Home Assistant AI Deep Dive: LLM Vision, Copilot, OpenClaw & More

0:04 [SPEAKER_00]: Hello and welcome to the home misses and podcast. [SPEAKER_00]: My name is Phil and joining me as usual. [SPEAKER_00]: I've got Ruham. [SPEAKER_00]: Hey Ruham, hey Ruham. [SPEAKER_02]: Hey good, how are you? [SPEAKER_00]: Yeah, good. [SPEAKER_00]: Thank you. [SPEAKER_00]: This podcast is sponsored by home misses and cloud by Nabukasa.

0:17 [SPEAKER_00]: For some more monthly fee, you'll unlock powerful features, like secure effortless access to our system from anywhere. [SPEAKER_00]: Your choice of voice assistance, off-site backups, and more. [SPEAKER_00]: All configurable in the UI, and there's no gamma needed. [SPEAKER_00]: It also supports the development of home assistant, SP home, and other open home foundation projects.

0:36 [SPEAKER_00]: Click the link in the description to learn more. [SPEAKER_00]: We'd also like to give a shout out to our Patreon members, including our executive producers, Benny, Jason, and Rob. [SPEAKER_00]: You can help support the show and get early access to episodes all in and out for your feed. [SPEAKER_00]: To support the show, check out home assistance.fm and click a patreon in the menu.

0:54 [SPEAKER_00]: And of course, if you can't do that, we'd appreciate a like, comment, subscribe or rating wherever you're listening or watching us today. [SPEAKER_00]: It really helps us out. [SPEAKER_00]: We're hard. [SPEAKER_00]: It's just you're not here today. [SPEAKER_00]: No guests and no release today.

1:06 [SPEAKER_00]: We thought a lot of feedback would be in receiving recently is about, uh, we want to hear about your, uh, our use of the smart home and how we're using home assistance. [SPEAKER_00]: So I thought, you know, let's, [SPEAKER_00]: little do a little dive into AI today, I don't know, AI is the buzzword at the moment, I don't know, a lot of engagement on episodes where we're covering AI, either, you know, release, or if I guess it's using AI, I don't know, I've used a bit of AI in my smartphone, I was sort of tinkering around the edges with it still.

1:36 [SPEAKER_00]: Yeah, your opinions were on what you're doing, but yeah, I thought you'd [SPEAKER_00]: come at you first, if I can, I like what have you done with it in your home? [SPEAKER_00]: Is it something that you want to bring into your business install or you're trying to avoid bringing it in? [SPEAKER_02]: Neither really, the one thing I did do is I have Gemini hooked up to my home assistant voice P.E.

2:13 [SPEAKER_02]: So, when I give it the trigger word, I can ask it. [SPEAKER_02]: whatever, right? [SPEAKER_02]: Which is nice. [SPEAKER_02]: Um, you know, again, this is standard examples of like how tall so I have a tower, whatever rate that I can answer that because even without context, right? [SPEAKER_02]: But, uh, so that's uh, that's something I've uh, I did actually quite a while ago, I think when the functionality first came out.

2:37 [SPEAKER_02]: Um, but I'll just in that of actually not been doing a ton [SPEAKER_02]: solid foundation for AI tool, so like for example, like running stuff either at home or in the cloud. [SPEAKER_02]: And when I say in the cloud, I don't mean like [SPEAKER_02]: Like, I don't know, chat to you PT or one of those, I mean, like, private models in the cloud, right?

3:08 [SPEAKER_02]: So trying to build a base level that can actually start doing things intelligently, like, intelligent routing and all that stuff. [SPEAKER_02]: And then once I do that, the idea is just hook it into a home assistant and then that way, it's kind of, it's cheaper and it's just more optimized and all those kind of things.

3:27 [SPEAKER_00]: What are you using to do that? [SPEAKER_02]: Right now, at least what I'm playing with is Olamma with models running an Olamma cloud. [SPEAKER_02]: Yep. [SPEAKER_02]: And then light LLM as well. [SPEAKER_02]: There's a couple of other tools. [SPEAKER_02]: None of them have really hit the mark for me.

3:46 [SPEAKER_02]: One of them is called OptLLM. [SPEAKER_02]: I think I'm just using it wrong. [SPEAKER_02]: So again, I'm still in that phase where I'm like, let me go through. [SPEAKER_02]: At least get everything working and then figure out. [SPEAKER_02]: You know, what I need to tweak or what I don't, right?

4:00 [SPEAKER_02]: So the idea is, and I know you've done this a little bit too fill, it's just like, you know, throw like, I don't know, whatever, five bucks or 10 bucks into some of the paid services, like, like, Jackie between to the API or whatever. [SPEAKER_02]: Yeah. [SPEAKER_02]: And then run stuff through Olauma Cloud, just based on which specific models I want to use.

4:23 [SPEAKER_02]: And again, I'm still figuring that part out. [SPEAKER_02]: And then, as a fallback, it can go through to a paid model, whereas this is just a 20 bucks a month or something and from a subscription instead of paying for API usage, right? [SPEAKER_02]: So, yeah, so that's kind of the idea of, like, hey, what am I trying to do?

4:45 [SPEAKER_02]: What am I trying to prioritize? [SPEAKER_02]: Whatever, and throw everything into one system and then have it go that way. [SPEAKER_02]: I'm sure there's stuff that exists, but I'm doing it in hard mode. [SPEAKER_02]: right now. [SPEAKER_02]: And just stumbling through it and figuring it out.

5:00 [SPEAKER_02]: At least for me, that's just I'm using it as a learning opportunity, right? [SPEAKER_02]: That have no pressing need for that today, but I'm sure it'll come up at some points. [SPEAKER_02]: I'm just trying to do that right now. [SPEAKER_02]: It's just build that foundational layer, and then just whatever services I want on top of it.

5:17 [SPEAKER_02]: So it's like chat. [SPEAKER_02]: I could throw on an open web UI, radar, or whatever, and keep it all kind of within [SPEAKER_02]: My quote on quote, my own four walls, but even though the models are running in the cloud, right? [SPEAKER_02]: But yeah, and then eventually, if I ever get a GPU run on myself.

5:36 [SPEAKER_00]: Well, that's it also right, like the pricing of everything has just gone up so crazy now, like getting even your hardware in your four walls is a bit expensive at the moment. [SPEAKER_00]: I know I've certainly been [SPEAKER_00]: repurposing hardware and I've haven't changed my, I'm Mrs. Hardware in, you know, more than three or four years, right?

6:00 [SPEAKER_00]: Like it's still running on an old, I'll let over desktop PC as a Ubuntu server. [SPEAKER_00]: So there's no AI capability on that thing. [SPEAKER_00]: I'm not running models on that thing. [SPEAKER_00]: It's just, I'm a distant and a couple of other self-housing containers, right? [SPEAKER_02]: So there is that there's that one model that I said Microsoft somebody released that apparently it's optimized for CPU and do it that kind of thing.

