cai

Why we included this project

CAI is an agent-based framework for building cybersecurity AIs that take on both offensive and defensive security work. Security researchers and ethical hackers can assemble specialized agents for jobs like reconnaissance, exploitation, and privilege escalation, with built-in tooling and support for more than 300 LLMs from providers including OpenAI, Anthropic, DeepSeek, and Ollama. The project has been put through HackTheBox CTFs and real bug-bounty engagements, and its design includes guardrails against prompt injection and dangerous command execution, which is worth having when an LLM has access to a shell. It is still actively developed and carries explicit ethical-use terms, so treat it as a research-grade foundation with proven workflows rather than a finished commercial product.