Aegis Analysis & SDK Docs
Complete guide for integrating Aegis security analysis, multi-layer policy enforcement, and real-time LLM risk scoring.
Documentation Topics (10)
Architecture & Security Layers
Overview of Aegis multi-layer defence-in-depth security model for AI agents.
Aegis provides a runtime security shield for LLMs and autonomous agents. Every prompt, tool execution, and response passes through three non-blocking security layers to guarantee real-time risk assessment and policy compliance.
Layer 1: Input Guard
Jailbreak detection, prompt injection shielding, system prompt isolation, and static regex policies.
Layer 2: Execution Gate
Dynamic tool authorization matrix, argument inspection, risk scoring, and stateful Human-in-the-Loop (HITL) approval barriers.
Layer 5: Cloud Telemetry
Asynchronous event streaming to Aegis Cloud dashboard with audit logs, risk timelines, and token consumption metrics.
Non-blocking Architecture: Aegis executes Layer 1 checks synchronously under 5ms, while Layer 5 telemetry is dispatched in background worker tasks without adding HTTP latency.