cvkit
Open-source camera AI

Cameras that tell you what matters.

cvkit is a multi-camera streaming and AI toolkit. Point it at security cameras and have it tell you about the things you care about: a car or a person you don't recognise, an animal, smoke.

View on GitHub How it works

How it works

01 · Edge

Watch locally

A lightweight agent on your network ingests the camera feeds and detects motion, kept thin so nothing slows the ingest path.

02 · Gate

Filter the noise

Off the ingest path, candidate events are checked and tuned so a moth on the lens doesn't become an alert.

03 · Understand

Look closer

Vision models take a richer pass over the clip to work out what is actually there.

04 · Alert

Tell you

Set an alert like "cars I don't recognise", whitelist your own, and get a description, a still and the video.

Components

edge
The on-network agent: camera ingest and motion detection.
autotune
Post-processing and parameter tuning for the edge detector.
understand
Detailed analysis of images and video.
worker
The frame-processing chain the detectors compose into.
video-synthesizer
Synthetic test footage, so tests don't need real cameras.
go-media-repackager
TS to fMP4 repackaging in Go.