Towards Responsible Generative AI: A Reference Architecture for Designing Foundation Model based Agents

TL;DR

This paper tackles the lack of a holistic, trustworthy architecture for foundation model–based agents. It develops a pattern-oriented reference architecture derived from an empirical literature review and validates it by mapping to two real-world agents, MetaGPT and HuggingGPT. The RA identifies core components—interaction engineering, memory, planning, execution, RAI plugins, and model choices—and multiple patterns to support responsible AI. The work provides a practical design template and patterns for building, evaluating, and governing FM-based agents, with future work on decision models for pattern selection.

Abstract

Foundation models, such as large language models (LLMs), have been widely recognised as transformative AI technologies due to their capabilities to understand and generate content, including plans with reasoning capabilities. Foundation model based agents derive their autonomy from the capabilities of foundation models, which enable them to autonomously break down a given goal into a set of manageable tasks and orchestrate task execution to meet the goal. Despite the huge efforts put into building foundation model based agents, the architecture design of the agents has not yet been systematically explored. Also, while there are significant benefits of using agents for planning and execution, there are serious considerations regarding responsible AI related software quality attributes, such as security and accountability. Therefore, this paper presents a pattern-oriented reference architecture that serves as guidance when designing foundation model based agents. We evaluate the completeness and utility of the proposed reference architecture by mapping it to the architecture of two real-world agents.

Figures (2)

• Figure 1: Architecture of an agent-based ecosystem.
• Figure 2: Reference architecture for designing an autonomous agent.