6:25 [SPEAKER_02]: I haven't played with it. [SPEAKER_02]: I don't know how good that is but [SPEAKER_00]: Yeah, I can, I can't get out of got a fancy enough CPU for that. [SPEAKER_00]: Right. [SPEAKER_00]: To Shunt. [SPEAKER_00]: So with that, yeah, like I guess, you know, if then it comes to how do you wish all your AI stuff to the cloud home assistant, so out of the box, I guess, comes with some AI now built in.

6:46 [SPEAKER_00]: That came in in the 2021 or the 2025 releases. [SPEAKER_00]: So if you add in an open AI API key, and I think there's another couple of providers that are also supported. [SPEAKER_00]: home misses and other box we'll have some very basic AI features so you can do AI suggestions based on automation names, potential labels and areas for those automations when you create them.

7:11 [SPEAKER_00]: So if you create an automation in the UI and it has the steps of you know, 20 on the garage light, it will prompt you, you can say suggest and it will then come out and give you okay. [SPEAKER_00]: Let's go ahead and name this automation turn on the garage light and [SPEAKER_00]: going a little bit deeper into the online is where you get to the cutting edge stuff and just sort of start heading into, you know, you really need to be able to deep out yourself.

7:42 [SPEAKER_00]: So there's a few things I've started playing with. [SPEAKER_00]: Straight off the bat, like I think for me, get help, co-pilot, and then by extension, Claude Code, has been a big game changer for me in terms of maintaining my YAML missions, though. [SPEAKER_00]: I'm not really a UI guy, I would love the ability to have my [SPEAKER_00]: UI automations and yummy automations in sync all the times that I could switch between the two, but for the moment, I'm got 50, 50, mainly, or a lot of 50 exactly, but you know, mostly, my automations are done in yummy, I'm a few that are done in AIX, in UIX, exclusively.

8:31 [SPEAKER_00]: But having all my automations in YAML already has really been able to make me leverage AI, particularly around GitHub co-pilot. [SPEAKER_00]: So using VS Code, I can get co-pilot to create me automations now. [SPEAKER_00]: It will also use the smart suggestions to suggest, not only entity ideas, which, if you're using VS Code, there are a home business and extensions for it.

9:00 [SPEAKER_00]: free fill, oh, yam or blocks of what I would like it to do based on. [SPEAKER_00]: And now, um, I usually didn't write, I am automation descriptions, but now I'm finding myself writing description. [SPEAKER_00]: So, you know, I've write down an ID and name and I'll give the automation and description.

9:16 [SPEAKER_00]: And when I give it enough of a description that allows a VS code and GitHub code pilot to then, [SPEAKER_00]: brief fill a whole bunch of that automation for me, using conditions, using triggers based on the description of the automation I've already given. [SPEAKER_00]: So yeah, that's good.

9:37 [SPEAKER_02]: That's a good use of it. [SPEAKER_02]: I mean, I know that's not like again, copilot within BS code has been there for for a decent amount, right? [SPEAKER_02]: And even auto-filling entity names and stuff like that has been, [SPEAKER_02]: has been part of the home assistant VS code integration fairly correctly.

9:56 [SPEAKER_02]: But yeah, it's, it's, that's a good use of, like, again, cloud code and stuff like that, right? [SPEAKER_02]: And I haven't, I haven't been brave enough to point cloud at my configs yet. [SPEAKER_02]: The one thing I do want to play with is the homelessness in MCP, right? [SPEAKER_02]: Where you can plug that in, or run it, I guess, on your, on the same system and go and plug it into cloud and say, hey, you know, how's my house looking, how are my plants looking, how are, like, whatever, right, like, [SPEAKER_02]: where it could be like, hey, you really need to fertilize your, whatever.

10:33 [SPEAKER_02]: I mean, it's a state of my plan to typically not do it due great, so that they need a lot of just a lot of love. [SPEAKER_02]: But yeah. [SPEAKER_00]: There's, I believe there's two MCP because they have community driven server. [SPEAKER_00]: I think there's hash, so hash AWS, hash MCP, and there's another one, H-A, MCP.

10:56 [SPEAKER_00]: But yeah, they allow agents like Claude to go in and discover the tools, the entities, and then query for the home assistant API to find out states of those. [SPEAKER_00]: So yeah, it looks promising. [SPEAKER_00]: haven't gone down the MCP path too much yet, I'll talk about it soon, like my struggles with it.

11:18 [SPEAKER_00]: Yeah, I was mentioning just before like GitHub co-pilot and Claude code. [SPEAKER_00]: One, [SPEAKER_00]: really good extension I've come to use a lot now, both at work. [SPEAKER_00]: And for home assistant is called Open Code. [SPEAKER_00]: So our code is a terminal based application. [SPEAKER_00]: They've just released in beta, a UI version, [SPEAKER_00]: Um, that you can, uh, basically have as Claude Coats and what you would write, uh, you would basically interview or talk to your agent and say, hey, this is what I want to do.

12:00 [SPEAKER_00]: You're going to like a big planning session, um, and then, uh, the agent then go, uh, and run that for you. [SPEAKER_00]: So I've been able to, um, um, [SPEAKER_00]: set up a planning session with OpenCode and tell it that I want to, but I have the other day. [SPEAKER_00]: So I previously had a notification color system, you went back to my rented place.

12:24 [SPEAKER_00]: I sort of left that early to certain, [SPEAKER_00]: The lights around the house, so I wanted to really pull that out into a proper, understandable thing. [SPEAKER_00]: So I just said to the AI agent in Open Code, I would like to introduce notification and accent lights. [SPEAKER_00]: I want to use tagging, so now I label a device in home as an accent light and any accent light that is labeled as an accent light.

12:53 [SPEAKER_00]: And also, I think I've introduced another condition like a label type of notification light. [SPEAKER_00]: When the dishwasher is cleaned, when the washing machine is cleaned, when the internet goes down, I think there was something else I added, change the accent lights to the specific color for that event.

13:15 [SPEAKER_00]: So I think I've got green for the washing machine pink for the dishwasher, for example. [SPEAKER_00]: And then if the future I add another light strip or another colored light bulb somewhere, I don't have to then automate that specific bulb. [SPEAKER_00]: I just go into the automatic, the UI of Home Assistant and go label this as a notification light as an accent like done.

13:36 [SPEAKER_00]: Yeah. [SPEAKER_00]: And yeah, I was able to plan it out. [SPEAKER_00]: We actually drink interview process. [SPEAKER_00]: I came up with some logic errors. [SPEAKER_00]: For example, I only want a light to change color if it's already on and make sure that when the light turns off or when the notification clears, that it returns the light to the previous state.

