modelfuzz

ModelFuzz is a runtime guardrails tool designed to intercept and prevent unsafe tool calls made by AI agents due to prompt injection attacks. It checks each argument against defined policies before execution, ensuring that only permitted actions, such as API calls to whitelisted domains, are executed, effectively blocking any malicious attempts. Notable features include support for both synchronous and asynchronous function wrapping, configurable policies, and logging of blocked actions for auditing purposes.