Watching the wild, quietly, together.
WildWatch is a network of public wildlife camera livestreams, monitored by AI in real-time. Every camera streams live, free, to anyone with an internet connection — and every animal that walks into frame is automatically catalogued.
How a detection becomes a data point.
Frame extraction
The agent uses yt-dlp to resolve livestream URLs and ffmpeg to grab a single frame. URLs are cached for 3 hours to avoid rate limits. Frames from multiple streams are extracted in parallel.
AI classification
Frames are sent to Google's SpeciesNet model — an ensemble of MegaDetector v5 (object detection) and EfficientNet V2 (species classification). Inference runs on a local GPU, grouped by country code for geographic filtering.
Smart upload
Results are uploaded to Supabase only when the detection changes — consecutive blank frames are skipped. Old detections are pruned to keep the 5 most recent per stream, with storage cleanup.
Real-time dashboard
The Next.js dashboard subscribes to Supabase Realtime. When a new detection arrives, the grid updates instantly — no polling, no refresh needed.