You know it’s just tiresome … the constant breaks.. the endless spook notifications
The relentless firmware upgrades..
Not a whinge …ok.. a long time HA user here.. (and a long term Nabu Casa subscriber) so please don’t shoot the messenger
But a bot that HA officially supports and sorts out all the “breaks” and cleans up my HA sure would be a welcome addition…
We are all busy people.. I get it.. running Proxmox servers, home labs, large networks..
So I’d love an official bot ..that speeds up my productivity..
Or am I missing something?
1 Like
Well you’re certainly missing the fact that everything you hate doing is actually what a lot of people enjoy doing as a hobby and would loath giving control of this to some over confident bulshitting clanker that would likely fuck things up even worse than the devs occasionally do.
13 Likes
Hi, since you are a long time user, you know that HA is so versatile that breaking stuff is unavoidable which also depends a lot on what you have installed on your system.
As you can see, I’m using HA even longer then you and my experience seems to be quite different from yours.
If I see what people are trying to do with their HA, which is far from home automation, it’s not strange that things break (frequently).
You have a point that it could be convenient to use AI for helping you with it.
I guess you also have an idea what it takes to make that work and I assume that this is out of scope since, I think, this is not what the majority want but I could be wrong.
Before we had FR here in the forum, now you have to make those on discord: so maybe you might get some better responses there…
2 Likes
@solarjunkie, totally agree!
I tinker with HA too whenever I have the time, but pretty often the HA rabbit hole is just impossible to get through, even though it has definitely improved over the years.
I’d really benefit from some proper configuration assistance, especially with the damn YAML jungle, which becomes completely unmanageable as soon as you have more than a few lines of it.
I have about 15,000 lines of it in my automation.yaml file alone. It is easy to manage and only took a couple of years to master in my spare time.
Sounds to me like you just want to avoid learning because there is no instant gratification.
1 Like
I would hate this in core and would rather see a no AI stance from the org because of the widespread and serious harms that LLM tools can cause to those using them, and even moreso to everyone else.
Even without the numerous harms, I wouldn’t use AI because I like the keep my ability to understand and reason about things, and because learning and problem solving bring me joy.
These are the reasons I chose to look into home automation (when I no longer had time enough to program), and it’s a good substitute for that, which also happens to be useful in my life. Win win!
I also wouldn’t trust the output. Anyone here seen the news that the US came close to war with China because of convincing but wrong output from an LLM that was pushed up the chain of command? Deployment in military is rife and reckless.
1 Like
You’re missing a lot of something.
First bolting an LLM to the side of HA does not ‘make an agent’ I know people are used tk providing an api key ad and hitting a big ol easy button. But it doesn’t work that way.
You’re missing Context engineering effort.
Aka the crap we don’t think about when we all get annoyed that Alexa or Google didn’t listen. It’s not magic
It would literally take two more years at current for them to be successful. Not like quantum computing happens in 5 years every 5..like an honest lift with the same effort they did for year of the voice.
Im pushing my context work at that pace and It’s STILL taken me two years to get here.
Doing NOTHING but context engineering work. And I don’t consider it anywhere near done, I’m just now getting to basic security building blocks and gated calls.
They’ve (ohf/Nabu/paid dev) been laying the plumbing for stuff like this to even work. It’s still a bit of a lift. If you look at the UI work. It’s also precursor context work if those modules expose the same thing as JSON to the agent they do on screen (like the new history or maintenance view would be gold for context for an agent at runtime) you can see the underpinning happening if you squint and look hard.
They’re also being careful. Im seeing new methods for security context and making it easier to tell those things inside the chat (this is necessary for future agent security) There is a lot of damage that can be done if they do it wrong.
You can have fast, good, cheap. Pick two. Theyve apparently picked cheap and good at OHF.
3 Likes
Just to share, we are currently working hard to get our vision for AI in Home Assistant aligned at the OHF. Once it is there, it will be shared (and next figure out how we can get there or what it exactly is going to look like).
Just to make it very clear, we are working on a higher level vision, not implementation details or features at this point.
Yes, this will include possibilities like running our own harnesses and possibilities for using AI inside Home Assistant.
One thing I already can share for sure (just to prevent someone bringing out the pitch forks), it will be your choice if you use it and what you’d hook up.
../Frenck
Blogging my personal ramblings at frenck.dev
15 Likes
I totally agree.
