Skip to main content
Protector Plus deploys as an inline runtime defence layer between AI-driven applications and their LLM endpoints. The platform is LLM-agnostic — supporting cloud-hosted models (OpenAI, Anthropic, Google Gemini), open-weight models (Llama, Mistral, AI Singapore SEA-LION), and customer-self-hosted models on the same architecture.

​
Logical layout

Protector Plus + ShieldPrompt architecture

ShieldPrompt™ protects prompts and responses end-to-end across first- and third-party LLMs; the AI Security Portal manages policy, surfaces analytics, and forwards events to the customer's SIEM.

​
Two services, both over HTTPS

Management Console

Web UI for tuning security policies, reviewing alerts, generating API keys, and configuring SIEM forwarding.

Guardrail Service API

HTTPS endpoints for prompt inspection (/input-check) and response inspection (/output-check). Can be called directly by the application or through an LLM gateway.

​
Design principles

No agents on web servers or inside the GenAI application. Integration is HTTP-level — fast, low-friction onboarding without infrastructure changes.
Designed for horizontal scaling. In the Kubernetes variant, additional capacity is provisioned through standard Helm chart configuration. Updates ship through CD pipelines — infrastructure as code.
The LLM-powered engine catches evasive threats in near-real-time. The Threat Vector Database provides a continuously updated library of malicious prompt patterns without requiring model retraining.
Role-based, fine-grained permissions support administration by multiple parties — typically separating the platform Super Admin from per-application owners.

​
Where to next

Deployment modes

Forwarding and non-forwarding topologies.

Detection layers

Standard Guardrails and ShieldPrompt.