Article

One RPi CM5 for Sensing, Automation, Analytics & Media at Home

May 27, 2025

The Home Assistant dashboard running on the Raspberry Pi CM5
The Home Assistant dashboard

The computer that runs my house is smaller than a credit card. A Raspberry Pi Compute Module 5 sits quietly at the centre of my Home Assistant setup, sipping power and working around the clock, and it has yet to feel stretched.

Its quad-core ARM Cortex-A76 and up to 8 GB of LPDDR5 ECC memory keep everything responsive, even with dozens of sensors and automations running at once. Every room has temperature and humidity sensors keeping comfort in check; motion sensors look after lighting and security; and the existing security cameras have been given something close to judgement.

That judgement comes from an LLM vision integration. The cameras no longer just see movement; they describe it. They can tell one vehicle type, make and model from another, and they can describe the people who come to the door.

A sample notification generated by the LLM vision integration describing a camera event
Sample response from LLM vision
Instead of “motion detected”, the house now says: “Person in blue jacket approaching with a package.”

I know which car has pulled up and what colour it is. When a delivery arrives, a snapshot and a short description are on my phone before the doorbell rings. And because the little module still has headroom, it also runs Plex Media Server, streaming and transcoding films around the house without breaking a sweat.

The garage is where things get more interesting. An mmWave radar sensor there detects and analyses human movement with remarkable precision, and it has quietly become my workout log, recording movement patterns and how long each session lasts.

All of this produces a lot of data, so it lands in a MariaDB database: exercise sessions, temperature readings, security events and automation logs, all in one place. Home Assistant’s built-in Jupyter notebook integration turns that archive into something useful. With Python and the usual data-science libraries I can chart how temperatures drift through the day, how my exercise habits change over months, and where the house spends its energy.

Air quality gets the same attention. VOC sensors pick up the volatile organic compounds released by cleaning products, paint and furniture, while NOx sensors watch for pollution drifting in from outside, especially at busy traffic times. The CM5 processes those readings in real time and acts as soon as a threshold is crossed.

Next on the list is the weather: a system that studies wind patterns and predicts when a storm is likely to take out the power. Not bad for a computer you could lose in a drawer.