13:54 [SPEAKER_00]: Right. [SPEAKER_00]: So yeah, we did some back and forward. [SPEAKER_00]: I end out those sort of edge cases and then I [SPEAKER_02]: Yeah. [SPEAKER_02]: Have you, have you done anything with like using, you know, whatever cloud code open code, anything like that with, uh, with building a UI or like a building a dashboard.

14:17 [SPEAKER_00]: But yet, um, [SPEAKER_00]: you know, I haven't really thought about that. [SPEAKER_00]: I didn't really think about that until I installed a couple of other extensions that AI and I miss this, which I'll get to be later. [SPEAKER_00]: But no, like I think because I know like there's a new home dashboard that obviously doesn't release it's now the default as of 2026.2.

14:40 [SPEAKER_00]: I just [SPEAKER_00]: Yeah, for me, like, I haven't, I don't really try to use, there's only myth uses the UI, right? [SPEAKER_00]: I don't look for anything to look nice. [SPEAKER_02]: I mean, even in terms of, like, hey, give me, like, optimal placement, place it out nicely, so that it's, like, easy for me to use a mobile, like, those kind of things, right?

15:02 [SPEAKER_02]: It's funny, I just re-did my tablet dashboard, but it's just using the sections layout because my tablet dashboard was just only focused on areas that my tablet is around. [SPEAKER_02]: So for example, my tablets in my kitchen, so the kitchen and the living room, and those kind of adjacent areas, and just my, my.

15:22 [SPEAKER_02]: My son's room, just again, I guess we need to do anything there, and so, you know, I rebuilt it out and included the whole house, right, all of my main entities, and trying to make that my new main dashboard, but it's like, I don't know, it's. [SPEAKER_02]: It's not perfect, you know what I mean, like even even as I use it is like there could be a little better like space A and they could be a little better like so it's like hey you have access to all of these plugins you have access to all of these you know whatever go do it essentially right like and build me the perfect dashboard and based on my usage and stuff like that so it's like oh you see I actually turn on my range of like manually every every time.

16:06 [SPEAKER_02]: go in and maybe place my kitchen stuff at the top. [SPEAKER_02]: So just in terms of that, right? [SPEAKER_02]: Because it knows usage, it knows all of that kind of stuff. [SPEAKER_02]: So great go, go talk to it and build me my most used for the top and down to my least use kind of thing.

16:21 [SPEAKER_00]: I haven't found A.I. [SPEAKER_00]: to be as good with UI as I would like. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: I think it can sort of add a whole bunch of, at least in my professional life, it can add a whole bunch of low colors, especially around CSS and margins and all that, for for not.

16:41 [SPEAKER_00]: what it needs to do. [SPEAKER_00]: So I don't really trust it on the UI yet. [SPEAKER_00]: I maybe I should try and, because I guess, like, for the home Mrs. and Dashwell is 50, it is YAML at the end of the day. [SPEAKER_00]: Right. [SPEAKER_00]: Right, exactly. [SPEAKER_00]: So yeah, maybe I should try it.

16:58 [SPEAKER_00]: Yeah, but that'll be. [SPEAKER_00]: So one of the good things about open code, and I'm guessing you can do an enclosed code, is you can have specialized subagents. [SPEAKER_00]: So you can give a specialized probe to someone. [SPEAKER_00]: So for example, if you're in a web development world like I am, you might have a database expert, you might have a testing expert, a front end expert.

17:21 [SPEAKER_00]: Maybe I need to create a few agents or subagents. [SPEAKER_00]: for home visits, specific stuff, you know, like a home visits and dashboard expert, and they can just know the intricacies of, you know, what's a view? [SPEAKER_00]: What's a section and all that. [SPEAKER_00]: Yeah. [SPEAKER_02]: Yeah, exactly.

17:38 [SPEAKER_02]: Just give it a context of everything and be like, now be something else, design something. [SPEAKER_00]: Yep. [SPEAKER_00]: I will say if you are looking at getting into like any of these like coding agents or anything. [SPEAKER_00]: I can code is great. [SPEAKER_00]: It works with a whole bunch of AI providers.

17:54 [SPEAKER_00]: I think you can throw like an opening IT at it. [SPEAKER_00]: You've got GitHub access. [SPEAKER_00]: Like if you use GitHub, you can use GitHub, you can use GitHub, [SPEAKER_00]: The there's also an extension or a plug-in to I can code called OmyOpenCode, which I would highly recommend as well.

18:13 [SPEAKER_00]: It sort of just adds like a whole, but it basically allows you to just type a prompt, get it to start work and then leave it for a few hours and it will just keep iterating and checking over itself, ensuring it's reached the task was given. [SPEAKER_00]: So it takes it to the next level.

18:29 [SPEAKER_00]: So we'll highly recommend those. [SPEAKER_02]: It's the one thing with open code as opposed to that cloud code and stuff is because cloud code will take your subscription if you have the, if you're doing like, I don't know, 20 bucks a month, or even the bigger, 100 bucks or whatever it is, whereas open code uses API.

18:49 [SPEAKER_02]: Do you find that there's a more of a cost there if you do it that way or for me? [SPEAKER_00]: No, for me, I'm lucky that my employer pays for GitHub wallet. [SPEAKER_00]: So yeah. [SPEAKER_00]: For me, like I, the models I try and use are like opus, opus, from Claude, I find, yeah, like, I think everyone has their own story with each model, like some people go up to me and be like, [SPEAKER_00]: Oh, the GPT codex models are absolutely brilliant.

19:21 [SPEAKER_00]: Like, you've got to use that. [SPEAKER_00]: Other people swear by the Gemini models for coding. [SPEAKER_00]: Apparently, rocks really good at coding. [SPEAKER_00]: I just try stick with the Claude Ope's models for me. [SPEAKER_00]: I've got whatever is in the others, just to be a mixed bag.

19:37 [SPEAKER_00]: Maybe it's the AI prompted. [SPEAKER_00]: Maybe it's just what I'm working out of time, right? [SPEAKER_02]: Yeah, and that's the other thing worth mentioning too, right? [SPEAKER_02]: It is, again, sometimes different models are better at different things, but sometimes better models work for you because, I mean, you're either great at prompting it one way or you're not right and that part is true too.

20:03 [SPEAKER_02]: There's, I know there's a lot of, even on YouTube and stuff, there's a lot of prompt engineering, [SPEAKER_02]: courses you can take right in terms of like, hey, how do you want to structure it and do it those kinds of things? [SPEAKER_02]: I don't know if you're if you're doing this enough, then maybe there's value in investing your time to do that.

20:20 [SPEAKER_00]: Yeah, and I think I honestly not going anywhere. [SPEAKER_00]: So I might be a good opportunity to do it. [SPEAKER_02]: Yeah, yeah. [SPEAKER_02]: I mean, I hope it gets more efficient. [SPEAKER_02]: I just, I don't love how inefficient. [SPEAKER_00]: But, um, so there was a couple of, sorry, we were mentioning before on, like, you were sent a build range of models and, you know, you know.

