Agentic Radar is a security scanner for agentic AI workflows, designed for developers, researchers, and security professionals. It provides a comprehensive HTML report detailing workflow visualizations, identified tools, MCP server usage, and mapped vulnerabilities against frameworks like OWASP Top 10 LLM Applications, enhancing transparency and security assessment for complex agent systems.

How It Works

The tool offers two primary commands: scan for static analysis and probe for runtime vulnerability testing. scan generates a report by analyzing code for agentic workflows, identifying components and potential security risks. probe executes simulated adversarial inputs against running agentic workflows to detect vulnerabilities such as prompt injection, PII leakage, and harmful content generation, aligning with OWASP LLM Top 10 threats.

Quick Start & Requirements

• Install via pip: pip install agentic-radar
• Optional extras: pip install agentic-radar[crewai] (requires Python >= 3.10 and < 3.13 for full functionality) or pip install agentic-radar[openai-agents].
• probe and prompt enhancement features require an OpenAI API key (set via .env or environment variable).
• Documentation: https://agentic-radar.readthedocs.io/en/latest/
• Demo: Google Colab Notebook available.

Highlighted Details

• Supports langgraph, crewai, n8n, and openai-agents frameworks.
• Runtime vulnerability probing includes tests for prompt injection, PII leakage, harmful content, and fake news generation.
• Optional prompt enhancement feature refines system prompts using LLMs.
• Generates detailed HTML reports with workflow graphs and vulnerability mappings.

Maintenance & Community

• Active development with a roadmap including support for more frameworks like LlamaIndex and AutoGen.
• Community channels available via Discord and Slack.

Licensing & Compatibility

• The LICENSE file is linked but not explicitly stated in the README. Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

• Prompt enhancement and probe features require an OpenAI API key and may send data to LLMs.
• CrewAI tool descriptions are limited on Python versions outside the 3.10-3.13 range.
• Framework support beyond the listed ones is under development.