AI Chat Before the OS Boots

A developer has created a system that lets you talk to an AI immediately upon powering on a PC — no operating system, no kernel. Demonstrated on a Dell E6510 laptop, the project runs LLM inference directly within UEFI boot services mode.

The Full Stack in Freestanding C

The entire application is written from scratch in freestanding C with zero external dependencies, implementing:

Tokenizer
Weight loader
Tensor math engine
Inference engine

The boot flow is straightforward: power on → select "Run Live" → type "chat" → talk to an AI. Everything operates in UEFI boot services mode, completely bypassing the traditional OS layer (with the exception of Wi-Fi drivers still in progress).

Current Limitations and Roadmap

The developer acknowledges that the system is currently quite slow due to lack of optimization. The priority is network driver implementation first, with performance optimization to follow. Plans include evolving it into a small-model server.

Why This Matters

Beyond being an impressive technical feat, bare-metal AI inference opens intriguing possibilities: ultra-lightweight edge devices, embedded systems with no traditional OS overhead, and secure environments where OS exposure must be minimized. The 394-score community reception shows strong interest in this unconventional approach to AI deployment.

#uefi

Bare-Metal AI: Running LLM Inference Directly in UEFI, No OS or Kernel Required