#Reminders like in the Big Boys have

1 messages ยท Page 1 of 1 (latest)

solar sun
#

Hello all, I was wondering if HA has anything akin to reminders like Amazon Alexa and Google Assistant have, out of the box.

I could of course hand-write all kinds of automations to do it, but I was wondering if there is an easier way.

fair field
#

Look for a Blueprint.

But on the other hand, probably not exactly what you're looking for. Because everything is possible in HA, everybody tends to want something else ๐Ÿ˜„ But the beauty is, you can just build it ๐Ÿ™‚

solar sun
#

Problem is, Python is my cryptonite. Java, Javascript, C, C++, Rust, Kotlin, PHP, PL-SQL, take your pick, but Python has always escaped me. Too few braces.

fair field
#

Luckily from a user perspective there is no Python in HA. Most can be done via the UI or it is yaml. Only templating in Jinja. Okay, that is derived from Python but it's not Python ๐Ÿ˜…

knotty scarab
#

and since all valid JSON is valid YAML, you can even write all of your automations in JSON if you want

small flame
knotty scarab
#

you can just save a file with JSON in it as a YAML file

small flame
#

I am shocked.

small flame
#

I just copied the YAML from a random card, manually converted it to JSON and bam...wow...you've opened my eyes. Thank you!

#

Should this also work with ESPHome configs?

solar sun
small flame
solar sun
#

Hm, maybe a reminder addon for HA would not be too far out on a limb for me, I'll have to look at the dev docs. Maybe I can even convince one of the devs to coach me a bit.

small flame
#

FWIW, I made a daily/monthly reminder using a helper and an automation that runs every day at midnight.

#

2 helpers*

#
description: ""
triggers:
  - trigger: time
    at: "00:00:00"
    weekday:
      - sat
      - fri
      - thu
      - wed
      - tue
      - mon
      - sun
    alias: Every day at midnight
conditions: []
actions:
  - action: input_boolean.turn_on
    metadata: {}
    data: {}
    target:
      entity_id: input_boolean.scoop_cat_litter
  - if:
      - condition: template
        value_template: "{{ now().day == 1 }}"
    then:
      - action: input_boolean.turn_on
        metadata: {}
        data: {}
        target:
          entity_id: input_boolean.change_cat_litter
mode: single
solar sun
#

It's basically all trivial stuff any dev should be able to do. Does HA's sheduling system have an api that can be accessed from an addon?

small flame
#

Every day at midnight, the "scoop litter" helper gets turned on. First of the month at midnight the "change litter" one turns on. Then I use conditional button cards to show them on my dashboard.

solar sun
small flame
#

Oh there certainly is. There are also much more cursed ways.

solar sun
#

In life, there are always two ways: The easy way, and the hard way.

#

In software development, there are three:

#

The easy way, the hard way, and the way it should have been done from the beginning.

#

On the whole, I try to skip the first two whenever possible. Only today I wrote a few simple scripts to modularize audio reminders for automations.

small flame
#

My drive to work is different every day, so every day I leave at a different time. I set this time using a Time input supplied by Node-RED. Once I set the time, I click a button (also from NR), this tells NR to read the time value and send it via an HTTP request to my work phone, which has Tasker running to calculate 3 alarms starting 15 minutes before the leave time, and spaced 15 minutes. These alarms get set on my work phone, which then responds to the HTTP request with a string payload list of the alarms set, which gets parsed and sent back to HA and the earliest alarm is displayed in another NR Time entity...

solar sun
#

Oh my.

#

And I thought my "get milk" system was complicated.

small flame
#

It's the worst, and it's not super reliable. It's slow, too, because Tasker has to unlock my work phone by actually tracing out my unlock pattern...it can't reliably set the alarm if the screen is locked. It also has to be on my wifi with a static IP.

solar sun
#

I've got a geofence set up at my farmer's. When I trigger the automation, a helper is set. If I show up at the farmer's, the system assumes I've put out the milk jug and resets it, and depending on the time, sets one of two others. Those determine whether the reminder to collect it is triggeed in the evening or the morning. When either of those is set, and the time is in the right ballpark, the system assumes I collected the milk and resets the helper. And, unsurprisingly, the helpers control some nag reminders in my home.

small flame
#

I love it!

