🕵️
No visibility
Unknown how many agents run — including unregistered Shadow AI — and which data they can access.
Private · Production-ready
Security and control for AI agents in enterprise infrastructure
AegisAI combines agent registry and Shadow AI discovery, OPA/Rego policies, Prompt Guard, hybrid ML anomalies, audit, HITL, and monitoring — so security teams can see autonomous agents and stop what they must.
Demo: demo@aegisai.com / demo123 · read-only mode
Why it matters
Without centralized control, AI agents become a security blind spot
🔓
No policies
Agent actions bypass GDPR, HIPAA, local privacy law, and corporate rules.
📊
No audit
Events are not recorded in a signed trail or forwarded to SIEM.
💉
Prompt injection
Jailbreaks and data leaks through prompts pass through unfiltered.
👤
No HITL
Critical operations execute without a responsible human in the loop.
Built for SecOps and platform teams
From Shadow AI discovery to SIEM export — one control plane for agents
🤖
Registry & search
Registration, heartbeat, semantic search, and predicted type / risk from embeddings.
🕵️
Shadow AI
Discover unregistered agents: fingerprinting, HITL confirm → register, batch clustering.
🛡️
Policy Engine
OPA + Rego: GDPR, HIPAA, privacy templates, HITL, and custom rules.
🧱
Prompt Guard
Regex, banlist, TF-IDF + Isolation Forest, and OPA guard.main on ingress.
📋
Audit & SIEM
Signed audit trail, JSONL/CSV, Syslog, CEF, Splunk HEC.
⚠️
Anomalies
Hybrid supervised / Isolation Forest / ONNX + rules, Redis hot-cache, p95 ≤ 300 ms.
✋
Human-in-the-Loop
Route risky actions to manual approval workflows.
📈
Monitoring
Prometheus, Grafana, alerts — API, Guard, anomaly latency, Shadow TTC, and policies.
🔌
SDK & integrations
Python SDK, LangChain/OpenAI (with discovery fallback), webhooks, SSO/OIDC, multi-tenant RBAC.
Product interface
Dashboard, Prompt Guard, HITL, and Grafana monitoring in one platform
Defense in depth
Multiple independent layers before an agent action executes
1
JWT / API key / SSO authentication
2
RBAC and tenant isolation
3
Prompt Guard middleware
4
OPA policy check + hybrid anomaly / risk score
5
HITL for critical operations
6
Signed audit + rate limits
Platform stack
FastAPI, OPA, Redis, PostgreSQL, Vite SPA, and observability built in
FastAPI
REST + WebSocket API
OPA / Rego
Policy decisions
PostgreSQL
Agents, policies, audit
Redis
Cache + Guard stats
Vite SPA
Dashboard + i18n
Prompt Guard
Rules + ML + OPA
Prometheus
/metrics endpoint
Grafana
Dashboards + alerts
25+
agents in demo
15+
active policies
10 000
audit records
3
demo orgs
Quick start
Run locally in minutes with Docker Compose
# Run (private repo access required)
cd AegisAI
cp .env.example .env
docker compose up --build -d
# Health check
curl -s http://localhost:8000/api/v1/health# Python SDK — Prompt Guard
from aegisai import AegisClient
client = AegisClient(base_url="http://localhost:8000")
result = client.guard.validate("user prompt here")
print(result.is_safe)See AegisAI in action
Open the interactive demo with seeded data or deploy the full stack on your own infrastructure.