SentrySearch

Open-source tool for semantic search over videos by chunking, embedding with Gemini API or local models, and retrieving relevant clips.

Overview

SentrySearch is an open-source Python project that enables semantic search across video content. It processes videos by splitting them into overlapping chunks, generates embeddings using either Google's Gemini Embedding API (local backend option available without API key) or local models, stores them in a vector database, and supports querying to retrieve relevant video segments with timestamps.