My Home Assistant install has about 1,300 entities, and on a typical day about 150 of them are unavailable. Most of those are harmless, like a laptop that’s asleep or leftovers from something I removed. The few that need fixing get lost in the pile. An integration that fails to start is worse, because it only shows an error in the integrations list, so I could miss it for weeks.
So I built Gut Check. It looks over your install once a week and puts anything that needs you in Settings > Repairs, where you act on it or ignore it.
What it checks
- Home health check: Sorts unavailable entities into expected, worth fixing, and safe to remove. Only the worth-fixing ones get a Repairs card, and the card clears itself when the entity comes back.
- Update review: Reads the release notes on pending updates and flags the ones that might break something (renamed or removed features, manual steps). It never installs anything.
- Area suggestions: Suggests one of your existing areas for each device that has none. You confirm or decline each card.
- Device class suggestions: Suggests a device class for sensors that report a unit but have none, so they get the right icon, unit conversion, and a spot on the Energy dashboard.
- Stuck integration check: Sorts integrations that failed to start into passing glitch, sign-in problem, and broken for good.
- Critical label suggestions: Finds smoke, CO, gas and leak sensors and shutoff valves you haven’t marked critical yet.
- Diagnostic sensor suggestions: Suggests hiding sensors about the device itself (Wi-Fi signal, SIM carrier, phone storage) that clutter auto-generated dashboards.
- Recorder suggestions: Counts which entities write to the recorder most and suggests which ones you could exclude, with a snippet to paste into configuration.yaml yourself.
1-5 are on by default. The other 3 can be enabled in the integration’s Configure flow.
How it decides
The judgment calls go to Jev, a decision model from TypeSafe AI. Gut Check sends a short description of each item and a fixed list of answers, and Jev picks one and says how likely it is to be right. It can’t write free text, invent a new option, or tell Home Assistant to do anything. When it isn’t confident, Gut Check marks the item unsure and leaves it alone.
Prereqs
A free TypeSafe AI API key is required. There’s no add-on or local model to run. On my install, one run of every check costs about a cent. You set a daily token budget in Gut Check (500,000 tokens is the default – plenty of overhead), and Gut Check refuses any run that would go over it. Gut Check exposes 2 sensors that show tokens used and cost today.
In building this integration (running the checks multiple times, using about 3.2MM tokens total), I’ve used $0.114 of the $5.00 free balance I started with when I created my Jev account. For a typical user running every check weekly on an install like mine (~1,300 entities) cost $0.009 each run (9/10 of a cent). 
Install
Add https://github.com/funkadelic/ha-gutcheck to HACS as a custom repository (category: Integration), install it, then add Gut Check from Settings > Devices & services and paste your TypeSafe AI API key.
If it sorts something wrong on your install, please open a GitHub issue and I can take a look! Loved to hear any feedback on how it worked out for your install.