20:46 [SPEAKER_00]: getting a little bit wet. [SPEAKER_00]: There's a couple of hacks plug-ins that you can install that might be up your alley. [SPEAKER_00]: One of those is called the AI agent. [SPEAKER_00]: So the AI agent is available and hacks to install it. [SPEAKER_00]: It creates a new menu into home-missident, which can be used to talk to your home [SPEAKER_00]: to assist.

21:14 [SPEAKER_00]: Whereas assist tries to work out what you're trying to do. [SPEAKER_00]: This is purely talking to the LLM about your home misses and instances. [SPEAKER_00]: So the idea is that you can talk to it and go okay, what lights are on at the moment or how much electricity did I use today? [SPEAKER_00]: You can also ask it to suggest some automations.

21:39 [SPEAKER_00]: Help me build out an automation to turn on [SPEAKER_00]: Yeah, and it will go ahead, give you the demo code that you can then copy and paste. [SPEAKER_02]: So it's a chat interface with an AI. [SPEAKER_02]: Yep, exactly. [SPEAKER_02]: Back end. [SPEAKER_02]: That's informed. [SPEAKER_02]: Interesting.

21:56 [SPEAKER_00]: So it uses its own mechanisms to communicate with the APIs. [SPEAKER_00]: So there's no, so if you've added your OpenAI integration to home, it doesn't use that. [SPEAKER_00]: You'll have to open AI keys separately to this integration. [SPEAKER_00]: But it supports a few providers. [SPEAKER_00]: I think there was Gemini, OpenAI, and a couple of others.

22:18 [SPEAKER_02]: Yeah, I'm looking at it right now. [SPEAKER_02]: OpenAI, Gemini, and Thropic, OpenRetter, Alter, SetAI, and Lama. [SPEAKER_00]: So, yeah, you can chuck your API key into that. [SPEAKER_00]: And then you essentially get a new menu, the left hand side, you know, home assistant, you can then fire up a, yeah, and then you've just got a prompt, especially like a free-per-familiar, free-per-familiar, free-per-views, chat-chip, and even before it's just a chat window.

22:40 [SPEAKER_00]: So then you can start talking to the LLN about your home assistant instance. [SPEAKER_00]: Right, right, right, right. [SPEAKER_00]: That's kind of neat. [SPEAKER_00]: Another small one, or one or a small bit, another hacks out on that you can get is called AI Suggester. [SPEAKER_00]: So this one is, this is actually pretty cool.

23:03 [SPEAKER_00]: It's always running in the background of your home Mrs install and it uses the notifications platform in home Mrs to then make suggestions for you. [SPEAKER_00]: So for example, when I installed it, I've only been running for a few days just testing it out. [SPEAKER_00]: It's suggested things like, oh, it looks like you're opening the garage door at this time of the day.

23:28 [SPEAKER_00]: Let's make it automation to automate that. [SPEAKER_00]: It can then go ahead and go, oh, it looks like when the house goes into the waymo, we could turn the TV off automatically. [SPEAKER_00]: So it's looking over your automations and your entities trying to work out where it can add value.

23:48 [SPEAKER_00]: And then they just pop up as a notification into your instance and then gives you the yellow code to then plug into your automations. [SPEAKER_02]: Nice. [SPEAKER_02]: I've bought both of these as you've said it. [SPEAKER_02]: I've downloaded them. [SPEAKER_02]: So let's let's play with them.

24:04 [SPEAKER_00]: And once again, I also just need OpenAI keys. [SPEAKER_00]: It doesn't use, you've already added OpenAI keys to home assistant. [SPEAKER_00]: It doesn't leverage those. [SPEAKER_00]: You have to then add your own keys into that integration. [SPEAKER_02]: Yeah, and the cool little hack here is if you have, again, stuff like, like I mentioned like LLM earlier, it's actually open API compliant.

24:28 [SPEAKER_02]: So as long as you can change the URL of the API endpoint, so to say, instead of. [SPEAKER_02]: you know, uh, opening i.com slash if you have whatever it is, uh, if you can do, uh, like point it to your instance, then that'll work too. [SPEAKER_02]: So that's how I'm going to be trying on these guys.

24:50 [SPEAKER_00]: Now, if you're looking to upgrade your smartphone, you can't go past the zoos. [SPEAKER_00]: Zoos have just released their brand new Z-Wave a long range RGBW dimmer. [SPEAKER_00]: You can now automate 12 to 24 volt LED strips with full brightness and color control. [SPEAKER_00]: It's designed to easily install behind your memory wall switch, or you compare it with the Z-Wave scene controller for more flexibility.

25:12 [SPEAKER_00]: You can use the RGBW dimmer for accent lighting, under cabinets, behind screens, or wherever crazy you can think to put an RGBW light strip. [SPEAKER_00]: Zeus also in official works with home assistant partner. [SPEAKER_00]: Helping network cast-up by providing long range devices for testing the home assistant connect is at WA2 radio, so you know you're in friendly hands.

25:33 [SPEAKER_00]: Zeus offer a range of affordable and innovative devices, including water leak sensors, water valve controllers, smart plugs, and scene controllers. [SPEAKER_00]: There's N78 High Power Relay, allows you to control and monitor higher load appliances, up to 40 amps, which is perfect for pool pumps, air compressors, and other high power outlets.

25:53 [SPEAKER_00]: So for the best prices on all zoos products head over to the smartesthouse.com, that's the smartesthouse.com. [SPEAKER_02]: All right. [SPEAKER_02]: So now again, we're talking about AI. [SPEAKER_02]: Obviously there's a lot of folks doing things like, you know, your doorbell notifications and stuff where they goes back and rewrites it.

26:15 [SPEAKER_02]: adds an image and you know those kind of things where it says like okay you know I see a person wearing a hat and a black sweater with a snowy background or whatever right like are you are you doing any of those right now and and or where do you want to go with that? [SPEAKER_00]: Yeah, so, and I think I mentioned it a few times in parts of my other episodes, but yeah, so the one I would recommend to use is LLM vision.

26:43 [SPEAKER_00]: So that's another hacks component you can leverage. [SPEAKER_00]: The idea is once again, you put in your opening like keys or whatever your flavor of choices to AI model, put the AI keys in, give it a camera entity, and I think you can also [SPEAKER_00]: put prompt in as well. [SPEAKER_00]: So it's several op modes to it.

27:04 [SPEAKER_00]: For me, I'm using it when the house is in an arm to state. [SPEAKER_00]: So I then night away vacation and any about external cameras. [SPEAKER_00]: We use real-it cameras here. [SPEAKER_00]: They have a, they're in built in AI for person and animal detection. [SPEAKER_00]: So whenever a person is detected via a camera, it then triggers [SPEAKER_00]: entity in a home assistant to go after his presence detected.

