Right now I’m trying to fully switch from Google Assistant/Nest to HomeAssistant as Voice Assistants. But one key feature is missing in my opinion: being able to ask questions about current or future events and information on the web in general.
E.g. when’s the next election for xyz or something like that.
I’ve tested a few things with different AI and the only thing that is able to produce accurate results for this is (who would have thought) Google.
Using it’s Gemini AI you can turn on “Grounding” as a feature so that it’s fact checking with a web search.
As far as I am aware, you can also use this feature in API Requests as a tool:
The “dynamic_threshold” should also be configurable in my opinion, since the use of grounding could get pretty expensive.
Can you add this to the official Google Generative AI Integration?
Either as a configurable model option, or to use as a own manual request to the API, so we can use it in scripts as tools for assist?
If I overlooked something or you have better solutions, please let me know.
That would be a gamechanger for using AI in combination with assist in my opinion!
Edit: for anyone wondering: yeah, ChatGPT (and most likely others) also has this feature, but only the App itself - not the OpenAI API.
In case anyone is interested in my current workaround:
I’m using ChatGPT as assist agent and created a script which calls a Perplexity API Request (Initial Setup - Perplexity) via RESTful Command - Home Assistant. Then i configured the context of my assist to always use this script if current / upcoming information is requested or the knowledge could be outdated. This is somewhat solid, but still a workaround and I think an out-of-the-box solution like Gemini with Grounding would result in better responses.
We can close this issue/feature request in my opinion, since it’s now fully supported for the Gemini and also OpenAI API Integrations for HomeAssitant since the latest 2025.4 Release!
How did you manage to have the assist use the script exactly? I also have created the script based on the description in the official integration documentation, but I am struggling to forward the voice assistant to it, when necessary.
Right now you could just use your OpenAI Assist with web search enabled - this works pretty good for me without such a perplexity script workaround now.
I don’t know if that’s the best approach, but what i did was roughly:
Add something like this to my assist context:
“Always answer questions about events and points in time, whether future or past, exactly correctly and on the basis of a current web search if you are not sure.
You should recognize the following keywords for current searches: “current”, ‘still’, “at the moment” and synonyms of these.
Even when it comes to questions about public figures (who are still alive), events or similar, always check the current facts.”
You could also add the name of the script you want it to use, but I’m not quite sure if that’s necessary.
For the Script: Set a description like “Search the web for specific, up-to-date information. Use this script if you need information on a topic that may require current data.”
Make sure that the script is exposed to your assist
PS: you could also ask the assist which tools it used for the request (also shown in debugging the assist requests) or which tools it knows in general.
