Operating power grids is becoming increasingly challenging. Decentralized power generation, fluctuating loads, and new regulatory requirements are making decisions in grid control centers more complex than ever. At the same time, many operators face a shortage of experienced personnel. Against this backdrop, the Fraunhofer IOSB-AST presented "GridCompanion" at the VDE FNN/ETG Tutorial on Protection and Control Technology 2026 in Leipzig at the end of April – a new AI assistant designed to support control room staff in their analyses without actively intervening in grid operations.
A key difference compared with many other AI solutions lies in its mode of operation: GridCompanion runs entirely on local infrastructure. As a result, data remain with the grid operator, and no external access or cloud connection is required. Among the technologies used are European language models such as Mistral Small 4, which can be operated directly on site. “For grid operators, it is crucial that AI solutions meet the strict requirements for IT and OT security and integrate into existing processes without disrupting them. In addition, data security and data sovereignty remain top priorities,” explains Dr. Dennis Rösch of the Fraunhofer IOSB-AST. By integrating a European language model locally, the project deliberately emphasizes digital sovereignty.
Technically, the system is connected to existing applications in a read-only mode. It collects and structures information from different sources and supports tasks such as fault analysis, the evaluation of grid data, and documentation. Responsibility for all decisions remains with human operators. Initial real-world applications are already underway in operational grid environments, where GridCompanion is being used to test various AI-supported use cases in control center operations, including load-flow analysis and the evaluation of complex grid data. In addition, Fraunhofer IOSB-AST offers "GridCompanion Discovery", a structured program that gradually introduces grid operators to the use of such systems. Through several phases, specific use cases can be identified, tested, and evaluated with regard to their benefits and integration requirements before any potential deployment in operational environments.