27:31 [SPEAKER_00]: If the house is in an armed state, LLM vision will then be activated to take a snapshot of the camera feed and then it's sent over to my case OpenAI and it asks what's going on in this frame. [SPEAKER_02]: Interesting, so it's only when you're not home to save on tokens and stuff. [SPEAKER_00]: Correct, yet.

27:53 [SPEAKER_00]: There's no point in telling me that it's seeing myself, all right? [SPEAKER_00]: Because yeah, processing images can be like a burn through credits, pretty quickly. [SPEAKER_00]: And yeah, and I think it has some built in prompts or pre-loaded prompt. [SPEAKER_00]: You can then refine it, you know, like, tell me, you know, talk like a pirate, you know, or explain, like, you know, give some, I've seen some funny people making some prompts out of it, right?

28:22 [SPEAKER_00]: Like, you know, make sure you're rude to this in description, right? [SPEAKER_00]: You know, making room common about their clothes or something like that, which can be funny, if you know, getting an notification, you're doing a little like, you know, this guy thinks he looks cool in his backwards hat, you know, but really, he, yeah, which is does not just hair today, right?

28:34 [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah.

28:36 [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah.

28:38 [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [UNKNOWN]: Yeah. [UNKNOWN]: Yeah. [UNKNOWN]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah.

28:40 [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah.

28:42 [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [UNKNOWN]: Yeah. [UNKNOWN]: Yeah. [UNKNOWN]: Yeah. [UNKNOWN]: Yeah [SPEAKER_00]: So for me, that's what I'm primarily using from a security point of view.

28:52 [SPEAKER_00]: I also use it, so I've mentioned before, we have a gauge at the end of our driveway. [SPEAKER_00]: It doesn't record yet. [SPEAKER_00]: I could put a door sensor on there, but then I need to, like, 3D print, whether it's housing for a door sensor, and plus, you know, the range issues, like I have to get in, is the B down to the end of the driveway.

29:14 [SPEAKER_00]: But I do have a camera pointing down the end of the driveway. [SPEAKER_00]: So I've taken some snapshots from the camera of the gate to the driveway being opened, and the gate to the driveway being closed. [SPEAKER_00]: I can then put that into the LLM vision memory. [SPEAKER_00]: with a description.

29:30 [SPEAKER_00]: I then tell LNVision. [SPEAKER_00]: This is a photo with the driveway gate closed. [SPEAKER_00]: This is a photo with the driveway gate open. [SPEAKER_00]: And then I can query the LNVision icon answer to question. [SPEAKER_00]: Is the gate to the driveway open or closed? [SPEAKER_00]: and it will then consult its memory, take the live snapshot, and compare all three images and give me a open or closed state.

29:54 [SPEAKER_00]: So there is a prompt, I think I written it in a way of like, you know, tell me you must only answer in open or closed, do my answer, you know, do not. [SPEAKER_00]: Give me a full sentence, yes or no, basically. [SPEAKER_00]: That is the only way you must respond. [SPEAKER_00]: And yet, that has worked really well.

30:10 [SPEAKER_00]: I, once again, to say one tokens, I leverage the person detection event of the AI in the camera, so that five minutes after a person has been detected, it then checks if the driveway gate is opened or closed. [SPEAKER_00]: Um, because otherwise, if you're detecting it on every event, um, you know, the wind moves the trees or, um, if you're doing it every 15 minutes, like the chances are, the dates are not going to be moving unless the person there, right?

30:40 [SPEAKER_00]: Um, so that's how I try and save tokens there. [SPEAKER_00]: that's cool. [SPEAKER_00]: And what else have I got? [SPEAKER_00]: And the next one, I also have an interesting use case where we have washing a washing line outside, particularly in summer, so it's summer at the moment, down in the summer hemisphere, and we've got washing outside.

31:04 [SPEAKER_00]: When the wind blows the washing around, it can actually be marked as a person detected by the AI of the Reeling cameras, whatever reason. [SPEAKER_00]: Yeah. [SPEAKER_00]: And when we're away, that obviously then triggers the AI to go ahead and go, hey, what's happening in this frame? [SPEAKER_00]: And then all I get is a photo of my backyard of washing and the description is no activity observed.

31:26 [SPEAKER_00]: So now what I'm doing is I'm adding in more conditions to that. [SPEAKER_00]: So don't check the backyard camera if there's washing outside. [SPEAKER_00]: I know if there's washing outside based on my washing machine being done or completing a cycle today and the laundry door being opened.

31:45 [SPEAKER_00]: that tells, I mean, this is that, okay, the washing machine is being done today. [SPEAKER_00]: Someone has walked outside and that means there's high probability that washing is now on the washing line. [SPEAKER_00]: So I can shut that camera, location is off essentially. [SPEAKER_00]: I can also then do a check based on the weather station, if it starts raining and the washing machine is run today, [SPEAKER_00]: Yeah.

32:11 [SPEAKER_00]: Check the washing line. [SPEAKER_00]: Is there clothes hanging on the washing line? [SPEAKER_00]: And if so, send it to the Lord, go bring the washing. [SPEAKER_00]: Because in winter we may do a washing load, but may not hang it outside. [SPEAKER_00]: We might put it around on a wheeler to dry it.

32:24 [SPEAKER_00]: So, yeah, there's no point. [SPEAKER_00]: We may need to not check it in that situation. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah, so that's really how like I would recommend Ellen Vision. [SPEAKER_00]: Other people I know are using it for garbage bins, you know, is the garbage bin. [SPEAKER_00]: Yeah.

32:44 [SPEAKER_00]: Been checking out, you know, take a photo of the garbage bins inside. [SPEAKER_00]: Oh, yep, they're still here. [SPEAKER_00]: That means they're on the street. [SPEAKER_00]: They need to go out. [SPEAKER_00]: I guess you could maybe do it for parcels as well. [SPEAKER_00]: I know some doorbell cameras have the cameras that tilt down to look at patches.

32:58 [SPEAKER_00]: Yeah, they go bottom of the camera. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah, exactly. [SPEAKER_00]: If you've got one of those cameras, we could potentially use Alan Vision to go how you have some of the lift parcels here. [SPEAKER_00]: Yeah, sir. [SPEAKER_00]: The front door. [SPEAKER_00]: Or even if you don't maybe don't have one of those cameras, maybe you've got a camera just facing your door.

33:13 [SPEAKER_00]: It's wide enough lens to do it. [SPEAKER_00]: And you could do that. [SPEAKER_00]: And I guess you could also like to try and count the number of cars in the driveway. [SPEAKER_00]: It's another occupancy or another presence, you know, is there a car on the driveway? [SPEAKER_02]: Don't want to close the the driveway gates today, for example, but it doesn't do, oh, I guess it is going through, it is going through and all, I'm so I can do OCR too, so you can check the plates and say, hey, this is, you know, romance car, as opposed to, you know, a friend's car or something.

