AMD bets on local AI with OpenClaw running on Ryzen and Radeon hardware
AMD unveils OpenClaw, a local-AI framework with two configurations—RyzenClaw and RadeonClaw—that run large language models on consumer hardware via Windows WSL2 and LM Studio (llama.cpp), aided by Memory.md for local context. RyzenClaw targets CPU-based inference with roughly 45 tokens/sec, a 260k token context, and up to six concurrent agents; RadeonClaw uses the Radeon AI PRO R9700 for around 120 tokens/sec, a 190k token context, and supports two agents. Priced from about $2,700 for a Ryzen-based desktop and $1,299 for the GPU, OpenClaw is aimed at developers and enthusiasts who value autonomy, privacy, and on-device AI over cloud-scale solutions.