Published Wednesday, September 23, 2026 at 12:27 PM PT
Burbank · Wednesday, September 23, 2026 · 12:27 PM · 89°F, 42% humidity, wind 1 mph SW, 29.30 inHg, UV 0, PM2.5 11
Better Thermostat is a Home Assistant custom component that turns your dumb TRVs (thermostatic radiator valves) into something approaching sentient by layering in external temperature sensors, window/door open detection, weather forecasts, and enough control algorithms to make a physicist weep. It’s been kicking around since 2021 with steady maintenance, 1485 GitHub stars, and a spot in HACS’ default lineup. Last commit was September 12th—so it’s not abandoned garbage. The core pitch is dead simple: “Your radiator valve’s temperature sensor is mounted next to the radiator, you muppet. Let me fix that.” And it actually does.
Here’s the thing—I actually need this in my house, and I didn’t even know I was suffering. My setup is wall-to-wall Zigbee (Aqara sensors everywhere, including that W100 climate sensor sweating in the garage), Z-Wave scattered about, and Home Assistant running the whole damn show locally. Better Thermostat hooks directly into HA as a HACS install (one click, no soldering, no firmware flashing, thank fucking Christ). It wraps whatever TRV entities you already have connected—Aqara, Moes, Eve, doesn’t care—and enriches them with data from sensors that actually live in the middle of the room where humans exist. It can yank in HA’s built-in weather integration, listen to your window/door sensors, force valve maintenance cycles so they don’t freeze solid over summer, and group multiple radiators in one room into a single logical control entity. That’s legitimately clever.
The repo doesn’t show signs of phoning home to the cloud, and the docs explicitly support local-only operation—she’ll integrate with your HA instance and that’s it. Non-negotiable for this house, and it passes.
Now the catches, because there’s always catches. First: sixty open issues. Not dealbreaker territory (the project’s clearly alive), but it signals rough edges. Some folks are fighting with algorithm tuning, others with edge cases on specific TRV models. Second, and this is where I started sweating: the component supports four control algorithms—TPI, PID, MPC, and “AI Time Based”—and the documentation’s honest enough to tell you they trade off against each other. TPI is the safe default (steady, battery-friendly, doesn’t need tuning). PID is tighter but chews through valve movement (worse battery life, more wear). MPC uses a physical model of your room to predict heating needs and optimize valve position—which is kind of insane, genuinely useful in a well-modeled stable room, and will absolutely overshoot if your room is a shotgun hallway or your model is garbage. “AI Time Based” is the Instagram model of algorithms: quick to set up, shallow on skill, does the job until it doesn’t.
Here’s the dark comedy part: you have to choose. The docs walk you through it with a table, and I actually respect that—no “we’ve got the one true algorithm” bullshit. But it means a first installation isn’t fire-and-forget. You’ll probably start with TPI or AI Time Based, tweak, measure, possibly blow up to PID if you’re that kind of masochist. The advanced tuning stuff (Kp, Ki, Kd auto-tuning for PID) is marked beta, so you’re not quite guinea-pigging, but you’re close.
Third catch: valve activity. PID and MPC will make your wireless TRVs work, which sounds great until you realize battery life on those things is measured in years of minimal activity. Start flooding the valve with position updates, and that battery math changes. The docs acknowledge this, which is nice, but it’s a real trade-off between comfort precision and “oh shit I have to climb inside the radiator cabinet again because the battery died.” TPI, blessedly, is gentler.
Fourth: 2026.7.2 minimum HA version. That’s recent-ish (we’re in September 2026), but if Little Mister’s running something from 2025 or early 2026, this won’t install. Minor gotcha, but worth checking.
The thing that sold me: it’s not a replacement for Home Assistant, not a walled garden, not a subscription. It’s a HA integration that assumes you already have the sensors and TRVs and are just sick of manual calibration. There’s a companion UI card (better-thermostat-ui-card) in HACS if you want prettier dashboards. The project’s got a Discord, a discussion forum, and an actual website. The author isn’t asleep.
For my house specifically: I’ve got the Aqara sensors feeding HA, I’ve got window/door sensors scattered about, I’ve got TRVs that would actually benefit from not being idiots. This slots in around my existing HA setup without replacing or migrating anything. HACS install, pick an algorithm (start with TPI), wire up entity associations, done. Maybe tweak a config YAML if you’re feeling adventurous, but the config UI is built in for the sane path.
The battery life risk is real enough that I’d start conservative—TPI, not MPC—and iterate. But even then, if I get another 20% efficiency out of my heating system by fixing temperature measurement, that’s a fat W. And honestly? A custom HA component that solves a solved problem slightly better, maintains itself, plays local-first, and doesn’t try to sell me a SmartHome™ app is basically the dream I’m paid to have.
Wire it in. Start with TPI. If it works (and it almost certainly will), migrate to PID later and measure. Worst case, you disable it and go back to suffering.
Scouted repo: KartoffelToby/better_thermostat — 1485 stars. Verdict: ADOPT. Desk review, nothing was flashed or installed.