A zero-trust local AI agent security testing lab powered by NVIDIA OpenShell and Ollama to safely analyze code, prevent secret leaks, and isolate agent execution.
| Step 1: Target Input 📥 | Step 2: OpenShell Security Engine 🛡️ | Step 3: Verified Report 📊 |
|---|---|---|
Run python main.py (automatically initializes target inputs/ folder if needed). |
Performs static pattern scanning & sandbox attack vector simulations. | Generates structured security reports in the outputs/ folder. |
Install NVIDIA OpenShell, local Ollama engine, and python dependencies using these single-line PowerShell commands:
pip install -r requirements.txt; python main.pywinget install Ollama.Ollamawsl --install -d Ubuntu; wsl --set-default-version 2💡 WSL2 Setup: Windows requires WSL2 container configuration. Run
wsl --install -d Ubuntuin PowerShell (Admin) to initialize kernel isolation for NVIDIA OpenShell.
- 🔒 Kernel Isolation: Contains local AI agent file and process access inside strict container boundaries.
- 🛡️ Default-Deny Networking: Blocks unapproved socket connections and external credential leaks.
- ⚡ 100% Offline Autonomy: Pairs with local Ollama models for zero cloud API costs and total privacy.
- 📊 Automated Report Generation: Automatically audits scannable files and outputs executive Markdown & HTML security reports.
- Untrusted Code Inspection: Safely test third-party open-source AI scripts before executing them on host machines.
- Secret Leakage Prevention: Verify local AI agents cannot access environment variables or private host API keys.
- Network Exfiltration Audit: Ensure code agents do not make unapproved outbound web requests.
- Local Model Evaluation: Benchmark how local Ollama models perform when restricted by security policies.
- CI/CD Security Gate: Run automated security checks on agent code inside containerized build pipelines.
NVIDIA OpenShell/
├── main.py
├── README.md
├── requirements.txt
└── .gitignore
Built with NVIDIA OpenShell policy engine principles and Ollama local model inference.
NVIDIA OpenShell Ollama AI Agent Security Kernel Sandbox Zero Trust AI LLM Safety WSL2 Container Code Execution Sandbox Data Leak Protection Local LLM Isolation