AMD says Ross, its agentic AI assistant for embedded engineering, is available now. It connects natural-language requests with supported AMD tools, engineering knowledge, reusable skills and design examples to support work ranging from hardware design and debugging to software, edge AI and deployment.
What AMD Ross does
Ross is not itself a large language model (LLM). AMD describes it as an assistant that works with a user’s preferred LLM and a compatible integrated development environment (IDE) or command-line client. AMD lists VS Code, Cursor, Devin, Claude Code, Copilot CLI and Codex CLI as examples.
AMD identifies four parts of the system: servers built on the Model Context Protocol (MCP) to connect with supported tools, the AMD Knowledge Base, reusable agent skills and design examples. Together, these components are intended to link natural-language requests with embedded-development workflows.
AMD tool support and licensing
AMD lists support for these tool versions:
| Tool | Supported versions |
| Vivado | All versions |
| Vitis HLS | 2025.2 and later |
AMD says Ross follows a monthly release cadence separate from regular Vivado and Vitis tool releases. Ross itself requires no additional license, but users still need applicable licenses for the AMD tools they use, as well as access to their chosen LLM.
Local setup and output review
AMD describes an air-gapped knowledge-base configuration with the database, embedding model and MCP server deployed locally. The documented setup also includes cloud or local access to the knowledge base.
Ross workflows are non-deterministic: AMD says results can vary between runs, models and prompt wording. Users are responsible for reviewing outputs and proposed actions before relying on or implementing them.
