AgentHound: Autonomous Offensive Security for Agentic Infrastructure

AgentHound is an open-source offensive security framework designed for agentic infrastructure. It performs autonomous scans that capture credentials, discover AI services, and prove concrete access in a single run. Results are preserved as a single artifact that supports full attack-graph analysis. The framework targets AI infrastructure operators, offensive security teams, and security engineers. It includes an analysis server component (agenthound-server) with commands such as serve, ingest, query, and version. AgentHound reads configured Ollama backends and enabled Qdrant external-knowledge connections but never stores their backend authentication material. AgentHound writes an ingest-valid artifact before collection and checkpoints after every meaningful result or action transition. It supports cross-protocol path queries and helps validate process isolation, identities, authorization, and an authorized end-to-end call.