Pryv
A decentralized frontend inference for multiple language models hosted on IPFS and Fleek.
Demo
About
It's a simple next.js application that has a couple functions. The first option is to ping the LLM's API's directly from the browser, from the model of your choice, keeping a coherent chat history even when switching models. The models i have are GPT-4o, GPT-4 GPT-4 turbo, and llama through Kuzco's API.
How it's made
It's built using IPFS and Fleek, and the OpenAI and Kuzco Llama API endpoints. Pretty vanilly, as i started late, but a decent demo! I used react-speech-recognition for the voice input, though it doesn't work in the brave browser. Some of the things that are half baked in there is wallet auth to create the account, chat history stored on IPFS, private inference using lit actions encryption/decryption etc.