RAMPART

Security & Safety 💻 Python ⚖️ MIT 🟢 Actively maintained
403 stars

RAMPART is a pytest-native safety and security testing framework for agentic AI applications. It helps developers write and run tests for adversarial attacks, benign failures, and harm categories with evaluation-driven assertions. It is designed for developers building AI agents who need structured safety testing.

✨ Key features

  • Pytest-native integration for seamless test writing
  • Covers adversarial attacks and benign failures
  • Supports broad range of harm categories
  • Evaluation-driven assertions for clear pass/fail
  • Structured framework for agentic AI safety testing

🎯 Use cases

  • Testing AI agents for adversarial prompt attacks
  • Validating agent behavior against benign failure scenarios
  • Assessing compliance with harm category policies
  • Integrating safety tests into existing pytest suites

📦 Installation

🧰 Requirements: Requires Python and pytest; no API keys or accounts mentioned.

pip install ramp-art

(Note: The README does not provide explicit installation steps; this is inferred from typical Python packages. If not stated, use empty string.)

🚀 Usage

# Example from README (if any)
# No explicit usage example provided in README.

❓ FAQ

What is RAMPART?

RAMPART is a pytest-native safety and security testing framework for agentic AI applications.

What types of tests can I write with RAMPART?

You can write tests for adversarial attacks, benign failures, and a broad range of harm categories.

How does RAMPART integrate with pytest?

It is pytest-native, meaning it integrates seamlessly with pytest for writing and running tests.

Does RAMPART require any external services?

The README does not mention any required API keys or accounts.

📊 Repository

Stars★ 403
Forks🍴 51
Open issues🐛 11
Last commit🕒 Sep 3, 2026
Created📅 Mar 2026
Language💻 Python
License⚖️ MIT

🤖 Overview, features, install steps and FAQ were generated from the project's README on Sep 4, 2026. Always check the original source before running commands.