FURI | Spring 2026

GreyBox: An Autonomous Visual Testing Agent with Codebase Verification for Dynamic Web Applications

This research investigates whether an autonomous AI agent can replace traditional, brittle script-based web testing by visually navigating applications and verifying correctness against source code. Using Vision-Language Models (VLMs) to perceive interfaces like a human tester, combined with Retrieval-Augmented Generation (RAG) to pinpoint bugs directly in the codebase, the system moves beyond simply detecting failures to diagnosing their root causes. This Grey-Box approach promises greater software reliability while significantly reducing the maintenance costs associated with conventional automation frameworks. Future work will explore training smaller, locally-run models to make the framework accessible to resource-constrained development teams.