Even embedded systems need their own dedicated AI assistant. That seems to be the idea behind the new AMD Ross. This agent is designed to enable a full development cycle entirely within embedded IT. Engineers control AMD tools using natural language, and Ross handles everything from hardware partitioning to verifying PCB designs.
The tool is available starting today. By “embedded systems,” AMD refers to FPGAs, adaptive SoCs, embedded x86 chips, and specialized edge AI platforms. In other words, all types of AMD computing power not found in a conventional PC, server, or mobile system.
With Ross, engineers can consult AI in secure environments to search documentation, execute code, and debug. AMD does not specify a particular LLM powering Ross; rather, it functions as a kind of harness that uses provided models to deliver an assistant-like capability. This allows for the use of both cloud-based LLMs and on-premises AI.
Four building blocks
Ross is thus aimed at all developers working with AMD’s embedded portfolio. Users choose their own model, as well as the development environment and command-line interface. “AMD Ross brings together AMD Embedded tools, trusted knowledge, and expert-written workflows into a single agent-based AI experience,” says Salil Raje, SVP and general manager of AMD Embedded.
The foundation consists of four components. MCP servers, based on the open Model Context Protocol, connect AI agents of all shapes and sizes to the AMD tools. A Knowledge Base containing vectorized user guides, white papers, and application notes is available via the cloud or locally offline. In addition, there are agent skills: open Markdown files containing best practices, such as for timing optimization or restructuring C++ code in Vitis HLS. Ready-to-use design examples demonstrate how Ross can be utilized.
According to AMD, this means Ross is not simply an AI coding assistant built on top of existing tools. The company’s primary goal is to make the knowledge of its own engineers reusable. In many environments, this is a good thing, since the hardware available in embedded environments does not always support larger LLMs or allow for a connection to the outside world.
Read also: AMD gives its embedded chips 80 TOPS of AI computing power