AMD is promoting a new concept it calls the Agent Computer, a system designed to run AI agents locally on personal hardware instead of relying on cloud infrastructure. The company recently published guidance showing how developers can run the OpenClaw AI agent framework directly on Windows using AMD Ryzen AI processors and Radeon GPUs.
The approach focuses on keeping AI workloads entirely on the user’s machine. AMD says many users and businesses want more control over their data, predictable costs, and the ability to run AI systems continuously without depending on cloud services or remote APIs.
AMD describes two hardware paths for running OpenClaw locally. The first configuration, called RyzenClaw, uses a system powered by Ryzen AI Max Plus processors with 128 GB of unified memory. The setup reserves most of that memory as graphics memory to run large language models locally.
In AMD’s testing, a Ryzen AI Max Plus configuration running the Qwen 3.5 35B A3B model produced about 45 tokens per second. The system processed 10,000 input tokens in roughly 19.5 seconds and supported a large 260K token context window. The platform could also run up to six AI agents simultaneously, allowing developers to experiment with multi agent systems or so called agent swarms.
A second configuration called RadeonClaw takes a workstation approach. It pairs the OpenClaw framework with AMD’s Radeon AI PRO R9700 GPU, a professional graphics card equipped with 32 GB of VRAM.
This setup delivers significantly higher speed. AMD reports around 120 tokens per second and the ability to process 10,000 input tokens in about 4.4 seconds. The trade off is a smaller 190K token context window and support for two concurrent agents rather than six.
Both configurations run through Windows Subsystem for Linux using LM Studio for local model inference powered by llama.cpp. The environment also supports Memory.md through local embeddings, enabling AI agents to maintain memory without relying on cloud services.
AMD says the environment can be configured in under an hour and is primarily aimed at developers and early adopters exploring personal AI agent workflows.
However, the hardware requirements remain expensive. Systems based on Ryzen AI Max Plus processors with 128 GB memory can start around $2700. Workstation setups using the Radeon AI PRO R9700 GPU begin around $1299 for the graphics card alone.
The initiative reflects a growing industry shift toward local AI computing. As models become more efficient and hardware improves, companies are beginning to explore whether some AI workloads can move away from data centers and run directly on personal machines.
If the trend continues, agent based AI systems running locally could become a new category of computing alongside traditional PCs.
Image Credit: AMD
