yt-assemblyai-monitor

name

yt-assemblyai-monitor

description

YouTube channel monitor and video transcription using AssemblyAI cloud API. Pure Python + requests only — no ffmpeg, no Whisper, no extra tools needed. Monitors YouTube channels for new videos, extracts audio URLs via innertube API, submits to AssemblyAI for cloud transcription, and returns text + AI summary. Works on Mac, Linux, Windows. Only dependency: requests (usually pre-installed). Use when: user asks to monitor YouTube channels, transcribe YouTube videos, summarize video content, or set up YouTube content monitoring.

YouTube Channel Monitor (AssemblyAI)

Monitor YouTube channels and auto-transcribe new videos using AssemblyAI cloud API. Zero local dependencies beyond requests.

Prerequisites

1. AssemblyAI account: https://www.assemblyai.com/app/signup (free, 100 hours/month)
2. API Key: from Dashboard
3. requests library (usually pre-installed with OpenClaw/Python)

Setup API Key

Choose one:

# Option A: environment variable
export ASSEMBLYAI_API_KEY="your-key"

# Option B: config file
echo '{"api_key": "your-key"}' > data/config.json

How It Works

Channel page → innertube API → audio direct URL → AssemblyAI cloud → text + summary

No local audio download, no Whisper, no ffmpeg. AssemblyAI handles everything server-side.

Commands

# Transcribe a single video
python3 scripts/monitor.py now "https://www.youtube.com/watch?v=VIDEO_ID"

# Add a channel to monitor
python3 scripts/monitor.py add "https://www.youtube.com/@ChannelName/videos" "Alias"

# Check all channels for new videos + transcribe
python3 scripts/monitor.py check [count_per_channel]

# List channels
python3 scripts/monitor.py list

# Remove channel
python3 scripts/monitor.py remove "Alias"

Important Notes

• Audio URL extraction uses YouTube's innertube API (pure requests).
• Works on normal residential IPs. Datacenter IPs may get blocked by YouTube.
• If audio URL extraction fails with UNPLAYABLE, the IP may be flagged — this is a YouTube limitation, not a bug.
• Free tier: 100 hours/month of transcription.
• Transcription takes ~70-80% of video duration.
• Results saved to data/summaries/{video_id}.json.

Output Format

Each result includes:

• text: full transcription
• summary: AI-generated paragraph summary
• confidence: transcription confidence score
• utterances: timestamped segments with speaker labels

Files

Path	Purpose
data/channels.json	monitored channels
data/processed.json	already-transcribed video IDs
data/summaries/*.json	transcription results
data/config.json	API key (optional)