forest otter
In the update blog for 2025.9 in the Other noteworthy changes section there is a mention of fuzzy matching being added for non LLM pipelines.
What exactly does this mean?
Will it autocorrect me if I misspeak?
What does this feature allow me to do that I couldn't do before.
What does it prevent from happening?
round tusk
The PR has more information: https://github.com/home-assistant/core/pull/150595
forest otter
I still don't get it.
The doc says "The template 'set lights to (red|green|blue)' cannot match the sentence "red lights", for example."
When I say 'Set red lights to blue' or 'set lights to blue lights' it responds with 'An unexpected error has occurred'.
hollow geyser
question about the new intents. When I ask it " Office fan 33%" it works, but it's reponse back is just. "..."
and I get this in logs every time it uses the new intent:
ypeError: undefined is not an object (evaluating 'a.data.intent_output.response.speech?.plain.speech')
isToolResult (src/components/ha-assist-chat.ts:547:60)
event.type.startsWith (src/components/ha-assist-chat.ts:433:46)
message (node_modules/home-assistant-js-websocket/dist/connection.js:23:42)
forEach ([native code]::)
message (node_modules/home-assistant-js-websocket/dist/connection.js:15:34)
and for reference, I get this with the Home Assistant Cloud assist.
when I do the developer tools the sentence "office fan 33%". target matches a ton of things.