
export GROQ_API_KEY="your-key"  # Get at https://console.groq.com

# Local transcription (free, needs ~5GB RAM)
pip install faster-whisper

# Audio compression (for Groq's 25 MB limit — Windows: winget/scoop)
#   winget install ffmpeg  or  scoop install ffmpeg

# Taddy API (commercial, optional)
export TADDY_API_KEY="your-key"  # Get at https://taddy.org

2. Get a Transcript

The script auto-selects the best method. Tier 2 is the default recommendation:

Tier 1 → Tier 2 (RECOMMENDED) → Tier 3
(best-effort)  (Whisper)  (Taddy API premium)

Tier 1: Free direct sources (best-effort, limited availability)

• Lenny's: Clones ChatPRD/lennys-podcast-transcripts and searches by title
• Dwarkesh: Substack PDF scrape
• 20VC: Substack PDF scrape
• A16z: Website scrape
• Cheeky Pint: No Tier 1 sources available
• Note: Tier 1 sources are best-effort and limited. Tier 2 (RSS+Whisper) is the recommended approach.

Tier 2: RSS + Whisper transcription [RECOMMENDED]

• Downloads MP3 from podcast RSS feed
• Compresses if >25 MB (ffmpeg)
• Transcribes via Groq Whisper API (free tier, ~10s per hour of audio)
• Fast, free, and works for every podcast in the registry
• Default recommendation for all use cases

Tier 3: Taddy API (commercial/premium)

• Requires TADDYAPIKEY ($75/mo+)
• Use for large-scale or production transcript needs
• Covers all 5 podcasts with auto-transcription

3. Analyze with AI

Once you have a transcript, pipe it to the agent for analysis:

I have this transcript from [podcast]. Can you:
1. Summarize the key arguments
2. Extract 3 actionable insights
3. Identify any controversial claims
4. Compare with [other podcast] on the same topic

Supported Workflows

Single Episode

Cross-Podcast Search & Batch

Output Structure

Batch transcription saves to output/ with per-podcast subdirectories:

output/dwarkesh-podcast/Dwarkesh Podcast_2024-01-15_agi-is-still-30-years-away.md
output/20vc/20 Minutes VC (20VC)_2024-03-10_funding-round-analysis.md

Each file includes a YAML frontmatter header:

---
podcast: Dwarkesh Podcast
episode: AGI is still 30 years away
date: 2024-01-15
url: https://...
source: whisper
---

Podcast Registry

The registry at scripts/podcasts.json maps each podcast to its RSS feeds, transcript sources, and API endpoints. To add new podcasts:

{
  "id": "new-podcast",
  "name": "New Podcast",
  "rss": "https://example.com/feed.xml",
  "transcript_sources": {
    "primary": {"type": "website_scrape", "url": "https://example.com"}
  }
}

Troubleshooting

RSS Feed Status (as of 2026-06)

Common Mistakes

• Forgetting API keys: Set GROQAPIKEY in your env or .env file
• Relying on Tier 1 free sources: Tier 1 is best-effort and limited. Always fall back to Tier 2 (RSS+Whisper) which is the recommended method.
• Not cloning the Lenny's repo first: The GitHub archive must be cloned locally for Tier 1 to work
• Using --method taddy without TADDYAPIKEY: Falls through silently; set the key or use auto mode