P2P model hosting and inference WebGPU,libp2p,IPFS

This is a side project of mine demonstrating an application of my peercompute library. you can load a model locally or just run inference on peers. inference requests are routed automatically to peers hosting models capable of handling the request if a local one isn't available. image generation/analysis is supported as well as speech synthesis and recognition. peer traffic is unencrypted and uncensored and there are no logins. you can run any model you can find on hf that works. don't do anything stupid! the only backend is a libp2p relay that is also doing stun turn and ice.