33:47 [SPEAKER_00]: Yeah, potentially, if a camera is good enough quality, I guess the issue would be that the snapshot that is taken would have to have the plate in it. [SPEAKER_00]: So for me, the way I drive away and the camera positioning is the car may pull up, but between the snapshots taken, if the cars too far into the driveway, the license plate is already past the field of view of the camera, so it may not be able to capture the license plate.

34:15 [SPEAKER_00]: And once again, come on, it's my very, um, on how. [SPEAKER_00]: Yeah, you'll drop a user, right? [SPEAKER_02]: Um, yeah. [SPEAKER_02]: And the angles and and all of that stuff, right? [SPEAKER_02]: Yeah, you're being on there. [SPEAKER_00]: Yeah. [SPEAKER_00]: But yeah, Ellen Vision, I think, is it a little rabbit hole?

34:31 [SPEAKER_00]: It's got a hole. [SPEAKER_00]: It's got a really cool timeline feature as well that I love. [SPEAKER_00]: So you can see a timeline of events. [SPEAKER_00]: So, you know, this person, what a person was detected here. [SPEAKER_00]: He's a photo. [SPEAKER_00]: You can also manually add events to that timeline.

34:46 [SPEAKER_00]: So when the alarm panel arms or disarms, I add timeline events so that on my dashboard, I can see, okay, the alarm panels aren't at this time. [SPEAKER_00]: event was triggered at this time. [SPEAKER_00]: He's what happened. [SPEAKER_00]: Actually, really like that feature, like I think there would be a use case to have a timeline feature outside of LLM vision and the ability to add to it, like a custom, like through an automation, I think it would be very handy.

35:15 [SPEAKER_00]: I know there is the ability to add events to the home assistant logbook, but yeah, it's not going to [SPEAKER_00]: a time no fashion, you know, you don't have attachments to it. [SPEAKER_00]: You can't attach an image to it or give no description of the event to it. [SPEAKER_00]: Yeah, yeah, yeah, interesting.

35:33 [SPEAKER_00]: Hmm. [SPEAKER_00]: But it is February and I think everyone is talking about [SPEAKER_00]: plot, bot, open core, yeah, a multi bot. [SPEAKER_02]: Yeah, yeah, it's mold, right. [SPEAKER_02]: Yeah, it's changing. [SPEAKER_00]: Yeah. [SPEAKER_02]: Three times now, it's. [SPEAKER_00]: Yeah. [SPEAKER_00]: So it's now open core.

35:50 [SPEAKER_00]: Is it's a official name? [SPEAKER_00]: I hope you had to stay at the time. [SPEAKER_02]: Yeah, it was a time of recording. [SPEAKER_02]: Yeah. [SPEAKER_02]: It's I feel like we have to stay that. [UNKNOWN]: Yeah. [SPEAKER_00]: So OpenClaw is being hailed as the personal assistant or the AI assistant that was promised to us.

36:07 [SPEAKER_00]: For me, I've been playing around with OpenClaw this week trying to get it to work, arguing with it a lot. [SPEAKER_00]: It has a whole bunch of... [SPEAKER_00]: skills and integrations, which, and obviously there is a few for home isstent. [SPEAKER_00]: So yeah, I installed OpenClaw, I put it on it's own VPS, because if you put it on a, they'd recommend a Mac Mini in your house, but if you're giving it access to your personal files, I don't feel comfortable.

36:41 [SPEAKER_02]: And that's why I don't even know if I want to end my house. [SPEAKER_02]: Yeah, I go running the isolated things somewhere and then I'll only give you access to what you need. [SPEAKER_00]: Yeah, yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: So I installed it as a Docker container on a VAPS and that turned out to be a world of hurt.

36:57 [SPEAKER_00]: I just had trouble telling it to install skills. [SPEAKER_00]: I ended up arguing with it. [SPEAKER_00]: It was telling me that skills weren't in the folder that they were in. [SPEAKER_00]: Didn't have permission to install things. [SPEAKER_00]: So I eventually installed it natively on the Ubuntu operating system on the VAPS.

37:15 [SPEAKER_00]: I had a whole like open claw like sounds great and I'm sure it's just me like as a learning curve for me as a program I like to be able to go in and edit you know like config settings and you know see the source of truth By open claw it's a bit ambiguous on where things are there's main JSON config file but it's memory it's [SPEAKER_00]: persona, the way it operates, they're all marked down files in a workspace that it manages itself and it's a whole thing.

37:58 [SPEAKER_00]: Anyway, trying to keep this to home, this isn't focused. [UNKNOWN]: I [SPEAKER_00]: asked it to control some lights on my home. [SPEAKER_00]: This is an instance and it said, oh, sorry, I can't do that. [SPEAKER_00]: So I found, I was like, well, home, this isn't has MCP service. [SPEAKER_00]: And, yeah.

38:15 [SPEAKER_00]: Klocko can apparently talk to MCP service. [SPEAKER_00]: I have got this, it's got to be that hard, right? [SPEAKER_00]: It was hard for me to tell you, I kind of do, [SPEAKER_00]: I've been arguing with my AI on how to use MCP service too much this week. [SPEAKER_00]: I've burnt through so many credits trying to tell this thing just to how to use an MCP server.

38:34 [SPEAKER_00]: So I originally started off giving it a long-lived token with HAA MCP server at on. [SPEAKER_00]: We eventually were able to ping the MCP server. [SPEAKER_00]: I throw a bit of back and forward together. [SPEAKER_00]: only for it to find out that no it can't use a long leaf sokin for whatever reason.

38:53 [SPEAKER_00]: So I then I gave it the instructions from home assistant on how to get the OAuth keys, which is they're not real OAuth keys, it uses parts of your sub-domain or redirect URL. [SPEAKER_00]: We were able to eventually get a client ID for it to use for the OAuth flow and get a token [SPEAKER_00]: And after a lot of back and forth, I finally got it to be able to turn off my office lights.

39:23 [SPEAKER_00]: And it wasn't quick, you know, I said, you know, can you turn off the office lights? [SPEAKER_00]: And like 30 seconds later, the office lights turned off. [SPEAKER_00]: All right. [SPEAKER_00]: So it's clearly going through the motions in the background. [SPEAKER_00]: It's, you know, OK, I need to go find this MCP server, log into the MCP server, find the lights, the office down lights turn them off, and then report back.

39:45 [SPEAKER_00]: But being quicker for me to use like the raycast extension and turn off the lights. [SPEAKER_02]: Interesting. [SPEAKER_02]: And so the latency was actually that bad or that whole process was actually that bad or that slow. [SPEAKER_02]: I don't want to say bad, but that you could just or if you even had to to your point if you had a tablet, you could just go in and click.

