Skip to main content
All work

AEGIS

Multi-Agent AI Security & Guardrail Engine

An autonomous seven-agent security pipeline that detects, exploits, patches, and validates software vulnerabilities. It combines Semgrep static analysis, AST policy evaluation, prompt-injection guardrails, and isolated Docker execution.

Documented outcomes

Guardrail latency
<24ms
Injection recall
99.8%
Cooperating agents
7

How the system moves

  1. Payload Ingest
  2. Semgrep AST Scan
  3. Memory Isolation
  4. Threat Evaluator
  5. Sanitized Output

Run complete · 5 stages

The interface

AEGIS interface

Trace a payload through the guardrail

Outcome: AST policy passes; guardrail short-circuits in <24ms with no isolation spin-up.

Stack

  • FastAPI
  • Python
  • LangChain
  • Semgrep
  • React
  • Next.js
  • Docker

Engineering note

The project separates detection, exploitation, patching, and validation into explicit roles so the security workflow stays inspectable from end to end.