DogBot is an AI system that detects dogs from camera feeds, classifies their breeds, and tracks individual identities over time.

This is a live deployment running on home security cameras in Los Angeles. Create an account to explore the dogs that have been spotted, or subscribe to receive a daily photo.

Every dog
deserves to
be seen.

DogBot combines object detection, breed classification, and visual embeddings to identify dogs, determine their breed from 120+ possibilities, and recognize returning visitors across multiple sightings.

The entire system runs locally on a Mac with Apple Silicon—no cloud services, no subscriptions, no data leaving your network. Deploy it on your own cameras to build a catalog of your neighborhood dogs.

Start Exploring →

How It Works

Four AI models work together in a pipeline, each handling a specialized task.

01 Detection

Find the Dog

YOLOv5 / Scrypted

Motion triggers detection. The system identifies that a dog is present in the frame.

02 Extraction

Crop & Filter

OpenCV + PIL

The dog is cropped, quality-checked for blur and size, then saved for processing.

03 Classification

Identify Breed

Vision Transformer

A fine-tuned ViT model classifies the breed with confidence scoring.

04 Recognition

Track Identity

DINOv2 Embeddings

Visual embeddings match dogs across visits, building individual profiles.