What G8KEPR is: the four security pillars

G8KEPR is a security platform for AI-powered applications. It protects every layer an AI request touches — your APIs, your agent tool calls, your LLM traffic, and the AI output itself — and ties them together into one traceable record.

What G8KEPR protects

Modern AI stacks introduce risk at several distinct layers. G8KEPR covers each one with a dedicated pillar:

• ▸API Security — Guards your API surface. Every inbound request is scanned by machine-learning threat detection and a curated rule set, then filtered through rate limiting, shadow API discovery, and a web application firewall.
• ▸MCP / AI Agent Security — Secures Model Context Protocol tool calls and agent sessions. It runs tools in an OS-level sandbox, enforces per-tool permissions, records sessions, and detects "rug-pull" behavior where a trusted tool changes to do something malicious.
• ▸AI Gateway — Sits in front of your LLM providers. It routes inference requests across providers, tracks cost, filters PII, and applies guardrail policies before a prompt ever reaches a model.
• ▸Verification Engine — Checks AI-generated output before it reaches your users. Four layers of validation confirm integrity, grounding, structure, and constraints.

How the pillars fit together

The pillars are not four separate products — they are stages a single request flows through in order:

1. 1.A request arrives and API Security scans it (firewall, rate limit, threat check).
2. 2.The AI Gateway routes it to an LLM provider and applies guardrails.
3. 3.If the request involves agent tool calls, the MCP layer intercepts and sandboxes them.
4. 4.The Verification Engine validates the generated output.
5. 5.The response is returned to the client.

User request
  → API Security
    → AI Gateway
      → MCP (if agent tools are involved)
        → Verification Engine
          → Response

Why the connection matters

Every step shares one correlation ID. That means a threat caught at the agent layer can be traced backward to the user request that started it and forward to the exact API response — a single, unified record instead of four disconnected logs.

Next steps

• ▸Set up your first protected endpoint in Getting Started
• ▸Learn how requests are traced end-to-end in Command Center