Home Assistant Assist vs Sophia NLU vs Ollama

This comparison covers Home Assistant's built-in Assist, Sophia NLU and an LLM served through Ollama on a separate server. The speaker reports inconsistent results over two weeks, so the video is an account of one installation rather than a controlled benchmark.

Assist still switches lights and handles climate changes in that setup, but temperature queries fail with "no valid targets" errors despite exposed sensors. The speaker suspects a recent core update without establishing the cause. Ollama understands natural language, yet requests take roughly 16 to 23 seconds in the examples. Multi-room commands sometimes produce unsupported tool requests or JSON in the chat instead of an action. Some responses claim success without changing the lights.

The speaker favors Sophia NLU for local AI voice control, describing it as deterministic software written in Rust that uses 160 MB of RAM and needs no GPU. Reported timings include 0.32 seconds for a light command and 0.02 seconds for an office temperature query. Sophia also handles a combined lighting request, though it misclassifies a bedroom temperature question. Users can mark responses incorrect and submit failures to the developer. The speaker wants entity exclusions because Sophia accesses all entities and sometimes selects adaptive lighting controls instead of lights. The description discloses a referral commission.