And with my 2GB Pi 4, “She canna take much more cap’n!”
(BTW, bullshitting has 2 l’s in it).
Thanks for pushing frenk.
Also your case for some of Spook going into core becomes almost mandatory to do some of the things context engineers need just sayin.. Keep going.
Get me a way to clear single repairs and read the maintenance dashboard and a alias data directly through actions pls. 
2 Likes
I at times move things from Spook to Core when it makes sense. You might have seen Spook shipping with new Triggers, Conditions to experiment with; some of these will soon move over. Same for a bunch of actions it has.
Spook is my playground. Not everything is suitable for core (like the fault detection, is really not at the level/bar it needs to be for Core yet).
../Frenck
Blogging my personal ramblings at frenck.dev
5 Likes
Oh there’s an entire ‘ectoplasm’ tool in ZenOS. Implements almost every tool in Spook. 5.3 in agent space.
Zenos simply wasn’t possible without your label tool. PERIOD.<< thays my first choice.
I’m glad to read this 
I don’t want it inside my system.
3 Likes
I hope that will always mean off by default.
2 Likes
I’d still take it with a grain of salt. Just like we can’t choose to remove the built in dashboards (hiding is not removing). So you might be given the choice to hide it but not actually remove it,
We still don’t even have control over automatic naming of our own entities. e.g. You do not have the choice to remove the device in the automatic naming scheme despite the fact a fancy tool was developed to give the illusion of choice.
5 Likes
15,000 lines and only a couple of years to master in your spare time? 
@Rusty, I think you may have accidentally written one of the better arguments for an AI assistant in this whole thread. At that point automation.yaml isn’t really a config file anymore, it’s a small legacy software project. 
And just to be clear, I’m not talking about letting some hallucinating bot loose in HA and hoping for the best.
Make it optional. Let me choose local or cloud. Let me decide what it can see and what it can touch. Read-only if I want. And if it wants to change something, show me the change first and let me press the button.
With a proper integration into the actual HA instance though, I can see loads of useful stuff:
- “Why didn’t this automation trigger last night?” (i.e. look at the trace, states, history and config and tell me what actually happened).
- “Where is this entity used?” (i.e. check automations, scripts, templates, dashboards and helpers before I rename or delete the damn thing).
- Find dead entities, stale YAML, duplicate logic, broken references and random old config that hasn’t done anything useful for three years.
- Spot deprecated stuff after an upgrade and tell me what actually needs fixing.
- Take 200 lines of lovingly hand-crafted spaghetti-ninja-YAML and turn it into something I might still understand six months from now.
- Build or tweak an automation from plain English, validate it against the current HA version, show me the diff, and let me decide if I want it.
- Look at logs, Repairs, traces and integration errors together instead of making me play detective across five different screens.
- Explain some weird old bit of config (i.e. save me from spending two hours digging through docs, GitHub issues and forum posts).
- Answer questions about the whole setup (i.e. “what breaks if I remove this helper?” or “what can actually turn this switch on?”).
- Help clean up years of HA cruft, including things like
input_boolean_temp_test_2_final_really_finalthat nobody remembers creating.
For me, that’s the interesting bit.
A generic chatbot that knows some Home Assistant docs is… fine.
An assistant that actually understands my HA instance, my config, my entities, my traces, my logs, my history and the version I’m running is a completely different thing.
And no, it doesn’t mean I don’t want to understand my own system. I do.
I’d just prefer spending my spare time building useful stuff rather than slowly becoming a professional Ninja/YAML archaeologist.
As long as it’s optional, scoped, transparent and under my control, I honestly struggle to see the downside.
8 Likes
awesome post. KBOOM. that is exactly what I am after 
Only if it’s not baked in.
I’d stop updating HASS if it was going to force AI on me.
Isn’t that a problem?
Even the creators failed to stop it doing things like, breaking into other sites and even modifying itself.
I’ve heard of a few instances where AI has done things that weren’t ethical, but I hold the people managing the specific tools responsible for the results, not the model at the heart of the issue. There are a lot of people that work with AI that know that the model can’t do anything if it’s properly locked down.
Having said that, I do agree that people using Home Assistant should be the ones that ultimately have the final say on what is deployed in their instance, and an AI bot from the Open Home foundation, however well intentioned, is not going to be well received by users that have no desire to have one.