I like the ability to forward commands that are not understood to an LLM, it gives Sophia both speed and flexibility however at the moment the only way to know when Sophia is forwarding to AI is to set a specific command. It would be preferable if instead it would ask something like "I didn't understand, would you like me to ask custom ai name here?" That way it is clear when your commands were misunderstood.
On the topic of forwarding to LLM, it would be nice if instead of having to select an LLM from the available list and putting in the api key, you selected a home assistant voice assistant pipeline to fallback to. For instance, if you had 2 Home Assistant assist pipelines, one for Sophia NLU, and another for ollama, with Sophia as the default, when you ask the voice assistant for something, if it fails it then goes to the fallback pipeline which is Ollama. This would allow all your devices to be exposed (which is not possible at the moment), as well as giving access to any LLM setup through Home Assistant, not just the LLMs built into the Sophia NLU HA App. It also makes maintaining the code easier as you no longer need to update the app when companies (probably google) change how their API works.