AI-written message describing movement detected by a camera

[Archive] Feature request
,
21

No ^^ Not yet
The parts are good though ^^

image
image384×367 8.39 KB

22

Haha!! Better but not perfect (and a longer response):

image
image381×359 8.38 KB

image
image267×290 5.09 KB
image
image257×430 6.22 KB

1 Like
23

Is it possible to store the text in a variable so that Sonos can play it directly ??

24

Not right now, but that could be a nice evolution :slight_smile:

However, in terms of cost, on setups like yours it’s quite expensive to really send everything on each request

So basically, for the AI to be able to have the information about your whole home, I have to send the AI all the values from your sensors, and that creates a huge chunk of data on your side!

If I keep the same limit as currently (1000 requests / month / account), that exceeds the budget

Do you think we could send only certain devices?

25

Question

Wouldn’t it be interesting to select a request type within the scenes.

In the case of sensor retrieval, send an initial request with the appropriate context that returns only the categories/types to take into account and then send them back to it

Because indeed, sending the entirety of the sensors with every message
 I understand that becomes costly ^^ it necessarily has to be done in several steps.

Maybe another feature later for the sensors, right?

26

That’s the approach that’s already being used right now if you request a temperature.

Here, the idea was really to give the AI the full view of your home to allow the AI to answer very complex questions like:

« The alarm went off at my house — do you see anything strange at my place? Â»

The idea was really to allow the AI to answer open-ended questions, not just to respond to requests (we already know how to do that)

Still, it might be a bit premature, especially since we can see the AI hallucinates quite a bit, so it’s not very conclusive for a fairly high cost.

I think I’m going to remove this option for now

1 Like
27

@Terdious I rebuilt a Docker build without sending the devices!

Available here:

gladysassistant/gladys:ask-ai-in-scene

Do you see anything else that would be essential for this first release?

1 Like
28

Great !!

  • Implementing switches in the same way as lights would be great

  • being able to make Sonos speak the response, but like you said that can come later ^^

I think the current implementation is a good thing !! It’s just that we’re almost frustrated at not having more. But one thing at a time, and your usual approach of building small pieces step by step is, I think, a very good thing.

I, who wasn’t convinced at all at the start, am now pushing the thinking in that direction in my personal project. In the end it moves much faster and it’s a comfort for coding/ review

A big thank you to you for that :grinning_face:^^

29

Thanks for your feedback!

Regarding switches, I just looked and it’s a topic in its own right, so it’s outside the scope of this development. We can go into detail in another thread :slight_smile: I invite you to create it if you’re interested!

Making the Sonos speak isn’t that complicated but I think we can keep that for a later time. I’d welcome a feature request as well to gauge interest!

I think we’ll deploy this first development as is to get more feedback, thanks again for all your testing :folded_hands:

1 Like
30

This feature is available in Gladys Assistant 4.48:

I’m closing this thread to free up the votes!

1 Like
32

Requests created!

1 Like