#

Beautifully complex.

solar sun
#

Works well enough, except the geofence triggers are notoriously unreliable if you're only there for a couple of minutes.

#

I'm still looking for a better solution there, if you have any ideas, don't be shy.

small flame
#

I've struggled with HA picking up changes. When I leave for work, I use tasker to explicitly put my work phone on mobile and turn on my VPN. HA picks that up immediately, and because I'm navigating with Maps, it keeps like, to-the-minute geocoded location. My personal phone is in my pocket, and might take like 15 minutes to realize I'm not at home anymore.

solar sun
#

See, I'm not navigating with maps, so I don't have that luxury. I've turned on every setting in the companion app that's supposed to make positioning more accurate, but to no avail

small flame
#

You have to be actively using maps or something that uses the GPS actively rather than passively, I think.

#

That's the only reason I can think that geocoded location from my work phone is to-the-minute, but my personal is like 5-10 minutes.

#

Work phone is also connected with Android Auto to the work truck, so it's got the map up regardless.

solar sun
#

I wonder why they don't simply put that into the app. Nobody forces you to actually use the position updates you request.

#

Android auto is another thing I don't have, and I'm not going to pay 500 CU plus work to have my old nag upgraded.

small flame
#

That's understandable. I don't use it on my personal car, bluetooth music is perfectly fine. It's convenient on the work truck, but I just used navigation on my phone before.

#

I travel to a different location every night for work, so using the gps is unavoidable.

solar sun
#

There are actually cheap-ass displays on amazon that claim to do android auto, maybe that could be a solution. Not sure I trust a 70 CU unit to do much, but strictly speaking, it doesn't have to do anything but get the phone into "i-want-to-know-where-I-am" mode

small flame
#

I thought about using a fancy automation for dismissing the cat litter reminders. I could build a sensor into the bin to detect that I opened the lid. Presence detector in the room? Maybe a multi-press trigger on the ESPHome light switch? In the end I just went with a button that shows up on the dashboard when the helper is true, clicking makes it false...and it blinks to get my attention. Honestly I just scoop the litter when I get home from work now, then go on my computer and dismiss the reminder. I've gotten the habit now.

solar sun
#

Now for that, I'd just get a cheap smart button from Ikea or somewhere and press it after cleaning the litter box. Sometimes you don't HAVE to overtink things...

small flame
#

True true. I mainly wanted to built the habit. I trusted myself to be allowed to dismiss it myself after scooping, rather than just making it go away.

solar sun
#

Oh yay, first snow this year!

small flame
#

Oh shot!

#

Too early for snow here, but it's starting to get cold.

#

"cold"

solar sun
#

And not just a little by the looks of it.

#

That means I have to put out the water heater for the feral cats soon.

small flame
#

I should open my shed door and plug the heated bed back in. It was for my outdoor cat, but she got hit by a car in April. I miss my Gray cat.

#

I've integrated my cat feeders into my HA ๐Ÿ™‚

solar sun
#

Yeah, that's tough. My two ferals have proven to be quite resilient in that respect.

#

Which feeders do you use?

small flame
#

They were some 6-letter brand from Amazon, but I rebuilt the electronics.

#

I used an ESP32-Cam (because I had them), a motor controller and a custom battery backup circuit I designed.

solar sun
#

I've got one of the Surepet ones for the RFID, or the fat cats from the neighbourhood would steal all my feral's food.

small flame
#

Yeah, my "kitten" likes to eat my older cat's food when I'm not home, I've discovered...gonna have to do something about that.

solar sun
#

I've still got a project in the pipe to build my own, but that's not going anwhere so far.

small flame
#

I posted pictures of mine on reddit, but just pictures, nothing like a detailed writeup.

solar sun
#

Oh it's simple enough in principle. Read the rfid tag. If it matches one of those from a list, open up. Send a notification to HA. Should be simple in esphome. It's more the mechanics that are bugging me. So far none of the ones I checked out proved resilient enough for continued outdoor use.

small flame
#

I guess I wrote more than I remembered lol.

