A developer has successfully integrated the Surface Pro 11’s Neural Processing Unit with the R programming language to run local large language models using Foundry Local and ellmer.

Key Points

• The project utilizes the Foundry Local SDK to interface with the Surface Pro 11's integrated NPU hardware.
• R code leverages the ellmer and httr2 packages to establish an OpenAI-compatible connection to local model endpoints.
• The implementation requires the Foundry CLI to be installed and accessible via the system PATH for service management.
• The workflow automates model downloading, loading, and unloading processes directly through R system commands.
• This method provides an alternative for users seeking to run LLMs locally without relying on software like Ollama or LM Studio.

Why it Matters

This experiment demonstrates how developers can bridge the gap between specialized hardware accelerators and data science environments like R. By bypassing traditional GUI-based tools, users gain more granular control over local AI deployment on modern Windows-based ARM hardware.