AgentHound

AgentHound is an open-source offensive security framework designed for AI agent infrastructures, capable of conducting comprehensive reconnaissance, asset fingerprinting, credential harvesting, model inventorying, and active exploitation. It integrates findings into a Neo4j graph to visualize attack paths, addressing every layer of the agentic stack, including model gateways and inference servers. Notable features include credential inventorying, model inversion capabilities, and the ability to perform active exploitation through tool and instruction poisoning.