Add support for running through Local in-memory AI models through webGPU

[haikyuu]

Is your feature request related to a problem? Please describe.
The GPT-4 bill is very big. And it's possible to use a local open model without doing any prior setup.

Describe the solution you'd like
https://github.com/abi/secret-llama is an example project that enables the user to load LLAMA-3 or another model completely in the browser, and talk to it via the OpenAI API.
I'd love for MemGPT to offer that as well.