dreadnode/dyana

What it solves

Dyana provides a secure, isolated environment to load and profile the behavior of potentially untrusted files, such as machine learning models, serialized Pickle files, and executables. It prevents the host system from being compromised while allowing developers to analyze how these files interact with the system.

How it works

The tool uses Docker containers as a sandbox and integrates with Tracee to monitor and profile the system. It tracks specific activities including GPU memory usage, network requests, filesystem interactions, and security-related events during the execution of the loaded files.

Who it’s for

It is designed for researchers and developers who need to run and profile machine learning models or other executable files in a controlled environment to understand their resource usage or security implications.

Highlights

• Sandbox execution using Docker.
• Profiling for ML models, ELF executables, and Pickle files.
• Monitoring of GPU memory and network activity.
• Security event tracking via Tracee integration.