wildlifechorus/condenseit v2.7.0

2.7.0

on GitHub

What's Changed

New feature: YouTube audio transcription via OpenRouter Whisper

Videos that lack YouTube captions can now be transcribed using OpenRouter's
Whisper API, producing dramatically better summaries than RSS descriptions.

How it works:

1. youtube-transcript-api tries to fetch captions (free, existing behaviour)
2. If no captions → yt-dlp downloads the audio track
3. Audio is sent to openrouter.ai/api/v1/audio/transcriptions (Whisper)
4. Falls back to the RSS description if everything fails

Key details:

• Disabled by default — enable in Admin > Digest under "YouTube transcription"
• Uses your existing OpenRouter API key — no new subscription needed
• Spend is tracked in Admin > Budget and respects daily/monthly limits
• Default model: openai/whisper-large-v3-turbo (fast, cheap); whisper-large-v3 available for highest accuracy
• Configurable max video duration (default 30 min) to control cost
• yt-dlp and ffmpeg added to Docker images and provision-ubuntu.sh

Bug fix: YouTube video publish dates

YouTube videos now use their actual feed publish date rather than now(), so the age filter correctly excludes old videos.

Upgrading

If deploying to an existing VPS (outside Docker), install the new dependencies once:

pip install yt-dlp
apt-get install -y ffmpeg

Commits Since 2.6.0

• 866607d feat(youtube): add remote Whisper transcription via OpenRouter
• 02074f9 fix(youtube): use feed publish date for age filter