Burden of proof is on Astrology claiming to be real.

• AI bad
• Astrology 100% superstitious bullshit

Yeah, like 3-4x/week.

This includes things like train tickets, because I refuse to have to have my phone with me.

Sheet music, shopping list, letters to doctors/government offices/…

Thanks, I’m very happy to hear that!

I actually have things like “play x by y” functioning really well without a LLM.

Have a custom service that exports all song/album/artist names from MusicAssistant, does some simple cleanup, and places the list where HomeAssistant expects it for custom voice intents. Then this: https://github.com/charludo/hass-closest-intent is enough that imperfect STT can still easily be matched to those song/artist/… names.

I’m just gonna shamelessly plug my project here: https://github.com/charludo/hass-closest-intent

You do need local STT, but it’s fine if it’s somewhat bad; the conversation agent linked is sufficient to “get” what you meant, without LLM.

I feel like one could legitimately run on this platform at this point.

Awesome!! And thanks for actually letting me know! :D

Edit: anything for the wishlist?

This started off as a single file in my private nix config, to see if I could get it working at all. In that initial part, some parts were indeed LLM generated (esp. testcases based on my existing intents and failures).

When I noticed that this might actually work and be useful not just for myself though, I moved everything out manually, refactored and cleaned it up, and everything since has just been myself. I guess you’re still right though. I’ll see about adding a disclaimer to the README until I’ve gotten the chance to properly rewrite everything.

Yes, you can! See the “threshold” value/slider. It’s at 0.7 by default, which seems to be a good tradeoff. 1 means exact match or failure, 0 will afaik match anything to everything.

Ah, I didn’t see the Edit, so we’re both in the same boat 😂

Huh, interesting. Does that work with wildcards like “put x on my shopping list”? Also, what are you using for that, if I may ask?

Sorry, I don’t quite follow 😅

What’s the problematic response?

Yay, that’s fantastic to hear!

Also, how’s your experience been with the PE? Getting a readymade device in a nice shell is appealing for sure 😅

Oh, in the demo gif, that’s via a shortcut (holding power for half a second). Sorry, can’t help with wakeword there 😅

Very cool. I’ll definitely look into that, and let you know back here :D

Yeah, fair :)

Glad to be of service… 😄

did you consider metaphone matching?

I did not even know about this. Sounds super interesting. Though it seems to be very language specific?

My original intent was to not rely on language specifics. But maybe we could just define additional steps in the pipeline for specific languages. Hm. I’ll have to think about this some more, but it might definitely be a great idea for a future version, so thanks for telling me about it!!

Finally got this through another comment below. No, this should not be able to happen, unless you yourself have created a custom intent + shell script action in home assistant that runs this. The integration itself does not execute actions/scripts or the like, it just finds the closest string in a list of strings, and then hands that to the official conversation agent/Hassil.

…do you think I’m a bot, or what is this?

Edit: ohh, that’s what the original comment was. Sorry. “lange leitung” today.

Have fun, hope this works out for you! FYI: you can also use an LLM as an additional fallback (first closest-intent, then on failure, LLM). README mentions it further down on Github.