aegis-x5

AEGIS-X5
Autonomous Agent Governance Platform

Observe, guard, evaluate, and autonomously govern AI agent fleets — at scale.
From first trace to full autonomous loops, in under 2 minutes.

Quick Start · Modules · CLI · API · Templates · Request a Demo

---

Why AEGIS-X5?

Organizations deploying AI agents face a governance gap: once agents are live, there is no unified way to monitor their behaviour, enforce safety policies, detect drift, and respond autonomously.

AEGIS-X5 closes that gap with a single SDK that wraps any agent — Claude, GPT, LangChain, CrewAI — and provides real-time governance from day one.

---

Quick Start

1. Install

pip install aegis-x5

2. Initialize your project

aegis init

3. Add two lines to your agent

from aegis import Aegis

aegis = Aegis()  # local mode — no API key needed

@aegis.observe("my-agent")
@aegis.protect("safety-check", level="N2")
def my_agent(prompt: str) -> str:
    # your agent logic here
    return result

4. See your first metrics

aegis dashboard
# Open http://localhost:4005

That's it. Your agent is now governed — traces, latency, cost, and guard status visible in real time.

---

Modules

Autonomy Modes

---

Developer Mode vs Enterprise Mode

AEGIS-X5 operates in two modes depending on configuration:

---

CLI

aegis init          Create aegis.yaml in current project
aegis status        Show agents, traces, and stats
aegis dashboard     Launch local dashboard (port 4005)
aegis test          Run evaluations on golden set

---

API

REST API served on port 4000 (Docker) or embedded.

Authentication via X-API-Key header. See docs/api.md for details.

---

ML Predictive Analytics

AEGIS-X5 includes a built-in prediction engine (zero external ML dependencies):

• Health Score — 0-100 score per agent combining 7 weighted signals (latency, errors, cost, faithfulness, guard blocks, drift, memory)
• Drift Predictor — Predicts metric degradation 48h before critical threshold using linear regression + exponential smoothing
• Cost Forecaster — 7-day cost projection with spike detection and budget alerts
• Anomaly Detector — Z-score + IQR ensemble detection on sliding windows
• Calibration Tracking — Compares predictions vs reality (MAE, RMSE)

---

Templates

Industry-specific governance profiles that pre-configure validators, thresholds, and evaluation sets.

HSE (Health, Safety & Environment)

Built for occupational safety agents in Quebec/Canadian regulatory context:

• 4 specialized validators: SSTFactCheck, EPIValidator, CNESSTCompliance, HazardMinimizer
• Guard level N4 for safety-critical assertions
• Faithfulness threshold 97%
• 20-case golden set covering PPE, confined spaces, height work, hazardous materials, noise
• 6 pre-configured sources: CNESST, IRSST, APSAM, CCHST, ISO, OSHA
• Regulatory references: ISO 45001, OSHA 1910/1926, CNESST RSST, Loi 25, EU AI Act

from aegis.templates import load_template

tpl = load_template("hse")
# tpl.validators, tpl.golden_set, tpl.sources, tpl.regulations

---

Docker Compose

Full platform deployment with PostgreSQL, Redis, API, and Dashboard:

cp .env.example .env
make up

---

Examples

Ready-to-run examples in examples/:

• claude_agent.py — Claude agent with @aegis.protect
• langchain_rag.py — RAG pipeline with observe + evaluate
• crewai_team.py — Multi-agent team with guard on each agent
• openai_agent.py — OpenAI agent with observe
• fastapi_endpoint.py — API endpoint protected by guard

---

Architecture

aegis
├── core/        # SDK foundation, config, multi-tenant, tracing
├── observe/     # Telemetry, tokens, cost, latency metrics
├── guard/       # Validator pipeline, PII, injection, hallucination
├── evaluate/    # Metrics, drift detection, eval runner
���── collect/     # Source registry, web collection, scheduler
├── remember/    # Agent memory, provenance, GDPR erasure
├── predict/     # Health score, drift/cost/anomaly prediction
├── loops/       # Autonomous closed loops, orchestrator
├── templates/   # Industry profiles (HSE, ...)
├── dashboard/   # Local web UI
├── api/         # REST API (FastAPI)
├── local/       # SQLite standalone storage
└── cli.py       # CLI entry point

---

Request a Demo

AEGIS-X5 is a commercial platform by Preventera.

For enterprise licensing, custom templates, or a guided demo:

contact@preventera.com

---

© Preventera · GenAISafety · ReadinessX5™