I know the OpenAI Conversation integration has been around for a while but it’s rather limited as it’s just giving you the states. So I decided to try to let it actually call stuff, hacked together some stuff in SvelteKit, and Home GPT is the result. It doesn’t work half the time, but it can work.
2 Likes
Changed it to use templating to get information. It might be better to just give all the info in the prompt so it never needs to get anything, but tokens cost money, and there is a limit on tokens.
Since the AI seems to not be that good at querying info, I’m experimenting with looking for keywords in the message (eg “weather” for weather entities) and giving the matching context.