Ollama Assist integration | Home Assistant | Page 1

ashen rivet Jun 22, 2025, 3:56 AM

#

Hey all,

I picked up a Home Assistant Preview Edition the other day, and i've been playing around setting it up to act like GLaDOS from Portal. The only issue that I am encountering is the tts reading out the Assist commands from Ollama instead of executing them. The attached image shows an example of this, the HassListAddItem command gets read out through the TTS instead of being executed.

Ollama is running Mistral, which supports tools. I followed the guide on setting up Ollama on the Home Assistant site.

This is probably just a simple misconfiguration on my part as I am new to Home Assistant, but help would be appreciated!

willow acorn Jun 22, 2025, 5:00 AM

#

ashen rivet Hey all, I picked up a Home Assistant Preview Edition the other day, and i've b...

have you given it assist permission?
on the integration press configure on your model connection and ensure "control home assistant" is ticked

dense quest Jun 22, 2025, 6:48 AM

#

What are you running Ollama on?

ashen rivet Jun 22, 2025, 12:22 PM

#

i am running ollama on Windows 11, and i have given it assist permission

calm blade Jun 22, 2025, 9:43 PM

#

what mistral you are exactly using

ashen rivet Jun 23, 2025, 3:53 AM

#

just the default 7b
https://ollama.com/library/mistral

willow acorn Jun 23, 2025, 3:57 AM

#

ashen rivet just the default 7b https://ollama.com/library/mistral

try using the instruct variety - mistral:7b-instruct

ashen rivet Jun 23, 2025, 4:10 AM

#

#

i'm going to read through the documentation more, i'm probably just missing a step 😅

calm blade Jun 23, 2025, 7:38 AM

#

ashen rivet just the default 7b https://ollama.com/library/mistral

with that you wont have any success. Mistral-small-3.2 is the way to go, but require some amount of VRAM

remote quarry Jun 23, 2025, 11:09 AM

#

How many entities/scripts/automatons are exposed to assist? What context window size are you using? You might need to bump context window size up as it may be losing the tool knowledge from the window if it is too small and being truncated. I have my context set to 16384, the default is 8192. Don't people even go 32768. You may also want to lower the max history messages to something like 3 to save on context space.
There are also some settings on the ollama instance itself that can help with vram usage for the context window.
Note that as you increase context size, ollama will need more vram to store the context in addition to the llm model itself.

#Ollama Assist integration