SentrySearch enables semantic search for raw MP4 video
A Python CLI that leverages Gemini Embedding 2 to natively map raw video directly into vector space for instant retrieval. SentrySearch bypasses frame-by-frame captioning to enable natural language queries like "red truck cutting me off" across hours of footage.
SentrySearch is a textbook example of "pixels-to-vectors" search, proving that intermediate text descriptions are becoming obsolete for video retrieval. Gemini Embedding 2's ability to project video and text into the same 768D vector space removes the heavy compute overhead of OCR or frame-captioning, while local indexing via ChromaDB makes it practical for privacy-sensitive footage. Automated clipping with ffmpeg provides a seamless bridge between text queries and relevant segments, though the $2.50 per hour indexing cost remains a friction point for massive archives.
DISCOVERED
18d ago
2026-03-25
PUBLISHED
18d ago
2026-03-25
RELEVANCE
AUTHOR
Github Awesome