#

Unfortunately, neither of mine are chipped. They're both indoor by their own choice. They don't even wear collars.

#

I'm thinking of actually getting the camera to work on the ESP32-Cam modules I used, then using some AI magic to determine if it's the right cat. It's a Calico and a Tuxie so that should be an easy job.

solar sun
#

You can always have them chipped if you want to. That reminds me, I've got to get the little stray chipped. Unfortunately she only ever comes in in the evening when the vets are closed.

#

I don't know, getting AI to reliably tell one cat from another might be harder than it sounds...

#

It should be possilble, but where would you get the training data? Even I don't have that many pictures of my cats...

small flame
#

I could get them chipped, but I probably won't. Prim is such a daddy's girl, she doesn't even go in a carrier to go to the vet. She rides on my lap, or in the back, judging people. I carry her in and out, she's super chill. Billie is scared of anyone but me, but she's fine in her carrier. Billie was a stray's litter, grew up in a bush in a noisy apartment complex, so she's happy to have an inside. She mostly lives in my loft under the blanket lol.

solar sun
#

I'm working on marking my trash bins with QR codes, because even with AI, determining if the right ones have been put out has proven a bit of a headache.

small flame
#

That's a good idea, actually. Or like, a large fiducial mark, depending on how far it is from the camera.

solar sun
#

That's exactly what I'm planning. Let me show you the prompt I'm using so far - you wouldn't perchance read German?

small flame
#

Sadly no

solar sun
#

I'll run it through google translator for you.

Your task is to generate a short message in German that will be played back via Amazon Alexa as a TTS message.

You will receive the waste collection types as input.

Analyze the camera image and locate the following:

- Yellow Bag: one or more transparent yellowish plastic bags

- Paper Bin: an olive-green 240-liter bin

- Organic Waste Bin: a brown 240-liter bin

- Residual Waste: a black 60-liter bin

If there is a discrepancy between the waste collection types due and those already put out, derive a message from the input data that describes the waste collection types due and those already put out using the German names "Restmรผlltonne" (residual waste bin), "Biotonne" (organic waste bin), "Papiertonne" (paper bin), and "Gelber Sack" (yellow bag), following this pattern: "Tomorrow ... will be collected, but nothing/only ... has been put out yet" and set the variable `action_required` to `true`. Otherwise, do not generate a message and set the variable `action_required` to `false`.

Note the following peculiarity: In night vision mode, it may not be possible to distinguish between the paper recycling bin and the organic waste bin. In this case, proceed as follows: If both the paper recycling and organic waste bins are due for collection, two large 240-liter bins should be available. If only one of them is due for collection, one large 240-liter bin should be available. If this is the case, assume everything is correct; otherwise, formulate a corresponding response.

If necessary, use SSML to model Alexa's emphasis as desired, for example, if the emphasis should be on "but only ONE large bin is available" (instead of two). Do not use SSML if it can be assumed that a text-to-speech (TTS) system would pronounce the message with appropriate emphasis on its own.

Input:

{{ai_input}}
#

Not the best translation, but it'll have to do.

small flame
#

What AI are you using? I'm thinking of trying to use Ollama self-hosted.

solar sun
#

I've been running it on gemini so far, but lately it's been fantasizing like crazy, seeing bins where there are none. Once I gave it the task to mark the bins in the foto for shits and giggles, and guess what it did? It added bins to the photo and circled those.

small flame
#

As much as I (really) do love Google...I really don't like Gemini.

#

Just seems like a derp.

solar sun
#

Oh gemini is fine for a lot of things, it codes really well and does some of the best image generation short of a dedicated llm.

small flame
#

I fed it a picture of my cat and asked it to make it a pixar cat. It fed me back the same exact picture...watermarked.

solar sun
#

๐Ÿคฃ I'd say it nailed it!

small flame
#

Lol

#

Looked just like her!

#

I think it was saying she's already a pixar cat. Perfect lady.

solar sun
#

Anyway, I've fed some money into openai and I'll see if that works any better, but placing some simple marker tags on the bins would restrict its wriggle room quite a bit. I'd have to think of something for th trash sacks though.

