I personally look at the issue from a different point of view.
A human keeping a detailed household inventory current is hard. It will ALWAYS BE hard. I have absolutely no desire to scan every battery, update every drawer, or tell a database whenever I move a screwdriver. As someone said above, that system will remain beautifully accurate for approximately two weeks.
So I don’t want the human keeping track of it at all frankly, the agent should.
You know, the one you’re all going to have running in your home in less than five years. (we’re at the beginning of an IBM PC type event; I’m fairly convinced this is about to become a thing as price and capability converges…)
This is also why I firmly believe that agent belongs INSIDE Home Assistant.
You are constantly generating signals you don’t even realize are signals. Power consumption, temperature, flow, occupancy, room activity, schedules, purchases, completed chores, camera observations, device state, conversations, and a hundred other tiny events individually look like noise.
Over time, they form patterns your system can use to keep track of that sh- stuff for you.
A real example from my system.
ZenOS has a physical Plant manager. It can setup practically anything in the house that is a thing with a process as a managed machine: an HVAC compressor, air handler, water heater, freezer, dishwasher, washing machine, garage door, sump pump, water softener, vacuum, or hot tub.
Each machine can have two connected views.
The first is operational. Home Assistant can see that the machine is running, drawing power, changing temperature, moving water, completing a cycle, or behaving differently from normal. (This is through binding the ‘machine’ together through labels and such…
The second is physical. The agent knows what the machine is, where it lives, which parts and consumables belong to it, what stock is available, which chores are due, and which documents or service records relate to it. This is care of Grocy.
Consider an HVAC system. The thermostat provides room temperature, mode, setpoint, and demand. The physical model contains the compressor, air handler, filters, their locations, replacement stock, and maintenance chores.
Those pieces become one managed thing from the agent’s point of view.
With a single unified tool view, the agent can see that runtime has increased, connect that with weather and room behavior, know which filter the air handler uses, determine whether one should be in stock, check when it was last changed, find the manual, and create or surface the appropriate work.
YOU don’t need to remember to “update the inventory database” as a separate hobby. The inventory just becomes part of HA automating the machine.
That is the whole ball o wax.
A useful household model emerges from signals the home is already producing. The agent maintains the relationships, remembers the history, notices the patterns, and asks a human only when uncertainty actually matters.
It doesn’t need omniscient, centimeter-perfect knowledge of every object. It needs enough useful truth to answer practical questions:
-
What do we probably have?
-
Where does it normally live?
-
Which machine or activity is it associated with?
-
When was it last seen, used, purchased, or serviced?
-
How confident is the answer?
-
Does the uncertainty ACTUALLY matter right now? (Usually no)
If I ask for a furnace filter, the agent should be able to say, “There should be one on the garage shelving. You bought a two-pack in March and changed one in April.”
That is useful. If her confidence is low, she can ask me once and learn from the answer.
Lest you think I’m describing some distant AI future, my Home Assistant agent already does variations of this today.
Friday maintains structured household memory, understands rooms and the people using them, connects equipment and supplies to where they live, searches manuals and documents, manages household work, and reasons across the systems operating the place. She can compare food inventory with a recipe, identify shortages, handle unit conversions, estimate cost, consume stock, log waste, and return leftovers to inventory. The agent now tracks chores, maintenance, service requests, schedules, machine consumables, equipment wear, guests, household preferences, and the live state of rooms. She can retrieve supporting information from manuals, notes, tickets, documents, and the household knowledge base whenever she needs more detail. She already works across HVAC, energy, water, lighting, shades, security, cameras, media, plants, and the spa. She can assemble the relevant context proactively, so I don’t have to interrogate every system separately.
She is also capable of doing the household books, including inventory costing, waste, depreciation, and COGS, well, as soon as I finish supplying the source data and setting up the starting books.
The limiting factor in that last one is currently the human, who has apparently constructed an AI accountant so he can procrastinate about accounting more efficiently. 
The home develops a living semantic model of its spaces, people, machinery, supplies, knowledge, and activity. The agent uses that model to quietly remove clerical work from the humans living there.
We get to remain our best lazy selves.
The agent gets the clerical job. My question is what actually is a “non-smart household item” in an actual smart home… 
I’d say none such exists…
