A local-first application for semantic video search and discovery. Index videos via YouTube URL or local upload and perform natural language queries to find, play, and download specific moments.

• Semantic Search: Query video content using natural language (e.g., "Where is the battery replaced?").
• Local Vector Storage: Uses ChromaDB for local indexing and storage.
• Hardware Acceleration: Optimized with hardware-accelerated transcoding (h264_videotoolbox) for near-instant clipping.
• Progress Tracking: Real-time feedback during the multi-stage indexing process.
• Segment Downloads: Export any discovered video segment directly to your disk.
• Project-Based Library: Each ingestion creates its own project index so previously indexed videos are preserved.

1. Ingest a video from YouTube (/process) or local upload (/upload).
2. Backend creates a new project_id and analyzes the video into semantic segments.
3. Segment descriptions are embedded and indexed in a project-scoped ChromaDB collection.
4. Queries (/query) search only within the selected project using semantic similarity.
5. Matching time ranges can be clipped and downloaded via /clip.

• Python 3.10+
• FFmpeg
• uv (Recommended for dependency management)
• Node.js & npm
• Google AI API Key (Set in backend/.env as GOOGLE_API_KEY)

The project includes a Makefile for simplified orchestration.

This will configure the Python environment and install frontend dependencies.

make setup

Create a .env file in the backend/ directory:

echo "GOOGLE_API_KEY=your_key_here" > backend/.env

Launches both the Backend and Frontend in a single terminal.

make dev

Press Ctrl+C to stop both servers.

• POST /process: Index a YouTube video via URL.
• POST /upload: Index a local video file.
• POST /query: Search within a project index (requires project_id and query).
• POST /clip: Generate a segment from a project video (requires project_id, start_time, end_time).
• GET /projects: List indexed projects.
• GET /projects/{project_id}: Get project metadata.
• DELETE /projects/{project_id}: Delete a project and its indexed data.

• AI: Gemini (Analysis), sentence-transformers (Local Embeddings).
• Database: ChromaDB.
• Media: FFmpeg, yt-dlp.
• Frontend: React, Tailwind CSS.
• Backend: Flask.

MIT