Overview
Getting Started
Install the CLI, pull a local judge model, and run your first scan. The CLI and judge model run directly on your machine.
Prerequisites
Python 3.11+, uv, and Ollama running locally.
1 - Install the scanner
# Install package (creates .venv automatically) uv pip install -e . # For demo apps (Flask chatbots) uv pip install -e ".[demo]" # Verify CLI is available llm-scanner --help
2 - Pull a judge model
# Good balance of speed and quality ollama pull llama3.2:3b # Or a larger, more accurate judge ollama pull llama3.1:8b
3 - Run your first scan
llm-scanner \ --target http://localhost:5000/chat \ --target-type url \ --judge-model llama3.2:3b \ --format html \ --output-dir ./reports
Only have a public URL and don't want to install anything? A hosted, no-install public-URL scan (preview here) is on the roadmap - for now, run the CLI above against any public or private endpoint.
Next steps
- Wire the scanner into your pipeline - see CI/CD Integration.
- Look up any flag - see CLI Reference.
- Understand what each OWASP category tests - see OWASP LLM Top 10.