AI in Your Pocket: Hands-on with LG EXAONE 3.5, the Top Korean Small Language Model
Zero Setup, AI Chatbot on a USB Drive
MONKOS has successfully run LG AI Research's EXAONE 3.5 (2.4B) on-device.
By layering EXAONE on top of the llamafile engine, we've put an entire AI chatbot onto a single USB drive. No internet, no installation, just plug and chat.
Why EXAONE?
Ranked #1 in Korean language performance among lightweight models with 2.4B parameters (KoMT 7.24). Unlike competing models such as DeepSeek and Llama, EXAONE avoids the frequent Korean-English mixing issue.
Measured Performance
| Environment | Speed |
|---|---|
| iMac (Intel CPU) | ~1 tok/s |
| RTX 4060 (GPU) | 30~50 tok/s |
| USB (Windows PC) | 5~15 tok/s |
What it Means for Small Business Owners
"A chatbot that knows my store better than I do" — AI trained on the owner's expertise, providing 24/7 customer support. Zero cloud costs, zero data leak concerns.
This is the reality of the AI chatbot tailored for small business owners that MONKOS is pursuing.