Unblink

// Overview

Unblink is an open-source, AI-powered camera monitoring application from Zapdos Labs. It watches your camera feeds, sends sampled frames to a vision-language model (the README names Qwen3-VL) for analysis and summarization, and lets you work with the result in plain language: a chat interface for asking about what the cameras see, and a search over recorded frames using natural-language queries.

It is built around a relay-node architecture. A small unblink-node program runs inside your own network, discovers and forwards traffic from local cameras that speak RTSP or MJPEG, and connects out to a publicly reachable server that manages nodes, handles authentication and stores configuration. That split means cameras on a private network can be viewed and analyzed without opening ports to them; on first run the node prints a URL you open in a browser to authorize it.

The server is written in Go with PostgreSQL for data and go2rtc for RTSP and WebRTC handling, and the web frontend uses SolidJS. The chat, fast and vision models are each configured as OpenAI-compatible endpoints, so the models behind the analysis can be swapped through environment variables. The repository includes a Dockerfile, a docker-compose file and a Render deploy button for running your own server, and prebuilt node binaries for Linux and Windows on x86_64 and ARM64.

// What it does

• Frame analysis and summarization of camera feeds with a vision-language model
• Chat interface for asking questions about your camera feeds in natural language
• Natural-language search over recorded frames
• Relay-node design: a local node forwards RTSP and MJPEG cameras from a private network to the server without exposing them directly
• Chat, fast and vision models configured separately as OpenAI-compatible endpoints
• Prebuilt unblink-node binaries for Linux and Windows (x86_64 and ARM64), or install with the Go toolchain

// Getting started

The README's quick start is about the node, the piece that runs on your network next to the cameras and connects to the relay server. You can download a prebuilt binary or install it with Go.

go install github.com/zapdos-labs/unblink/cmd/unblink-node@main

unblink-node

cp .env.example .env
make install
make dev

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Get plain-language summaries of what home, office or shop cameras have seen instead of scrubbing through footage
• Search recorded frames with a query like a description of a person, vehicle or event
• Monitor RTSP or MJPEG cameras on a private network from a web app without port-forwarding each camera
• Experiment with different vision-language models for video monitoring by pointing Unblink at your own OpenAI-compatible endpoint

// How Unblink compares

Unblink alongside other open-source physical sensing & perception tools AI/TLDR tracks, ranked by GitHub stars.