I am currently working on an idea of using ChatGPT to simplify voice queries like “would you please turn on the kitchen lights so I can see something in here?” to just “turn on kitchen lights” and then be able to use the official HomeAssistant Assist to handle the intents.
My goal is to use LLMs to allow for more “human” interactions with my home.
The current state of the HA intents are barely usable in French and even English is very limited (no criticism here, a lot of good work is being done I know).
I’ve looked a bit online but haven’t found anything about controling and sending requesting to Assist except from the official APP and WebUI.
Is there any API I could use in my 3rd party app ?
Voice is still a lot in alpha testing at the moment.
Rhasspy has the features you request and the developer, Mike Hansen, is also the one hired to developer Voice in HA.
The new version of Rhasspy is the one being integrated into HA, so you can probably expect all the features from Rhasspy to become available in HA.
Rhasspy is really modular built and in the latest Year of the Voice video the developers mention a new Whisper standard for communication between modules to make alternations like the one you suggest even smoother.
It is therefore mostly a matter of having patience at the moment.
Thanks for your message, I have in fact seen the use of this Wyoming protocol.
Following the Rasspy3 documentation, it mentions an HTTP API.
If what you said about Rhasspy being bundled with HA is true, it means that this API is probably running somewhere inside HA but is probably restricted for usage by HA only or even entirely disabled.
It would be great if a dev could provide more details about this situation, opening/enabling this API would be great for 3rd parties and Assist could be ran on low power devices like a Pi to allow for voice recognition with wake-word and stuff without using the App or the web interface.
You might be using the limited release of the Voice assistant then.
It is pretty locked down, but it is mostly a proof of concept for a voice assistant.
The development is going for an open modular design.
I want to use assist/conversation in my own dashboard using the home assistant websocket or rest API.
I got conversation working, but it only uses the “home assistant” assist pipeline. I want to use my own assist as I have it set up in home assistant with openAI.
