Thanks for the app @Arve_Bjornerud !
Been using it for a while now snd it works way better than my google home has ever performed. In the long run, im looking at replacing the googöe home, but in the meantime would it be possible to get this app to send commands to a spotify connect device or a chromecast device?
Hi. Thank you for the kind words 
Spotify and Chromecast integration is interesting. Everything is already in place in the App to extend it with new functionality like this. The only question is if there is an open API i can use to interact with these services. I’m thinking most probably there is for Spotify and most probably not for Chromecast. I’ll look into it 
@Arve_Bjornerud thanks for the app, it’s exactly what I’ve been hoping for.
I wish to run local voice AI.
I have rooted an Amazon Echo Dot 3 Gen3 and installed GitHub - Gamer92000/echo-dot-assist: Turn Amazon Echos (Echo Dot 3, Echo Dot 2, Echo 2) into local Home Assistant voice satellites. Keeps Amazon's mic processing and wake word engine, replaces Alexa with a small C daemon speaking ESPHome or Wyoming. Multiroom music via Music Assistant, Bluetooth speaker and proxy, no cloud. · GitHub on it.
After that I used a custom pipeline in your AI Voice Assistant app.
Everything is running in docker containers on Latest Debian (5700g, 32gb, with an nvidia 3060 12gb gpu)
speech-to-text: is (openAi compatible custom url) speaches-ai/speaches:latest-cuda
text-to-speech: is Wyoming-Piper (Local) rhasspy/wyoming-piper
Language: I have tried both LMstudio & Ollama with Qwen 2.5:7b and Qwen3:8b
All 3 custom inputs (STT, Language, & TTS) test correctly (i.e green), However…
The echo dot hears me, activates the echo in homey but NOT the device being switched on or off, and then answers me that it has switched the device on/off .(i.e it seems to work but actually fails)
I tried some of the other options as well… (Mistral & Gemini Live).
Mistral hears me but then does nothing - If I try Mistral as custom Url (I was going to use it just for the language part) it clears the test for both STT & TTS, but when I try to use it for language it fails.
Finally Gemini Live works perfectly… It hears me, and switches on the device, and then responds.
I haven’t tried OpenAI.
I really would like to get local language going… Is it failing because the models I’m using for my weak gpu aren’t good enough? I’m new to AI, so don’t really know what to expect from local processing..
Or am I just using the wrong models? or not tweaking them correctly.
Any help would be greatly appreciated. Thanks
@Arve_Bjornerud Hi! Would it be possible to make AI Voice Assistant work without requiring a physical voice device?
My use case is mainly AI-generated announcements in Homey Flows.
For example:
Dishwasher finished → generate a message → convert it to a natural OpenAI voice → expose the audio URL → play it on Sonos.
So I don’t necessarily need a microphone or ESPHome voice satellite, only the text-to-speech / audio generation part.
Maybe this could be added as a virtual device, headless TTS device, or an output-only mode?
That would make the app useful for Sonos, Chromecast and other speakers even for users who don’t own a dedicated voice assistant device.
Ideally the Flow cards would still provide something like:
For me this would be perfect for dishwasher notifications, weather summaries, goodnight messages and other Homey announcements.
@Mariomark1541 not the voice part, but for the text part of that: I made Grok Connect, which can write the message inside a Flow.
Something like: dishwasher done, then Ask Grok “Write one short, friendly sentence that the dishwasher is finished. Max 15 words, no emoji.”, then pass the Answer tag to whatever text-to-speech card you already use for your Sonos.
It won’t give you an OpenAI voice or an audio URL, so for that part you’d still need what you asked Arve for. Homey Pro only, and you need your own xAI API key.
Sorry for jumping into your thread, Arve. Just thought it might cover half of it in the meantime.
@Mariomark1541 Hi! Good idea, and most of the pieces were already in the app. I’ve added three Flow cards that belong to the app itself, so you don’t need a voice device:
Hand the URL to Sonos’ Play a URL card, or to any speaker that can play a URL. For the dishwasher, the Flow would be: dishwasher finished → Ask assistant and create speech audio → Sonos Play a URL.
The AI engine and model come from the app settings rather than from the card. What would you want to switch per Flow, and why? If there’s a real need, I’ll look at it.
It’s on a branch for now and hasn’t been tested on a real Homey yet (on vacation):
I’ll test it and put it in a release soon.
Hi @Dee_M-S, thanks for the detailed write-up!
A log will show what’s going wrong. Could you say the command once more with the custom pipeline (LM Studio or Ollama), then right after:
The file includes your engine and pipeline setup. API keys and what was said are removed automatically, and the file is deleted after 30 minutes, so grab it straight away.
When it says “done” but nothing happens, the model usually answered with text and never actually called the device tool. The log will show whether that’s the case.
