Coming soon.
APAi is a work-in-progress concept decenralized shared-compute uncensored Roleplay-AI service.
How it will work:So, how does it work? You may ask. Well it's hard to convey in words, but I'm too lazy to create a chart, so we'll try. Firstly APAi uses
WebLLM to run an LLM in the browser (In the case of current tests Tifa-DeepsexV2-7b-MGRPO-q4f16_1-MLC though this will likely change). While you
can just use this (***They'll be an offline mode) the main focus of this idea is that it's
not just your machine running this model - it's many, every client visiting the site in fact. Now, this is the part that usually breaks ideas like this: Compatability and processing power. Now the first is easy to tackle; if there's no WebGPU failback to WebGL. But the second, no, that isn't as easily solved... or is it? APAi aims to use a P2P (With possible server intermediary) network of clients. So if your machine is bogged down the generation of your chat can be
offloaded to a client who does have extra power. Now, you may think this is unfair, but fear not! With accounts and credits you earn credits by contribtung to the network. Then, when you need to generate on other clients you can use those tokens. One token is one chat message. Every account gets 100 tokens by default per-day.