I got this app but I can’t get it to work with my local middels, maby because I got about 15 different models.
It should work, even with 15 models. Have you configured the app in More->Settings->Ollama in the mobile app?
Having different models/providers has always been the plan. The main issue with the code is getting data to flow fast enough between the PE, Homey and the LLM. ( PE → Homey → LLM → Homey → PE). OpenAI had the easiest (at the time) API to do real time streaming, so that’s why i went with that. My plan is to have the “easy” option, where OpenAI is the default one click solution. They have Speech-To-Text, LLM and Text-To-Speech in one single “call”. But also a more advances option where you basically choose what service to use for each step. So for example, you could do all locally, or some locally and some at a third party. But again the issue is data flow speed. If you do the individual steps Homey will have to do a lot of data routing. Example:
PE --(byte steam)–> Homey --(byte steam)–> Local-Wyoming-whisper --(text)–> Homey --(text)–> Local-llm --(JSON)–> Homey --(text)–> Local-Wyoming-Piper --(byte steam)–> Homey --(byte steam)–> PE. Then at the same time, this isn’t one sequential flow of data, data has to flow from the PE to the LLM and back while one is actually speaking. That way one doesn’t have to wait so long for the response to come back once your done speaking. I really want this App to work, it would be so cool. I have trained a local Text-To-Speech model that i plan to use for my own house, based on the voice on my King Harald here i Norway which has a very iconic voice. Where i give the LLM a prompt to speak like the king does.
I’ve been refactoring the code a bit now. I’ve found a couple of bugs and so on. I think I’ve solved the issue with the old/new PE firmware. Will try to push a new test release soon.
is ThirdReality Voice & Music Assistant Dev Edition also supported? - It seems to be compatible with / close to Home Assistant Voice PE.
But any AI with a OpenAI compatible API could work?
Mistral’s API should be compatible with OpenAI from what I found (see e.g. “Access Mistral Using OpenAI-Compatible API”, Access Mistral Using OpenAI-Compatible API ). Because for me - and I think many others - the high european data & privacy protection standards really matter it would be great if you could have a brief look into that.
I’ve just updated to v.1.2.0. These are the changes:
-
Weather answers — ask about the current weather and forecast for your location. -
Knows the time and place — the assistant now uses your local time, time zone, and location, so its answers fit your situation. -
Smoother, faster conversations — quicker replies and better memory of what you just said during a chat. -
Better speech understanding — improved voice recognition and updated AI models for more accurate responses. -
More reliable smart-home control — steadier control of your devices and rooms by voice. -
Works with the newest voice hardware — support for the latest Home Assistant Voice PE firmware (2026 and up). -
Reliable device pairing — fixed a bug where devices wouldn’t show up when adding them a second time; they now appear every time. -
Stability and maintenance — updated the app’s foundations (newer Homey platform) for better reliability and security.
Oooh, nice! Yeah i think that should work. Looking at the device it costs a bit to get into Norway, might order one anyway. Will let you know.
The image looks like it. Should be the same as Voice and Music Assistant Dev Edition – ThirdReality
At least in HA it can be used as a substitute for Home Assistant Voice PE. - I am currently looking for apps & devices to Build a local (or if cloud then at least european)TTS Pipeline in Homey. Trying to avoid going to HA for that.
I don’t know if the speaker uses the same server/mobile app, but the ThirdReality Smart Hub Gen2+ that I use is cloud-only. As far as I know, all ThirdReality hubs (including the speaker-with-hub) use their server for connections
Then lets wait what @Arve_Bjornerud comes up with 
@Per-Erik_Sundberg Try with the new version, are you able to add your device now?
AI Voice Assistant v1.3.0 — Timers & Alarms
Your voice assistant can now set timers and alarms, fully hands-free:
- “Set a 10 minute timer”
- “Set a timer for the pasta”
- “Wake me at 7”
The LED ring counts down, the device chimes when time’s up. Alarms are just timers — it works out how long until the time you asked for.
Also new:
- Flow cards — triggers (started/finished/cancelled), a condition (timer running?), and actions (start/cancel), so timers work in your automations.
- Device tile shows the active timer’s name and time remaining, ticking down live.
One timer at a time, and it asks before replacing a running one. Feedback and bug reports welcome 
PS: There is a bug in this version: On the Voice PE, the first turn of a conversation plays fine, but follow-up turns in the same open session are silent — the assistant records and generates a reply, yet nothing comes out of the speaker. Starting a fresh conversation restores audio. Will look at this next.
@4x4_Pete I’ve ordered the ThirdReality speaker.
Version 1.4.0 ready for testing
Pick your voice engine — Google Gemini now works alongside OpenAI, each with its own set of voices to choose from.
More languages — the assistant speaks many new languages; just choose yours in Settings.- Smoother conversations — follow-up questions and natural back-and-forth work much better (this was a big one).
Lock the door by voice — lock and unlock supported doors with a spoken command.
Reliable time & date — ask what time or date it is and get the right answer, every time.
Better privacy — sensitive details are kept out of the logs.
New developer emulator — contributors can now run and test the app without Homey hardware, making it easier to improve and add features.
Next up is trying to run the AI locally Thankful for any feedback
Christmas is coming early this year. You guys get all your wishes 
Hosting everything locally and also Mistral is implemented and working.
ThirdReality speaker (TR) is also in the works. Initial testing proved it works with the App out of the box. Need some adjustments to accommodate the differences from the PE.
Will go for full App release in the Homey Store once TR is up and running.
Hi, @Arve_Bjornerud,
To have the app running with any of the 3 mentioned speakers, HA Voice PE, Thirdreality or Xiaozhi any outside paid account are needed for API tokens$?
When you mention Hosting locally that means having an extra PC running a server?
But it does look great, Thank you!
Yes, the speaker/mic hardware is relatively inexpensive, and running everything locally is completely private and “free”.
But Homey itself isn’t powerful enough to run anything AI related, so you would need to have a dedicated machine to do Text-to-speech, a powerful enough LLM and Speech-to-text. I have tested on my laptop that has a Nvidia 4080 with 12GB of ram, and that’s pretty snappy. But that laptop is anything but inexpensive. And it has to be power 24/7. In comparison, $5 of OpenAI credits last forever. The AI models used are very low-cost.
thanks,
Beside the dedicated machine and speaker mic hardware,
what software combo would you recommend for local control.
I’ve been testing LM studio. Nice interface, good for beginners. This is only for the LLM part, so you would need to do STT and TTS somewhere also, i just host those in docker and use Whipser and Piper. Look in no.arvebjoe.ai-voice-assistant/README.md at main · arvebjoe/no.arvebjoe.ai-voice-assistant · GitHub for more details.