40:07 [SPEAKER_00]: Yeah, it was like the, the LLM had to think like how do I perform this action. [SPEAKER_00]: Yeah. [SPEAKER_00]: Yeah. [SPEAKER_00]: As opposed to, yep, I know I need to call this that. [SPEAKER_00]: Now, there are skills that you can use in just all that I'm going to learn. [SPEAKER_00]: What's going to that's a whole other rabbit hole that I'm like fighting with LLM to do.

40:25 [SPEAKER_00]: And then I thought I'd get smart and introduce subagents to the mix. [SPEAKER_00]: and subagents then have their own world of pain that they're even more restricted and the main agent has to teach the subagents how to do things. [SPEAKER_00]: Right. [SPEAKER_00]: Yeah, it's it's been a [SPEAKER_00]: it's been a challenge.

40:47 [SPEAKER_00]: I thought like the, for me, this is my first use of an MCP server, like MCP to me was the go-to for like AI, so communicate with other APIs, all right? [SPEAKER_00]: But no, OpenClaw just could not stand, like, just struggled doing its thing to talk to this [SPEAKER_00]: I [SPEAKER_00]: I even, like, just before this recording, I was like, oh, I'm gonna maybe I'll try and do a live demo in this recording, I'm like, okay, let's go ahead and turn on the downloads while I'm in the recording.

41:28 [SPEAKER_00]: No, it's now giving me, like, please run this curl command on the VPS box. [SPEAKER_00]: So we can work out what's wrong with this server. [SPEAKER_00]: I'm like, copy and paste what it gives me and it's got syntax errors and extensions I installed and now we're getting into a fight about.

41:46 [SPEAKER_00]: Why is it running in Docker or is it running on the server like yeah, it's it's frustrating, but yeah, there are I do I think open claw like it's getting a lot of attention at least at the time at this recording I do think there is potential there for the ability for you know like an an agent to then run your home This is for you [SPEAKER_00]: But isn't that isn't the point of form assistance that you don't need an agent on home assistant?

42:20 [SPEAKER_00]: Thank you. [SPEAKER_00]: That was going to be my next point, right? [SPEAKER_00]: The agents, like, [SPEAKER_00]: You can do things like I can say to it like waking me up 6 a.m. tomorrow and I assume it can then go in and turn the lights on at 6 a.m. tomorrow for me. [SPEAKER_00]: But I could do that already.

42:37 [SPEAKER_00]: I could just set an alarm on my phone and the next alarm sensor. [SPEAKER_00]: In her misses will then be able to okay turn the lights on 15 minutes before. [SPEAKER_00]: I don't need an LLM burning tokens to do that for me. [SPEAKER_00]: Yeah. [SPEAKER_02]: Yeah, I think the, and you've touched on it a few times, right?

42:55 [SPEAKER_02]: The moral of the story is you kind of have to be efficient in how you use it either a, you're going to use a lot of tokens, right, which will eventually cost you. [SPEAKER_02]: And or be, it's, it's just inefficient, right? [SPEAKER_02]: It's slower, it uses more, like, again, electricity, even if it's about your electricity, somebody's somewhere, right?

43:16 [SPEAKER_02]: If it's a, whether it's an opening [SPEAKER_02]: And honestly, the net result, take a fully selfishly. [SPEAKER_02]: It's still slower, just having a automation go in and do stuff. [SPEAKER_02]: Sometimes I find that AI is, or at least AI tools are better at certain things. [SPEAKER_02]: You talked about LLM vision, for example, right to say, OK, let's actually compare these three images.

43:42 [SPEAKER_02]: And you know, it can, it, it, [SPEAKER_02]: knows a lot more around finding new ones and doing those kind of things and being like, hey, you know, if it's not exactly if the garbage bin, for example, is not exactly in this position, it's actually moved over a couple of feet. [SPEAKER_02]: Okay, well, the garbage bin is still in the frame, so you can still assume that it's a garbage bin, right?

44:03 [SPEAKER_02]: So those kind of things [SPEAKER_02]: And at least start, maybe not really good, but good enough at, right, as I do doing some kind of thing locally, right? [SPEAKER_02]: So, but for other things like you said, like, again, like open cloud for running our open cloud for running a full assistant, I don't know, yeah, much of a point.

44:28 [SPEAKER_00]: I think it will become, there is a use case for it when maybe you're out and you go, hey, can you just check that everything off is off at home or maybe you want to query things like, how much electricity have I used today or can you compare that? [SPEAKER_00]: What's my trends? [SPEAKER_00]: Like if there's asking deeper questions, maybe there's a use for it.

44:52 [SPEAKER_00]: At that moment, like yeah, just, [SPEAKER_00]: having it to have the ability to turn off and on things is great, but it's really slow at doing it. [SPEAKER_00]: And that's, yeah, I guess maybe I just need to help a bunch more training. [SPEAKER_00]: It's also not as a program, it's not intuitive for me, like I'm not used to having to talk to something to teach it, how to do things, like for at least a system, right?

45:20 [SPEAKER_00]: Like I [SPEAKER_00]: access like an omniscent webbook I set up specifically for it. [SPEAKER_00]: It was talking to the agent. [SPEAKER_00]: Okay, you're going to make a post request to this address. [SPEAKER_00]: Here's the JSON structure you need to do. [SPEAKER_00]: As opposed to just defining that somewhere in a config file, I needed to talk to the LN for it to then go and create those scripts and instructions somewhere.

45:47 [SPEAKER_00]: Um, my, with that said, I do have some plans for OpenClaw. [SPEAKER_00]: I would, um, my plan is to create some, uh, assistance, um, special assistance. [SPEAKER_00]: One's going to be at least at the moment. [SPEAKER_00]: I'm going to have like a meal plan or assistant. [SPEAKER_00]: Um, um, um, um, um, um, um, um, um, um, um, um, um, um, um, um, um, um, um, um, um, um, um, [SPEAKER_00]: planning meals, you know, my daughter's started school.

46:08 [SPEAKER_00]: We've got a whole bunch of meal prep we need to do now like lunches and getting home from work and dinners and all that. [SPEAKER_00]: I want to try and get more efficient with that. [SPEAKER_00]: If I can offshoot some of that mental load to an alienation, that will be great. [SPEAKER_00]: I would like to have a planner.

46:24 [SPEAKER_00]: or a scheduler, sub agent as well, now that we've got two kids at different places, you know, two both parents work each whole time. [SPEAKER_00]: Like it's a mission to work out who's at home, who needs to pick up drop-off. [SPEAKER_00]: If I can have an AI, like sort of sync your calendars and work out, who's going to be where who needs to be home, who needs to be on pick-up drop-off, that will take a lot of [SPEAKER_00]: planning away from us.

