Lab Notes

Frigate + Local AI Stack

Experiments with local-first surveillance infrastructure, MQTT events, and lightweight AI workflows.

Frigate + Local AI Stack

Related projects: Home Automation, Cameras & Local Monitoring and MQTT Learning & Packaging Automation Concepts

This project started as a basic self-hosted camera setup and gradually turned into a much larger infrastructure sandbox.

The overall goal is keeping everything local-first and self-hosted while experimenting with:

  • event-driven automation
  • lightweight AI workflows
  • WebRTC streaming
  • MQTT messaging
  • hardware acceleration
  • browser-based camera systems
  • remote access
  • Linux infrastructure

The stack changes constantly depending on what is being tested.

Current experiments involve:

  • Frigate
  • go2rtc
  • Docker
  • VAAPI acceleration
  • MQTT notifications
  • browser-based camera dashboards
  • remote networking through Tailscale

One thing that became obvious very quickly is how much browser behavior impacts real-time video systems.

A lot of debugging time has gone into:

  • stream stability
  • buffering behavior
  • WebRTC negotiation
  • hardware decode paths
  • browser compatibility
  • low-latency streaming

There is also ongoing experimentation around event systems and machine communication.

MQTT has become especially useful for tying together:

  • motion events
  • notifications
  • automation triggers
  • lightweight device communication
  • status reporting

The overall environment continues evolving into a general local AI and infrastructure experimentation platform.