AMD and Unsloth Enable NPU Support for Over 100 Agentic AI Applications
AMD and Unsloth have partnered to integrate AMD XDNA 2 Neural Processing Units (NPUs) with more than 100 agentic AI applications via the Unsloth Desktop API, enabling power-efficient local AI inference on supported AMD Ryzen AI PCs.

AI Tools Test
Part 1 of 6
Table of Contents
AMD and Unsloth announced a collaboration to support AMD XDNA 2 Neural Processing Units (NPUs) in over 100 agentic AI applications through the Unsloth Desktop API. This integration allows compatible AI applications to perform local inference on AMD Ryzen AI PCs using the XDNA 2 NPUs, offering a power-efficient alternative to GPU-based execution.
Unsloth Desktop’s API connects applications to models running directly on the NPU, enabling sustained low-power inference while freeing CPU and GPU resources for other tasks. The API provides an OpenAI-compliant endpoint supporting more than 100 AI tools, including Hermes Agent, OpenClaw, Open WebUI, Cline, LangChain, and GitHub Copilot.
When an application requests model inference, AMD’s Lemonade and FastFlowLM components serve the workload on the XDNA 2 NPU, accelerating model responses. This setup enhances AI workflow efficiency on PCs by minimizing resource contention and improving multitasking.
Unsloth Desktop also offers open-source tools for model fine-tuning and memory optimization, plus a desktop app for training and inference without coding. AMD’s addition of NPU support broadens hardware acceleration options beyond CPU and GPU, improving performance and energy efficiency for agentic AI applications.
This development reflects AMD’s efforts to expand dedicated NPU acceleration across AI tools and workflows, benefiting developers and users by enabling efficient local AI processing on Ryzen AI platforms.
This report is based on information provided by AMD.


