My most recent project is an attempt to remove the Google Nest Mini device I use and substitute it with a local solution. This way I can fine-tune what I would like, giving me granular control over what my assistant can do and customize the voice to one of my choosing. To do this, I'm using Ollama running on my personal tower computer, and opening the port to my mini-PC, where I have Proxmox hosting HA OS. Inside Home Assistant, I have download the Wyoming tools that allow for using your own devices for voice input, processing and command making. This included "Assistant Microphone" for selecting your input audio device, the "OpenWakeWord" for specifying how to activate the listening mode (in my case I chose Hey Jarvis), then the audio is processed by "Whisper", a speech-to-text software, which then sends the clear text to the Ollama model. The Ollama model will interpret the input and decide whether it's a command for Home Assistant, or if it is a plain question for the LLM to answer. Upon Home Assistant receiving the answer, it will interpret the response and make the appropriate choice.