PenLocal-AI Security Architecture¶
Overview¶
PenLocal-AI is designed as an "Offline-First", secure, self-hosted AI pentesting environment. The architecture prioritizes data sovereignty, network isolation, and defense-in-depth to safely run autonomous security agents (Kali Linux) alongside sensitive vulnerability data.
Core Security Pillars¶
1. Network Isolation & Segmentation¶
We utilize a strict 3-tier network model to prevent compromised agents from pivoting to the host or backend services. - Networking Security Details: Full details on subnets, Docker networks, and host port binding.
2. Encryption Everywhere¶
- At Rest: Sensitive credentials and findings are encrypted using Fernet (AES-128-CBC) before storage.
- In Transit: All service-to-service and user-to-service communication is encrypted via TLS/SSL (Nginx Proxies).
- In Memory: Credentials used by agents are stored in RAM-only (
tmpfs) volumes and never written to disk.
3. Manager Webapp Hardening¶
The central command center is built with modern application security standards. - Manager Webapp Security: MFA, Strong Hashing (PBKDF2), RBAC, and Input Sanitization.
4. Component-Level Security¶
Each microservice is hardened according to its function:
- Credential Vault: Isolated secret management service.
- Ollama (AI): Proxied access to LLMs, resource limits.
- n8n (Workflow): Dual-homed network isolation.
- Minio (Storage): Per-pentest bucket isolation.
- Qdrant (Vector DB): Authenticated, tenant-isolated vector storage.