Candle + WASM for fully local browser AI apps

[ferris-joy]

One of the most interesting parts of Candle to me is the WASM support.

Curious how far people have pushed:

• browser-based inference
• quantized models in WASM
• memory constraints
• streaming generation in-browser
• WebGPU integrations

Feels like there’s a huge opportunity for fully local/private AI applications without depending on Electron + Python stacks.

Would be interested to hear what architectures or limitations people are currently running into.