Practical overview
When it fits
Use it when you need timestamp-level discovery across a video collection and can run the indexing and search stack yourself.
Who it is for
- People who want to search a self-managed library of video by describing a moment in natural language.
- Developers or operators comfortable running a multi-service Docker application.
Documented uses
- Build a searchable index of local videos or videos fetched from yt-dlp-compatible URLs.
- Find relevant moments using speech and visual content, then jump to them in the built-in player.
- Filter searches to one indexed video or to audio versus visual matches.
Requirements & limits
The documented setup runs locally with Docker Compose and serves the web interface at http://localhost:8080. The project is licensed under GPL-3.0.
- The system indexes audio and visual content, and labels returned matches by modality.
- The documented stack uses Docker Compose, Qdrant, Whisper, FFmpeg, ONNX Runtime, and MobileCLIP2-S3-OpenCLIP.
- CPU setup is documented; the separate GPU configuration requires NVIDIA Container Toolkit.
- The repository describes local video uploads and yt-dlp-compatible URL ingestion.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Set up Sift Video and index a local file
Start the self-hosted service and add a local video to its searchable library.
Before you start
- Git
- Docker and Docker Compose
- A local video file
Steps
Clone the repository with its submodules and create the videos and output directories.
Start the CPU services with Docker Compose.
Open the local web interface at http://localhost:8080 and add a video through the upload panel, or copy a file into the videos directory.
Use the search panel to submit a natural-language query after the video has been indexed.
Expected outcome
Search results show relevant timestamps and audio, visual, or combined match types; selecting a result jumps the built-in player to that moment.
Guide limitations
- The documented NVIDIA GPU configuration requires NVIDIA Container Toolkit on the host.
- Files added to the videos directory are detected by a polling process described as running every 30 seconds.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 16 October 2026
- Original website title
- sourav4243/sift-video · Official homepage title · retained in its source language