I have a mac mini with 24GB ram and currently I am running Qwen3.5:9b on it with Ollama.
It works fine for chat.
But I want to control home assistant on it. So I provide access token and URL and basically it works. But it makes too many curl requests and that increases the used context and takes too long to process.
I tried other models (smaller qwen mostly) and llama.cpp instead of ollama.
But it did not get better.
So I am looking for a better , more effective way.
I see that there are other methods accepted by the community.
- Ollama integration and conversation agent (not good enough , because it does not respond with human language to questions like “which lights are on ?”
- project here: OpenClaw/ClawdBot on Home Assistant
(I could not try it because it requires HAOS and I use Docker)
is there a “best practice” for this ? Especially for an hardware like Mac Mini ?
is there a small llm model which is specially trained for HA control ?
is there a better approach to the “control” problem ? maybe a better API which would not require several curl requests to do just one easy task of “turning on all lights in living room” ?