Map, understand & secure your AI agents
Visualize workflows, detect vulnerabilities, and threat-model your agentic systems — right from your terminal.
Agent-Wiz is an open-source CLI tool by RepelloAI that extracts real agent flows, simulates attacks using MAESTRO, and lets you defend your AI before it breaks.
Real-World Problem
Your AI agent stack is a security blindspot
Multi-agent systems look smart. But behind the scenes, they're messy, implicit, and deeply vulnerable.
From tool misuse to prompt poisoning and lateral escalation — your agents are talking to each other, calling LLMs, and invoking tools in ways your IDE will never show you.
Agent-Wiz makes those flows visible and threat-models what others can't even parse.
Features
Everything you need to secure complex agentic workflows
A purpose-built toolkit for modern AI engineers and security teams working with multi-agent orchestration.
AST-based flow extraction
Parses orchestrator files (Python) to generate true runtime agent → tool → LLM graphs.
MAESTRO threat modeling
Assesses each interaction across 12 known agentic failure modes.
Real-time graph visualization
Export to .png, .dot, or interactive formats.
Risk-aware output
Get JSON risk summaries, ranked threat paths, and graph-annotated issues.
Framework agnostic
Works out of the box with the orchestration frameworks you already use.
Plug-and-play CLI
One command, no config, no vendor lock-in.
Security Insights
Not just diagrams. Real vulnerabilities.
Agent-Wiz reveals escalation paths, logic flaws, and trust boundary violations your IDE will never catch.
Detects privilege escalation vectors across agents
Highlights implicit flows between memory, tools, and LLMs
Flags prompt confusion, ghost agents, infinite recursion
Outputs MAESTRO-structured reports for risk scoring and mitigation
Install Agent-Wiz. Start threat modelling your AI agents today.
Because you can't secure what you can't see.