Autonomous penetration-testing agent that orchestrates security tools through a structured intent pipeline.
CLI Tool · AI Security
AI-Cyber is an autonomous penetration-testing agent with a structured intent IR pipeline. An LLM produces JSON intents compiled through a validator and command generator, preventing command injection across 40+ security tools (nmap, sqlmap, nuclei, hydra) executed in an isolated Kali Linux container.
Key features
- Plan-driven orchestrator with saturation detection — 70% response-token match threshold triggers re-planning
- 4-tier memory system — session NDJSON, hot KB with 7-day TTL, vector embeddings, 103-file static knowledge base
- Hard budget enforcement — 20 LLM calls, 100 tool calls per engagement to control cost
- Scope-aware capability guard parsing natural-language constraints into deterministic policies across 7 dimensions