Features
Prompt injection detection
Jailbreaking attack automation
Attack techniques: AdvPrefix, AutoDAN-Turbo, PAIR, TAP, FlipAttack, BoN, h4rm3l, CipherChat, PAP, Static Template
Pre-built benchmark datasets (AgentHarm, JailbreakBench, HarmBench, AdvBench, StrongREJECT, etc.)
Multi-framework agent support (Google ADK, OpenAI SDK, LiteLLM, LangChain, Ollama, vLLM)
Modular attack engine with generator and judge LLMs
Real-time terminal UI with attack progress and visualizations
Optional cloud sync via API key
Custom file dataset import (JSON/CSV/JSONL/TXT)
HuggingFace dataset integration
Autonomous strategic report generation (Sakana Marlin)
Multi-agent LLM orchestration matching frontier models (Fugu Ultra)
Recursive Self-Improvement (RSI) Lab for autonomous AI research
Conductor system for natural language agent orchestration
Japan-based data residency and export control compliance
Finance multi-agent proposal generation with SMBC Group
AI-powered information analysis with DEEP DIVE
CoffeeBench benchmark for LLM agents in multi-agent economic environments
Autonomous model orchestration avoiding export control risks
Research-stage speech-to-speech AI with KAME architecture