[Idea] Backend that could help make the AI interaction experience more natural #2
Description
I saw your Reddit post and I saw you mentioned something about having the LLMs/characters message you like humans at certain points in time instead of you sending a message and getting an instant reply (correct me if I got it wrong).
There's a certain backend that can be used to build logic for characters like that:
https://omnichain.zenoverflow.com/
https://github.com/zenoverflow/omnichain
It's not a loader like LM Studio or Ooba, it's more of a visual programming interface and workflow runner (they run on the backend) that exposes an OpenAI-compatible API which only uses the messaging parts, so can easily be wired into any frontend. It has certain nodes inside the workflows that connect to actual loaders like ollama / LM Studio / TextGenWebUI / OpenAI-compatible endpoints, so you can also mix and match.
The point being - you can make workflows that loop endlessly and add messages to a log whenever they want to, like real humans. And you could feasibly configure the workflow to accept requests with empty messages as a signal to send down any unread messages to your app.
Edit: It could also help you get easy RAG support for your app.
If you find this interesting/useful, I'd be willing to make such a workflow specifically for your use-case. Like you, I lost my regular day job because companies started adopting AI to do what I do, and freelance work started drying up too, so I had to build something to cope with the "death by a thousand rejections" situation on LinkedIn (still not enough to impress employers apparently).