Launching today
Agent Breaker
Break, test, and secure AI agents before production
2 followers
Break, test, and secure AI agents before production
2 followers
Agent Breaker is a chaos engineering and security testing framework for AI agents. 2000+Pypi Installs: https://pypi.org/project/agent-breaker Test LangGraph applications against prompt injection, tool misuse, and other adversarial scenarios before deployment. Features: • Automated adversarial testing • Failure detection and reporting • LangGraph integration • Security-focused evaluation • ML Based Judge for evaluation Build safer, more reliable AI agents by breaking them before your users do
Hey Product Hunt! 👋
I'm Gokul, the creator of Agent Breaker.
While building and studying AI agents, I noticed that most developers focus on making agents work, but spend far less time testing how they fail.
Traditional software has unit tests, integration tests, load tests, and chaos engineering. AI agents often have none of these protections despite having access to tools, memory, and external systems.
I built Agent Breaker to help developers automatically stress-test LangGraph applications using adversarial attacks and failure scenarios. The goal is simple: find vulnerabilities, unsafe behaviors, and reliability issues before users do.
This is an open-source project, and I'd love feedback from anyone building AI agents, agent frameworks, or evaluation systems.
What are the biggest failures you've encountered while building AI agents?
Here is the Interactive Demo Link: https://app.usehexus.com/flow/f83817a6-4dcd-4ecb-92b5-6fd8f35d4633?step=1