Agent Immune

Agent Immune is an adaptive security toolkit for AI agents, providing defenses against prompt injection, information exfiltration, and multi-turn attacks. It leverages both rule-based (regex) filtering and semantic memory to detect not just known threats but also rephrased or multilingual attacks that bypass static rules. Features include input assessment, output scanning for credentials/PII, rate limiting (circuit breaker), prompt hardening, threat learning, and threat sharing, all accessible via Python or as an MCP server for integration with agent governance stacks. Suitable for developers of AI agents, platforms, and governance toolkits aiming to enhance security and resilience.