aegis-x5 0.3.0

Autonomous Agent Governance — observe, guard, evaluate, and govern AI agent fleets at scale.

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

Module	Purpose	Key Features
Observe	Real-time telemetry	Distributed tracing, token counting, cost calculation, latency metrics
Guard	Policy enforcement	PII detection, injection prevention, hallucination checks, N1-N4 severity levels
Evaluate	Quality assessment	Faithfulness, relevancy, context precision, golden set testing, drift detection
Collect	Feedback ingestion	Structured sources, web collection, confidence scoring, scheduled polling
Remember	Memory management	Agent memory, provenance tracking (PROV-O), GDPR/CCPA erasure compliance
Predict	ML predictive analytics	Health Score 0-100, drift prediction 48h ahead, cost forecasting, anomaly detection
Loops	Autonomous control	Drift auto-correct, guard auto-tune, latency auto-scale, HITL approval gates

Autonomy Modes

Mode	Behaviour
monitor	Observe and alert only — no automated actions
semi-auto	Auto-correct low-risk; human approval for high-risk
full-auto	Fully autonomous governance with closed-loop control

---

Developer Mode vs Enterprise Mode

AEGIS-X5 operates in two modes depending on configuration:

	Developer (Local)	Enterprise (Cloud)
Setup	Aegis() — zero config	Aegis(workspace="org", api_key="ak_...")
Storage	SQLite (~/.aegis/local.db)	PostgreSQL + Redis
Dashboard	aegis dashboard (port 4005)	Hosted platform
Cost	Free for local use	Commercial license
Scaling	Single machine	Multi-tenant, multi-workspace

---

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.

Method	Endpoint	Description
POST	/api/v1/trace	Record a trace span
POST	/api/v1/guard/validate	Validate content through guard pipeline
GET	/api/v1/health	Health check (public)
GET	/api/v1/agents	List agents with health scores
GET	/api/v1/predictions	Active predictions and accuracy
GET	/api/v1/stats	Aggregate trace statistics
GET	/api/v1/traces	Recent traces

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

Service	Port	Description
API	4000	REST API + Swagger docs
Dashboard	4005	Real-time monitoring UI
PostgreSQL	5432	Persistent storage
Redis	6379	Cache + sessions

---

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:

[e-mail]

---

© Preventera · GenAISafety · ReadinessX5™