Not working on it yet but can a small box sitting on your counter do LLM, TTS all local with maybe outbound queries to the internet.
First iteration may be box is rpi based and local LLM runs in another room on beefier machine (or even before that just get it working with a cloud Llama).
What would make this cool is to use MemGPT for memory so you can talk to it Monday and then it remembers what you said Friday.
First iteration may be box is rpi based and local LLM runs in another room on beefier machine (or even before that just get it working with a cloud Llama).
What would make this cool is to use MemGPT for memory so you can talk to it Monday and then it remembers what you said Friday.
Being all local it could be always listening.