FRAME: A Model-Based Foundry for Engineering, Adapting, and Assuring the Quality of AI Agents
AI agents powered by large AI models automate interactions between humans and digital tools and are increasingly adopted to enhance productivity and integrate local and external data sources and services. However, current agent technology is still unable to provide the reliability, transparency and composability required for scalable and trustworthy deployment. The lack of engineering foundations and systematic quality control prevent agents from evolving into trustworthy, reliable, composable and maintainable systems.
FRAME aims at advancing the engineering of AI agents by creating a software-engineering foundry layered on top of existing agent technologies. The foundry will provide reference architectures, model-driven engineering methods, lifecycle governance, quality control and self-improvement to support the development of modular, composable and trustworthy agents able to reason about uncertainty. It will deliver the engineering abstractions, methods and tools needed to support reliable, scalable end-to-end development, deployment and evolution of AI agents. By establishing foundations for trustworthy agent engineering, FRAME will enable productivity, adaptability, reliability and foster both adoption and economic value across sectors.
Simula participates in FRAME as a partner and will focus on quality control, to manage and execute the testing of AI agents. Additionally, Simula will lead the task dedicated to understanding and characterising uncertainty within the project's framework.
To achieve the overall goal, the project will undertake research into unified abstractions, methods, architectures, models, and tools building upon lessons from software engineering.
The main research objectives include:
- Design a reference software architecture for AI agent integration to support modular, extensible, and interpretable reasoning over external services.
- Investigate standardised software service engineering abstractions and techniques (such as Model Context Protocol (MCP) and Agent-to-Agent (A2A) protocols) to simplify complex multi-agent workflows.
- Develop adaptation techniques that integrate AI explainability and user feedback, enabling flexible runtime reconfiguration.
- Develop parameterised adaptation templates and automated runtime weaving methods to support continuous alignment, transparency, and auditability.
FRAME's outputs will be validated on real-world use cases spanning three critical settings: robotics, healthcare, and software frameworks.
Funding
This project is funded by the Horizon Europe programme under Call HORIZON-CL4-2025-04, topic HORIZON-CL4-2025-04-DATA-03. It is a HORIZON Research and Innovation Actions (HORIZON-RIA).
Disclaimer: Funded by the European Union. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or CINEA. Neither the European Union nor the Granting authority can be held responsible for them.
Partners
- Fundación Centro Tecnolóxico de Telecomunicacións de Galicia (GRADIANT), Coordinator, Spain
- Bournemouth University (BU), UK
- Dublin City University (DCU), Ireland
- Simula Research Laboratory AS (SRL), Norway
- Universiteit van Amsterdam (UvA), Netherlands
- Charles University of Prague (CUNI), Czechia
- IRATEN Solutions (IRS), France
- PAL Robotics (PAL), Spain
- Process Talks (PT), Spain
- F6S Network Ireland Limited (F6S), Ireland
- Servicio Andaluz de the Salud (SAS), Spain
- University of Southern Denmark (SDU), Denmark
- Fundación Pública para la Gestión de la Investigación en Salud en Sevilla (FISEVI), Affiliated, Spain