#Ollama prompt is incomplete / missing exposed entities

1 messages · Page 1 of 1 (latest)

sleek geyser
#

Hi everyone!

Apologies if this has already been talked about, but I haven't found any existing posts or bug reports.

I've been using HA for a while now, but am relatively new to LLMs and the voice and chat components.

I'm trying to use Voice Assistant with Ollama and I've set up the Ollama integration and a pipeline, but the prompt (https://github.com/home-assistant/core/blob/dev/homeassistant/helpers/llm.py#L76) that I was expecting the Ollama server to receive doesn't seem to come through.

I've exposed a handful of entities, but even if I hadn't I would expect more of a preamble in the prompt. As it is, the LLM is unable to answer any question about entity states, when tutorials generally show it working out of the box.

This is what the Ollama server receives:

Oct 31 13:52:39 ollama-01v ollama[149]: time=2025-10-31T13:52:39.848Z level=DEBUG source=routes.go:1417 msg="chat request" images=0 prompt="<|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\n\nYou are a voice assistant for Home Assistant.\nAnswer questions about the world truthfully.\nAnswer in plain text. Keep it simple and to the point.\nCurrent time is 13:52:39. Today's date is 2025-10-31.<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nSummarise my smart home<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n"

And the reply I get is:

I don't have any information about your specific smart home setup. I'm a new instance, I don't retain data from previous conversations. Can you please tell me what devices and services are part of your smart home?

This is including the instructions I configured in the Ollama integration, but is missing all of the HomeAssistant magic that (I assume!) should also be there. Except for "Current time is" and "Today's date is". Or am I misunderstanding how Ollama should be receiving data about the state of all my entities?

HA version: 2025.10.4

Is anyone able to advise me?

Many thanks,
- Joe

GitHub

:house_with_garden: Open source home automation that puts local control and privacy first. - home-assistant/core

cinder ledge
#

Did you check off the assist button on the model config? Is that isn't enabled it won't add any smart home stuff

#

Like this (should look the same for ollama)

hasty shore
#

having the same problem this is how it looks Olama in HA is completelly different from above.

#

better tested right now and it started to work thinking think before resonse 🙂

sleek geyser
#

Hey @cinder ledge ! Thanks for the reply. In my case you're right, I checked the "Assist" box and now the prompt is properly populated 🙂

I want to say that, when I first tried setting it up a few versions ago, there was a radio box between "No control" and "Assist". So I totally assumed that this radio-turned-checkbox was referring to whether or not the LLM was able to trigger actions (like turning off a light) versus having read-only access to entity states.

As it stands, does this really mean that you either tick Assist and have your entities exposed + the LLM has the ability to make changes, or you don't tick Assist and the LLM has no information whatsoever about your smart home?

It's not so much that I don't want a voice assistant controlling my home, just that for now it's not great at understanding what I'm asking it for haha
So I was more interested in using it to make queries (save annoying my housemate by accidentally toggling lights in the wrong room)

Using templates I can add general info about the house and entities in the instructions box, which the model might find easier to parse than the full on HomeAssistant prompt anyway. I'll tinker more on the weekend and see