AMD Agent Computer System Local AI Workflows with RyzenClaw and RadeonClaw
AMD has developed its "Agent Computer" system which enables local artificial intelligence workflows through its RyzenClaw and RadeonClaw technology. AMD has published its Best Known Configuration (BKC) guide which enables users to operate OpenClaw on Windows through local installation. The new system from AMD represents a change from traditional Personal Computer systems because it uses its Agent Computer system to operate autonomous AI agents on local systems without needing cloud data centers. Through WSL2 and LM Studio implementation, AMD provides users complete data ownership rights along with no restrictions on usage and continuous system operational time.
AMD provides two separate hardware setups which enable local agentic computing with each setup designed to meet specific performance requirements and memory needs. The Qwen 3.5 35B A3B model serves as the common reference point for both paths which use it to perform local inference tasks.
1 RyzenClaw (The "Agent Swarm" Path)
This configuration is built for parallel tasking and large context windows. The Framework Desktop uses the Ryzen AI Max+ processor as its core processing unit.
- Memory capacity consists of 128 GB unified RAM with 96 GB allocated for Variable Graphics Memory storage.
- The system can process approximately 45 tokens every second.
- The system can handle a context window that extends to 260K tokens.
- The system supports operation of six agents at the same time.
- The complete system costs approximately $2,700 to obtain.
2 RadeonClaw (The High Speed Path)
The workstation setup uses high performance graphics processing units to deliver maximum speed and better performance during work tasks.
- Hardware: The system uses Radeon AI PRO R9700 which comes with 32 GB VRAM.
- The system can process approximately 120 tokens every second.
- The system can handle a context window that extends to 190K tokens.
- The system supports operation of two agents at the same time.
- The GPU costs $1,299 to purchase as a standalone product.
The local environment is built on a Windows first architecture using the following stack:
- Infrastructure: WSL2 (Ubuntu 24.04) manages the OpenClaw core and browser control.
- The system utilizes LM Studio for local LLM setups through its llama.cpp interface.
- The system maintains data integrity through local embeddings which enable agents to access memory.md support without needing cloud access to their entire knowledge base.
The graphical representation of the Agent Computer system describes how users will fundamentally change their approach to physical computer equipment. Users must perform application work through sequential steps in traditional PC systems whereas the design of Agent Computer systems allows users to assign work through direct commands. The apps Slack, WhatsApp, and Discord use background agents to operate which perform research and task execution through automatic summary creation.
AMD claims that the artificial intelligence agent represents the primary user for these computer systems. The two components require efficient processing to support continuous operation which the system provides through its unified memory and parallel computing capabilities.
The upcoming 2026 local AI agents revolution has become an actual technology in today services though users must pay high entry fees to access it. AMD recognizes its present product launch as a solution that offers developers and early adopters and small to medium enterprises (SMEs) access to secure computing resources which deliver both privacy and operational efficiency.