46:53 [SPEAKER_00]: We'll be able to get there. [SPEAKER_00]: I don't know, but I'm trying to don't try at least it make your attempt. [SPEAKER_00]: Yeah. [SPEAKER_00]: That's cool. [SPEAKER_02]: I mean, yeah, I like that idea of, I think that's also like in some whole automation realm that is a, I don't know, I don't want to call it the Holy Grail, but it's like one of the [SPEAKER_02]: Things that people want to really get to, right?

47:20 [SPEAKER_02]: It's just a say like, hey, like, plan this for me, right? [SPEAKER_02]: Where am I going? [SPEAKER_02]: What am I doing? [SPEAKER_02]: How am I doing this? [SPEAKER_02]: Who's your point? [SPEAKER_02]: Who's getting a kid? [SPEAKER_02]: Who's, you know, doing that kind of thing. [SPEAKER_02]: And open, or yeah, open claw might be good in the sense that it can access more things outside of your system, right?

47:41 [SPEAKER_02]: Like, and yeah, I don't know depending on how it is, how it looks, how it works. [SPEAKER_02]: things like to a fair and stuff like that, it might be able to handle. [SPEAKER_00]: Well, yeah, you can do an access to like your one password if you're that trusting with it, right? [SPEAKER_00]: Right.

47:55 [SPEAKER_00]: If you put something on password, then where it goes. [SPEAKER_02]: Exactly. [SPEAKER_02]: Exactly. [SPEAKER_02]: I'm not that trusting. [SPEAKER_02]: Nobody or nothing gets access to, you know, a password manager for me. [SPEAKER_02]: It's uh, unless there's a really good reason for it, right?

48:12 [SPEAKER_02]: And uh, [SPEAKER_02]: But yeah, so I don't know, there is a lot of cool things you can do it, right? [SPEAKER_02]: Like like what you said, but yeah, I don't know if you need open cloth for that specifically or even just like an MCP or sorry, not an MCP, just even like a chat interface where you just go in and say, hey, what am I eating today?

48:35 [SPEAKER_02]: And it goes in and crunches whatever, you know, hey, let's go look into, um, [SPEAKER_00]: No, look at grossy or grossy, yeah. [SPEAKER_00]: And if you find out, you know, we can do favorite recipes. [SPEAKER_00]: And then go see what items you've got in stock. [SPEAKER_02]: I would be cool to take it and say also like give me like nutrition information of this, right?

48:58 [SPEAKER_02]: Like I wanted to be healthy today and, you know, or whatever, maybe you have some kind of like diabetes is a perfect example, right? [SPEAKER_02]: Like hey, make sure this is, [SPEAKER_02]: friendly for an individual who lives with diabetes. [SPEAKER_00]: Yeah, and that's what that's a plan for my culinary expert on Colin Curtis, it's going to be my chef for the family.

49:25 [SPEAKER_00]: And his goal, one of his tasks I'm going to give him is to give me a shopping list at the end of the week that I need to go. [SPEAKER_00]: have to tick off his meal plan for the week, right? [SPEAKER_00]: Will we stick to that meal plan most likely not? [SPEAKER_00]: Will I get it over the line with the rest of the family?

49:39 [SPEAKER_00]: Probably not the kids will hate it. [SPEAKER_00]: Um, but at least it will live in attempt, right? [SPEAKER_00]: Um, because yeah, I know that when it comes to 5-3 in my car, crap, what's for dinner tonight, right? [SPEAKER_00]: It's all right, just doing this. [SPEAKER_00]: Um, and then Groundhog Day begins the next day, right, with what's the breakfast?

49:58 [SPEAKER_02]: Great. [SPEAKER_00]: So that, then my missions may go to have another one for proactive security, maybe something doesn't look right at home, maybe the instances gone down, it can't be accessible from open floor, maybe it goes like, I can't access home, this is not anymore, what was the last event that I received from home assistant, did I receive any notifications from LLNVsion about someone walking up to the house, maybe they've cut the power and broken in.

50:28 [SPEAKER_00]: Yeah, there would be some other ideas. [SPEAKER_00]: But I will also open to other people's suggestions if you've got some ideas on what you're planning to do with either OpenClaw or LLMs in your smart home. [SPEAKER_00]: I'd love to hear it. [SPEAKER_00]: Other than you should comment, so send us an email, feedback, and homes, do what I found, or would love to hear what you're going to be using AI for, or what you're hoping AI can solve in your home.

50:52 [SPEAKER_00]: Yeah. [SPEAKER_00]: I think that's all I had around AI and how I'm using it unless you've got any others bits here using it for [SPEAKER_02]: No, no, it's, I'm excited to see what comes from it. [SPEAKER_02]: Again, I, I'm on both sides of the fence. [SPEAKER_02]: I'm not a naysayer for AI, but I'm not also like, hey, like again, I use Cloud Code of Con in my personal life just for building out apps to solve specific problems and those kinds of things.

51:20 [SPEAKER_02]: Yeah. [SPEAKER_02]: In my home automation, I would love to see some of that stuff and I'm interested personally just to check out what you end up doing with your with Curtis [SPEAKER_02]: You know, it's those kind of agentic style, I guess, agents which we've already used it. [SPEAKER_02]: But, uh...

51:40 [SPEAKER_02]: Yeah, I'm excited to see where the technology goes. [SPEAKER_02]: Again, I just, we need to get to a point where it uses a lot less power and costs a lot less. [SPEAKER_02]: And that kind of stuff. [SPEAKER_02]: Because I know I've seen a lot of things on YouTube and stuff and a couple of friends that I know too.

51:58 [SPEAKER_02]: It says, Open Cloud can shoot through your tokens in no-time. [SPEAKER_02]: And there's ways to give it local memory and global blood. [SPEAKER_02]: Do a bunch of stuff to save on that, but it's still not cheap. [SPEAKER_00]: No, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no [SPEAKER_00]: Alright, let us know if there's another topic that you'd like us to cover to in sort of these episodes as well.

52:26 [SPEAKER_03]: Yeah. [SPEAKER_00]: We want to try and share what we're doing with homestead moron and give things sort of a homestead and focus as well. [SPEAKER_00]: You know, like AI specifically today was just about homesteads but maybe there's something else that we can try to lean and focus in on homesteads as well.

52:40 [SPEAKER_00]: So let us know in the comments as well. [SPEAKER_00]: Yeah, give us a show. [SPEAKER_00]: journey or come on as a guest. [SPEAKER_00]: Reach out to us at feedback at haskpodcast.io That's H-A-S-Spodcast.io The home assistant podcast is hosted by Phil Hawthorne and myself, Rohan Keremandi.

53:09 [SPEAKER_01]: For links to topics we discussed today, check out our show notes on haskpodcast.io