Fraunhofer Institute of Optronics, System Technologies and Image Exploitation IOSB 
Project description
LLM4OPC allows users to interact with industrial systems through a secure, natural language interface. Through the Model Context Protocol (MCP), Large Language Models (LLMs) can read live machine data and explain statuses in an understandable way. With approval, LLMs can also execute actions through OPC UA and Asset Administration Shells (AAS). The agent-based workflow design separates the user interface, workflow, and tools. A planner defines verifiable steps; a tool agent performs read/write operations and method calls; and a replanner enforces human-in-the-loop verification for every state change, confirming results with a confirmation prompt.
The result: Employees have access to an auditable voice interface that allows them to query system statuses, navigate AAS information, and safely trigger system functions. Standardized toolkits for OPC UA and AAS, as well as LLM-compatible modeling rules (e.g., consistent namespaces and qualified bro
Project status
As of today, we offer MCP toolkits for OPC UA and AAS with reference workflows and review mechanisms. We also offer secure reference architectures that ensure observability and audit trail integrity for language-model-driven interaction with digital facilities. A key focus of future development will be on-premises operation using specialized small language models to bring AI expertise and data sovereignty directly to the production site.