SentrySearch
A command-line tool for semantic search over video footage: describe what you are looking for in text or an image, and it returns a trimmed clip from the matching moment.
SentrySearch splits video files into overlapping chunks, embeds each chunk using Google's Gemini Embedding API, Alibaba DashScope, or a local Qwen3-VL model, and stores the resulting vectors in a local ChromaDB database. A search query, given as text or an image, is embedded into the same vector space and matched against the stored embeddings, and the closest match is automatically trimmed from the source video and saved as a clip.
It includes an MLX backend that runs the local model faster and with less memory on Apple Silicon, a dashcam metadata overlay for a specific vehicle brand, and companion tools for stitching or redacting footage. Most backends require an API key, except for the local and MLX backends, which need no API key at all.