Smart thermostats like Nest and Ecobee learn your habits — but they can’t handle a changed routine. If the kids are suddenly home from school, you can’t just tell your thermostat to figure it out. So How-To Geek’s Adam Davidson asked: could a local LLM do the job better? He handed his heating over to AI for a week to find out.
Smart thermostat experiment (Credits: How-To Geek)
Local LLM Thermostat Experiment: The Setup
Davidson runs a Tado smart heating system — a boiler module feeding radiators, with smart thermostats on individual rooms — all wired into Home Assistant. His plan: give a local LLM a detailed brief (home layout, room warm-up rates from past data, weekday and weekend routines), then let it run every 30 minutes.
Each run, the model checked room temperatures and humidity, phone geolocation for who’s home, and motion, presence, and Bluetooth sensors to see which rooms were occupied. It then decided which rooms needed heat, when, and how warm — a script translated those decisions into commands for the Tado system. Before going live, he spent a weekend shadow-testing so the model “said what it wanted” without touching the real heating.
Choosing the Model: Qwen 3.5 9B Won, Thinking Models Were Too Slow
Running on an old Dell OptiPlex with 16GB of RAM via Ollama, most small models flunked. Granite 4 7B and Nemotron 3 Nano 4B got fewer than half the test decisions right, and models with thinking enabled took over 20 minutes per decision — useless for a 30-minute control loop. The winner was Qwen 3.5 9B, which got 74% of decisions right in 2–5 minutes.
Local LLM Thermostat vs Nest and Ecobee: The Verdict
The moment of truth came on Thursday, when Davidson tested the feature he wanted most: he typed that the kids were off school and would be in either a bedroom (where the Nintendo Switch is) or the living room. The LLM “factored this into its reasoning but then did exactly the opposite of what it should have done” — keeping the other bedroom warm instead.
The conclusion, per Davidson: the models he can run on his modest hardware “just weren’t capable of making the required decisions accurately.” His smart thermostat, by contrast, “did a far” better job. A larger model on better hardware might genuinely work — and could even cut heating costs — but his local-LLM version “definitely isn’t” there yet.
Also Read: Wi-Fi Keeps Dropping for No Reason? Military Radar Could Be the Culprit — Here’s How to Fix It | That $5 ESP32 Board Isn’t $5: The Hidden Costs Every DIY Project Needs
Source: How-To Geek — read the full story there.