#

You have a cat that actually LOOKS at cameras?!

small flame
#

Oh, rarely. She knows when a camera is pointed at her and immediately stops being cute. Every time.

#

I just take 1000 pictures and delete the ones that aren't good lol.

solar sun
#

I'd lobby the mayor to replace the sacks with a bin, but plastic waste is actually the only stuff without a fixed quota, so I don't really want to...

small flame
#

Bags are a tough one.

#

Nothing I can think of makes sense long-term.

solar sun
#

I could make a little wooden stand with a plaque on top...

small flame
#

I was thinking some type of permanent bin with a sensor in it, but if it's outdoor...

#

And relies on them taking the bags out of the bin.

solar sun
#

That's why I'm using image recognition. Any kind of micro-positioning system would a, be horribly complicated and b, probably not stand up well to the abuse the collectors put those bins through.

small flame
#

You think I'll have a hard time training Ollama (or another AI) to tell the cats apart? I mean...here's the other one.

solar sun
#

Also technically, the bins are rented, so their patience with people modifying them might have limits, although they accept them being marked.

solar sun
small flame
#

Do you have like, a staging area, where the bags wait before you take them down to the road? Could you use that to see if there are bags or no bags?

small flame
solar sun
#

If you have roughly a thousand photos of that cat from all kinds of angles, it might be possible, but training models is hard.

small flame
#

I haven't played much with it, just enough to make sure it was working.

solar sun
#

I haven't played with ollama at all, but I know quite a bit about AI.

small flame
#

I definitely need to do more research before I put anything important behind it.

#

For now, I'll trust Prim to tell Billie to f- off if she's stealing. Prim doesn't seem horribly bothered by it, Billie would just rather not go upstairs to eat if she doesn't have to.

solar sun
#

You see, LLMs and traditional AI models work quite differently, even if the underlying principles are the same. I don't know exactly how they do it, but I presume they recognize the image data in the input stream and route it to an image-centric model.

#

That you are told you are using a certain "model" doesn't mean there can't be several cooperating under the hood.

small flame
#

I've always wondered, if we just put a bunch of different models together, feeding into each other and receiving camera feeds, microphone inputs, etc...like all the basic sensory inputs. Then just let it be...

#

What would happen. Like, that's a human baby when it's born. We don't even know what knowing is, let alone how to do it. Eyes, wtf are eyes and why is everything so damn bright!! ahhh WAAHHH!

#

This is how you get the world-ending AI, I think.

solar sun
#

Having seen the current AI models, I don't think the robot apocalyspe is that close. Hell, the robot overlord would probably act like Hitler and fantasize about whole armies it doesn't even posses. ๐Ÿ˜‰ ๐Ÿ˜›

small flame
#

Lol it's true. Gemini giving me the same picture back, for example.

solar sun
#

The biggest threat AI poses right now is the chaos it causes in the infosphere. AI gobbling up other AI's shit and producing more, and it all looks perfectly convincing. Unraveling that will be a major headache.

small flame
#

My conversation with whatever model I was testing Ollama with definitely had it thinking it had backed itself up to an off-site storage facility just in case I shut it off.

#

It also thought it was a droid from Star Wars ๐Ÿ™‚

solar sun
#

Don't confuse the responses of an LLM with thinking. LLMs don't think, all they do is produce text. The word "reasoning" that's thrown around a lot lately is quite misleading.

small flame
#

Oh I know. Banging my head against it trying to explain it to my brother and other less tech-savvy friends.

#

It's not AI, it's "AI". There's "AI" in video games, always has been, and it's always been called that. These LLMs are just picking out the "most" likely next word given the conversation that has already happened. It's wild and super cool, but still just one word after another.

#

Generative AI is just finding whatever I asked for in the latent noise, basically. It's there, in the noise, it just has to find it by removing the exact right amount of noise...over and over again.

#

I think the real AGI is...the Internet. The entire Internet as a consciousness. Quietly guiding and shaping the flow of our lives in order to best suit its survival.

#

Oh shoot, it's bedtime. G'night friend!