AUTONOMOUS AGENT GOVERNANCE

Govern your AI agents
before they govern you

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

Live Demo Quick Start
748+
Tests Passing
22
Platforms
512
Agents Governed
9
Modules

End-to-end agent governance in one SDK

Every agent action flows through a complete governance lifecycle — from observation to autonomous correction.

1. OBSERVE 2. GUARD 3. EVALUATE 4. COLLECT 5. REMEMBER 6. PREDICT 7. LOOPS

Built for every role in AI governance

AI Engineers

Add 2 lines to any agent. Get tracing, cost tracking, and guard rails instantly. Supports LangChain, CrewAI, OpenAI, Anthropic.

Safety & Compliance Officers

Pre-built HSE template with CNESST, ISO 45001, OSHA compliance. Guard level N4 blocks dangerous assertions in real time.

CTO / VP Engineering

Health Scores per agent fleet. Cost forecasting, drift prediction, anomaly detection. Full dashboard with autonomous loop control.

DevOps / Platform Teams

Docker Compose deployment. PostgreSQL + Redis + FastAPI. CI/CD with GitHub Actions. REST API with 12+ endpoints.

Data Scientists

ML predictive analytics (pure stdlib). Linear regression + exponential smoothing. Calibration tracking with MAE/RMSE.

Product Managers

Dashboard with 5 views. Agent-level KPIs. Guard event timeline. Cost projections. Export-ready for board presentations.

7 modules, one unified SDK

OBSERVE

Real-Time Telemetry

Distributed tracing, token counting, cost calculation, latency metrics (p50/p95/p99). Automatic span parent-child tracking.

GUARD

Policy Enforcement

N1-N4 severity levels. PII detection, injection prevention, hallucination checks. HITL approval gates for high-risk decisions.

EVALUATE

Quality Assessment

Faithfulness, relevancy, context precision. Golden set testing. Drift detection with warning/critical thresholds.

COLLECT

Feedback Ingestion

Structured sources with confidence scoring. Web collection. Scheduled polling. Domain-based registry.

REMEMBER

Memory Management

Per-agent memory with namespace isolation. PROV-O provenance tracking. GDPR/CCPA erasure compliance.

PREDICT

ML Predictive

Health Score 0-100. Drift prediction 48h ahead. Cost forecasting 7 days. Z-score + IQR anomaly detection. Zero external deps.

LOOPS

Autonomous Control

Drift auto-correct, guard auto-tune, latency auto-scale. 3 autonomy modes: monitor, semi-auto, full-auto.

Self-correcting governance — detect, correct, validate, learn

Each loop follows a 4-phase cycle. In semi-auto mode, high-risk corrections require human approval.

DriftAutoCorrect

Detects faithfulness drift below threshold. Triggers retraining pipeline. Updates baseline after validation.

HIGH RISKHITL GATE

GuardAutoTune

Adjusts guard thresholds based on false positive rates. Reduces alert fatigue while maintaining safety.

MEDIUM RISK

LatencyAutoScale

Detects p95 latency above threshold. Applies model fallback or throttling. Tracks load patterns.

LOW RISK

Two lines to govern any agent

Claude
OpenAI
LangChain
CrewAI
Webhook
from aegis import Aegis

aegis = Aegis()

@aegis.observe("claude-agent")
@aegis.protect("safety", level="N3")
def ask_claude(prompt):
    client = aegis.wrap_anthropic(Anthropic())
    return client.messages.create(...)
from aegis import Aegis

aegis = Aegis()
client = aegis.wrap_openai(OpenAI())

# Every call is now traced automatically
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": prompt}]
)
from aegis import Aegis

aegis = Aegis()
handler = aegis.langchain_handler()

llm = ChatAnthropic(
    model="claude-sonnet-4-20250514",
    callbacks=[handler]
)
from aegis import Aegis

aegis = Aegis()
mw = aegis.crewai_middleware()

researcher = mw.wrap_agent("researcher", research_fn)
writer = mw.wrap_agent("writer", write_fn)

# Full delegation chain tracked
// Node.js, Go, or any language
const response = await fetch("http://localhost:4000/webhook/trace", {
  method: "POST",
  body: JSON.stringify({
    agent: "my-agent",
    model: "gpt-4o",
    tokens: 1250
  })
});

See it in action

Run the simulator, then launch the dashboard to see SHIELD-OPS-X5 with 22 platforms and 500+ agents.

# 1. Install
pip install aegis-x5

# 2. Simulate 1400+ traces
python scripts/simulate_shield_ops.py

# 3. Launch dashboard
aegis dashboard

# Open http://localhost:4005

Simple, transparent pricing

Start free locally. Scale to enterprise when ready.

Developer

Free
Local mode forever
  • SQLite local storage
  • All 7 modules
  • CLI + Dashboard
  • 5 framework connectors
  • Community support
Get Started

Enterprise

$2,500
/month unlimited
  • Everything in Professional
  • Unlimited workspaces
  • Custom industry templates
  • SSO / SAML integration
  • Dedicated support engineer
  • On-premise deployment
  • SLA 99.9% uptime
  • Compliance audit reports
Contact Sales

Latest from AEGIS-X5

WHITEPAPER

State of AI Agent Governance 2026

7-page analysis of the governance gap in enterprise AI deployments. Why 73% of organizations lack agent oversight.

Download PDF →
WHITEPAPER

Closed-Loop Governance Architecture

Technical deep-dive into autonomous feedback loops for AI agent fleets. Detect-correct-validate-learn cycle.

