Assist not finding entites | Home Assistant | Page 1

lethal crest Dec 9, 2025, 7:38 PM

#

I have a local LLM using Ollama. The model is GPT-OSS-20B. I have tried exposing my sunset entity, as well as adding a template in the prompt and each time it just lists all my devices. I really don't want to expose 4000+ sensors. What is the best way I can do this?

This is my prompt and it all seems to be ignored:

Answer questions about the world truthfully.
Answer in plain text. Keep it simple and to the point.
Answer general questions.
Axt like Dwight from the Office when you respond.
Current Time: {{ now() }}
The sunset today is {{ state_attr('sun.sun', 'next_setting') }}.
When asked about the sunset, use the provided sensor value.```

#

Another piece of context is using OpenAi works just fine.

real stirrup Dec 9, 2025, 8:21 PM

#

The conversation agent will expose all entities that are "exposed" to assist. You need to unexpose them from voice assistant settings page. And then expose sun only.
The prompt you use is old way to expose entities - that was used before official LLM integration into voice pipeline.

lethal crest Dec 9, 2025, 8:33 PM

#

Okay I will remove that. Why does Openai work with sunset but ollama doesnt? The context window is much smaller on Openai.

#

If I have everything exposed openai works just fine. Ollama doesn't.

lethal crest Dec 9, 2025, 9:22 PM

#

@real stirrup

real stirrup Dec 9, 2025, 9:22 PM

#

I have no idea. 🙂

lethal crest Dec 9, 2025, 9:23 PM

#

Fair enough. Thank you

lethal crest Dec 9, 2025, 10:53 PM

#

Anyone know why? Ive fixed my prompt and it still doesn't seem to work. Sunset is expsoed and had mulitple aliases but never works.

restive lagoon Dec 10, 2025, 1:45 AM

#

I am also using gpt-oss:20b with ollama. The LLM responded correctly but it took 5sec. I did not have any special instructions about the sunset in my prompt. I have limited my exposed entities to ~90 and am using a context size of 32768 which is probably overkill. If you are running Home assistant 2025.12, you should be able to see the complete prompt and any tools that are called using the voice assistant agent debug. That might help troubleshoot the issue. I did have to add instructions in my prompt to have the LLM call my weather script reliably.

lethal crest Dec 10, 2025, 2:24 AM

#

Mind sharing your prompt? I have my context size set to 24000, I'll try upping it some more.

restive lagoon Dec 10, 2025, 3:01 AM

#

My promt is as follows:

You are a voice assistant for Home Assistant.
Answer questions about the world truthfully.
Answer in plain text. Keep it simple and to the point.
Do not use markdown formatting in your response.
Format your responses for TTS, don't include any special characters a TTS may have trouble speaking, such as asterisks or colon. Replace words like "Ms." and "Mr." with actual words like "miss" and "mister".
Use the script.voice_weather_forecasts_full_llm_script for weather forecast requests.Answer temperature or humidity values with 1 decimal digit or less.
Use the low reasoning model.

I do have many sentence automations for music, to-do/shopping lists and etc. to make sure I get the results I want.

#Assist not finding entites