RAG Box is designed with security and privacy as core principles. This document outlines our security practices, how to report vulnerabilities, and our supported versions.

• 100% Offline Operation - No data leaves your infrastructure
• Local Processing - All document processing, embedding, and LLM inference happens locally
• No External APIs - Zero dependency on cloud services for core functionality
• Configurable Logging - Control what information gets logged

• Clean Architecture - Clear separation of concerns with security boundaries
• Input Validation - All user inputs are validated and sanitized
• Error Handling - Secure error messages that don't leak sensitive information
• Resource Management - Proper cleanup of temporary files and memory

We actively maintain security updates for the following versions:

We take security vulnerabilities seriously. If you discover a security issue, please follow responsible disclosure:

For critical security issues that could lead to:

• Remote code execution
• Data exfiltration
• Privilege escalation
• Service disruption

Contact us immediately:

• Email: security@galezra.com (will respond within 24 hours)
• Subject: [SECURITY] RAG Box Critical Vulnerability

For lower-severity issues like:

• Information disclosure
• Minor input validation issues
• Configuration security improvements

Use GitHub Security Advisories:

1. Go to the Security tab
2. Click "Report a vulnerability"
3. Fill out the form with details

When reporting a vulnerability, please provide:

1. Description - Clear explanation of the vulnerability
2. Impact - What could an attacker achieve?
3. Reproduction Steps - How to reproduce the issue
4. Environment - OS, Python version, RAG Box version
5. Proof of Concept - Code or screenshots (if safe to share)
6. Suggested Fix - If you have ideas for mitigation

Our security response process:

1. Acknowledgment - Within 24-48 hours
2. Investigation - We'll investigate and assess the impact
3. Updates - Regular updates on progress (at least weekly)
4. Fix Development - We'll develop and test a fix
5. Disclosure - Coordinated disclosure after fix is available
6. Recognition - Credit in security advisory (if desired)

# Verify Poetry installation
poetry --version

# Use virtual environments
poetry install --no-dev  # For production

# config/production.yaml
log_level: "WARNING" # Reduce verbose logging in production
debug_mode: false # Disable debug mode
data_dir: "/secure/path/data" # Use secure directory

• Restrict data directory permissions to application user only
• Use dedicated user account for running RAG Box
• Monitor file access logs for unusual activity
• Regular backups of vector store and configuration

• Firewall rules to restrict access to Streamlit port (8501)
• VPN access for remote users
• HTTPS termination if exposing web interface
• Rate limiting on API endpoints

# Input validation
from pathlib import Path

def safe_file_path(user_path: str) -> Path:
    """Safely resolve user-provided file paths."""
    path = Path(user_path).resolve()

    # Prevent directory traversal
    if not str(path).startswith(str(DATA_DIR.resolve())):
        raise ValueError("Path outside allowed directory")

    return path

# Check for known vulnerabilities
poetry audit

# Update dependencies regularly
poetry update

# Review new dependencies carefully
poetry show --tree

# Test input validation
def test_path_traversal_protection():
    """Test that path traversal attacks are prevented."""
    with pytest.raises(ValueError):
        safe_file_path("../../../etc/passwd")

We use automated security scanning:

• Bandit - Python security linter
• Safety - Known vulnerability scanning
• Semgrep - Additional security patterns

• Poetry audit - Check for known CVEs
• GitHub Dependabot - Automated dependency updates
• License compliance - Ensure compatible licenses

# Use minimal base images
FROM python:3.12-slim

# Run as non-root user
RUN useradd --create-home --shell /bin/bash ragbox
USER ragbox

# Security headers in web interface
ENV STREAMLIT_SERVER_ENABLE_CORS=false
ENV STREAMLIT_SERVER_ENABLE_XSRF_PROTECTION=true

• Model Provenance - Verify LLM model sources (Ollama, HuggingFace)
• Model Poisoning - Use trusted model repositories
• Resource Exhaustion - Monitor CPU/memory usage

• Malicious Documents - Sandboxed document processing
• File Size Limits - Prevent DoS via large files
• Format Validation - Strict file format checking

• Index Integrity - Regular integrity checks
• Access Control - Restrict vector store file permissions
• Backup Security - Encrypted backups of sensitive indexes

# Security-relevant events are logged
logger.warning("Failed authentication attempt", extra={
    "user_id": user_id,
    "ip_address": request.remote_addr,
    "timestamp": datetime.utcnow()
})

• Failed authentication attempts
• Unusual file access patterns
• Resource usage spikes
• Error rate monitoring

• Critical vulnerabilities: Patched within 24-48 hours
• High severity: Patched within 1 week
• Medium/Low severity: Included in next regular release

• GitHub Security Advisories - For all vulnerabilities
• Release Notes - Security fixes highlighted
• Email notifications - For subscribed users (coming soon)

• OWASP Python Security
• Python Security Best Practices
• AI/ML Security Guidelines

• Bandit Security Linter
• Safety Vulnerability Scanner
• Semgrep Static Analysis

For security-related questions that are not vulnerabilities:

• GitHub Discussions - Public security questions
• Email: security@galezra.com - Private security inquiries

Security is everyone's responsibility. Thank you for helping keep RAG Box secure! 🔒