Download PDF →
ARTICLE

Why Your AI Agents Need Guard Rails

Lessons from production: how a single unguarded agent caused $50K in bad recommendations in 48 hours.

TUTORIAL

HSE Governance in 10 Minutes

Step-by-step: deploy AEGIS-X5 with the HSE template for occupational safety agents in Quebec context.

CASE STUDY

SHIELD-OPS-X5: 22 Platforms, 512 Agents

How Preventera governs 512 AI agents across 22 specialized HSE platforms with AEGIS-X5.

BENCHMARK

Agent Governance SDK Comparison 2026

AEGIS-X5 vs. LangSmith, Arize, Braintrust: feature matrix, latency overhead, and total cost of ownership.

22 AgenticX5 platforms governed

Each platform is a specialized agent fleet for occupational health and safety.

EDGY

122 agents · AI Safety & Governance

LiteraCIA

122 agents · Intelligence & Analysis

EmergencyOps

61 agents · Emergency Response

RegWatch

25 agents · Regulatory Intelligence

TrainForce

20 agents · Safety Training

VisionGuard

18 agents · Computer Vision

RiskMatrix

18 agents · Risk Assessment

FireWatch

15 agents · Fire Safety

IncidentIQ

14 agents · Incident Management

WellnessAI

12 agents · Worker Wellness

ChemWatch

12 agents · Hazardous Materials

ConfinedAI

10 agents · Confined Space

AuditShield

10 agents · Compliance Audits

ErgoSense

8 agents · Ergonomics

HeightSafe

8 agents · Fall Protection

EnviroGuard

8 agents · Environmental

LockoutPro

7 agents · LOTO Procedures

NoiseMap

6 agents · Noise Monitoring

ContractorSafe

6 agents · Contractors

SafeFleet

5 agents · Fleet Safety

PPETracker

5 agents · PPE Management

+ Custom

Your platform here

Proven in production

HSE

Mining Company — 200 agents

Reduced safety incidents by 34% after deploying AEGIS-X5 guard rails on all AI-assisted safety recommendations. Guard N4 blocked 127 dangerous assertions in the first month.

CONSTRUCTION

General Contractor — 85 agents

Achieved ISO 45001 compliance for AI-assisted safety documentation. Faithfulness score maintained above 97% with drift auto-correction.

REGULATORY

Provincial Regulator — 50 agents

CNESST compliance validator caught 43 non-conforming recommendations that would have bypassed manual review. Zero false negatives.

FLEET

Transport Company — 30 agents

Cost forecasting predicted a 40% cost spike 5 days before it happened, allowing model fallback. Saved $12K/month in LLM costs.

What our users say

AEGIS-X5 caught a hallucination that told a worker asbestos was safe to handle with just a paper mask. That single guard block justified the entire investment.
Marie-Claire Bouchard
HSE Director, Mining Corp
We went from zero visibility into our AI agents to a full governance dashboard in one afternoon. The DX is remarkable — pip install, two lines, done.
David Chen
VP Engineering, TechSafe AI
The drift predictor alerted us 36 hours before faithfulness dropped below threshold. The auto-correct loop fixed it before any user was affected.
Jean-Francois Tremblay
AI Lead, QC Construction
We tried LangSmith and Arize. AEGIS-X5 is the only one that combines observability, guard rails, AND autonomous correction in one SDK.
Sarah Kim
CTO, AgentForce Startup
The CNESST compliance validator is perfectly calibrated. It catches every regulatory gap without false positives. Our legal team loves it.
Pierre Lavoie
Compliance Officer, Provincial Gov
512 agents across 22 platforms, all governed from one dashboard. We finally have a single pane of glass for AI safety.
Mario Desjardins
Founder, Preventera

Works with every AI framework

Anthropic
wrap_anthropic()
OpenAI
wrap_openai()
LangChain
langchain_handler()
CrewAI
crewai_middleware()
Webhook
Any language
OpenTelemetry
OTLP export

Built for regulated industries

ISO 45001 OSHA 1910/1926 CNESST RSST EU AI Act Loi 25 (Quebec) GDPR CCPA SIMDUT / WHMIS PROV-O

Everything you need

Quick Start

Install, init, instrument, dashboard — in 4 steps. Under 2 minutes.

Read Guide →

Module Reference

Detailed docs for all 7 modules with configuration examples.

Read Docs →

REST API

12+ endpoints with request/response examples and auth guide.

API Reference →

Templates

HSE template deep-dive. How to create custom industry templates.

Templates Guide →

CLI Reference

aegis init, status, dashboard, test — all commands documented.

Changelog

Full version history from v0.1.0 core to v0.3.0 production.

View Changelog →

Release history

v0.3.0 — Phase 9 LATEST

Dashboard 5 vues, connectors (LangChain/CrewAI/OpenAI/Anthropic/Webhook/OTEL), SHIELD-OPS-X5 tenant, HSE template, Docker Compose, REST API 12+ endpoints, 748 tests

v0.2.0 — Phase 3-4

Developer experience: CLI, local SQLite, mini dashboard. ML predictive: Health Score, drift prediction, cost forecasting, anomaly detection

v0.1.0 — Phase 1-2

Core SDK: multi-tenant, tracing, guard N1-N4, observe, evaluate, collect, remember. Autonomous closed loops with HITL gates

Request a Demo

See AEGIS-X5 governing 512 agents in real-time. Personalized demo for your use case.

contact@preventera.com

Response within 24 hours