PROV-AGENT: Unified Provenance for Tracking AI Agent Interactions in Agentic Workflows

2025-08-04

AI Summary

AI agent workflows in heterogeneous federated environments (edge/cloud/HPC) suffer from coarse-grained provenance, metadata silos, and broken contextual continuity. Method: This paper introduces the first fine-grained provenance model tailored for agent workflows. It extends the W3C PROV standard by integrating the Model Context Protocol to enable unified, multimodal modeling of prompts, responses, decisions, and other agent artifacts, and implements an open-source, near-real-time provenance system supporting end-to-end contextual linkage. Contribution/Results: The approach ensures cross-environment decision traceability, context reproducibility, and auditable influence chains. Evaluated across multi-facility deployments, it significantly improves error propagation localization accuracy and decision reliability analysis—establishing a verifiable provenance infrastructure for trustworthy AI agent systems.

Technology Category

• Data Mining & Knowledge Management: Representing, Reasoning, and Using Provenance, Trust
• Multiagent Systems: Modeling other Agents
• Cognitive Modeling & Cognitive Systems: Agent Architectures

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• Security and Privacy: Data transparency and provenance
• Web Mining and Content Analysis: Web data provenance, reliability, and authenticity

Abstract

Foundation models, such as Large Language Models (LLMs), are increasingly used as core components of AI agents in complex, large-scale workflows across federated and heterogeneous environments. In agentic workflows, autonomous agents plan tasks, interact with humans and peers, and shape scientific outcomes. This makes transparency, traceability, reproducibility, and reliability essential. However, AI-based agents can hallucinate or reason incorrectly, and their decisions may propagate errors through the workflow, especially when one agent's output feeds into another's input. Therefore, fine-grained provenance is essential to link agent decisions, their end-to-end context, and downstream impacts. While provenance techniques have long supported reproducibility and workflow data understanding, they fail to capture and relate agent-centric metadata (prompts, responses, and decisions) with the rest of the workflow. In this paper, we introduce PROV-AGENT, a provenance model that extends W3C PROV and leverages the Model Context Protocol (MCP) to integrate agent interactions into end-to-end workflow provenance. Our contributions include: (1) a provenance model tailored for agentic workflows, (2) a near real-time, open-source system for capturing agentic provenance, and (3) a cross-facility evaluation spanning edge, cloud, and HPC environments, demonstrating support for critical provenance queries and agent reliability analysis.

Problem

• Tracking AI agent interactions in federated heterogeneous workflows
• Ensuring transparency and reliability in agentic decision-making processes
• Integrating agent metadata with workflow provenance for error analysis

Innovation

• Extends W3C PROV for agent workflows
• Integrates agent metadata with MCP
• Real-time system for provenance capture