SentinelFlow

SentinelFlow is a microservices-based platform for policy-aware AI orchestration.
It demonstrates how governance, risk, and compliance (GRC) controls can be natively integrated into AI workflows, ensuring that every model call is auditable, policy-driven, and aligned with enterprise requirements.

Traditional AI pipelines focus on speed and accuracy but often lack policy enforcement, risk awareness, and evidence logging. SentinelFlow fills this gap by embedding Microsoft Purview–style governance and AI Management System (AIMS) controls directly into the runtime path.

By doing so, SentinelFlow provides:

• 🔒 Security and Governance → All requests evaluated by a Policy Enforcement Point (PEP) against organizational policies before execution.
• 📊 Observability and Auditability → Every decision and model interaction is logged into an Evidence Store for compliance reporting.
• ⚖️ Risk and Impact Awareness → Dynamic checks against risk registers and Data Protection Impact Assessments (DPIAs).
• ⚙️ Pluggable AI Runtime → Works with Retrieval-Augmented Generation (RAG), tool brokers, and hosted models in a microservices design.
• ☁️ Cloud-Native Design → Deployable on any Kubernetes distribution (EKS, GKE, AKS, or on-prem), with CI/CD pipelines via GitHub Actions and GHCR container registry support.

In short, SentinelFlow is not just an AI service layer—it is a trust fabric that ensures every AI decision is explainable, compliant, and enterprise-ready.

---

✨ Features

• Gateway → Entry point, routes requests and enforces decisions
• PEP → Policy Enforcement Point, delegates to PDP
• AIMS (PDP) → Policy Decision Point, evaluates rules & logs evidence
• RAG Service → Adds retrieval-augmented context before inference
• Tools Service → Stub for external APIs/tools
• Models Service → Simple hosted model stub (echo + context)
• Audit/Evidence → Evidence store via AIMS (/v1/evidence)

---

📚 References

• Burns, B., Grant, B., Oppenheimer, D., Brewer, E., & Wilkes, J. (2016). Borg, Omega, and Kubernetes. Communications of the ACM, 59(5), 50–57. https://doi.org/10.1145/2890784
• European Commission. (2021). Proposal for a Regulation laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:52021PC0206
• Floridi, L., & Cowls, J. (2022). A unified framework of five principles for AI in society. Harvard Data Science Review, 4(1). https://papers.ssrn.com/sol3/papers.cfm?abstract_id=383132
• ISO. (2023). ISO/IEC 42001: Artificial Intelligence Management System (AIMS). International Organization for Standardization. https://www.iso.org/standard/81230.html
• Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V., Goyal, N., ... & Riedel, S. (2020). Retrieval-augmented generation for knowledge-intensive NLP tasks. Advances in Neural Information Processing Systems (NeurIPS). https://arxiv.org/abs/2005.11401
• Microsoft. (2023). Microsoft Purview: Unified data governance. Microsoft Docs. https://learn.microsoft.com/purview
• NIST. (2023). Artificial Intelligence Risk Management Framework (AI RMF 1.0). National Institute of Standards and Technology. https://doi.org/10.6028/NIST.AI.100-1
• OECD. (2023). OECD Framework for the Classification of AI Systems. Organisation for Economic Co-operation and Development. https://oecd.ai

---

🗂️ Architecture

flowchart TB
  subgraph User["User & Apps"]
    A1[Microsoft 365]
    A2[Teams]
    A3[SharePoint]
    A4[LOB Apps]
  end

  subgraph Gateway["Gateway Service"]
    G1[/Route API/]
  end

  subgraph PEP["Policy Enforcement Point"]
    PEP1[/Decide/]
  end

  subgraph AIMS["AI Management System (PDP)"]
    AIM1[Rules]
    AIM2[Risk Register]
    AIM3[Evidence Store]
  end

  subgraph Runtime["Runtime Services"]
    RAG[RAG Service]
    TOOLS[Tools Broker]
    MODELS[Model Service]
  end

  A1 --> G1
  A2 --> G1
  A3 --> G1
  A4 --> G1

  G1 --> PEP1
  PEP1 --> AIM1
  AIM1 --> AIM3
  PEP1 -->|allow| RAG
  PEP1 -->|deny| AIM3

  RAG --> TOOLS
  TOOLS --> MODELS
  MODELS --> G1

Loading

---

SentinelFlow — Quick Ops Cheat Sheet

📦 Local Development (Docker Desktop)

# from repo root
docker compose up --build -d

# health checks
curl.exe http://localhost:8080/healthz   # gateway
curl.exe http://localhost:8084/healthz   # pep
curl.exe http://localhost:8090/healthz   # aims
curl.exe http://localhost:8081/healthz   # rag
curl.exe http://localhost:8082/healthz   # tools
curl.exe http://localhost:8083/healthz   # models

✅ Test requests (via Gateway)

# Allowed path
curl.exe -X POST http://localhost:8080/v1/route `
  -H "Content-Type: application/json" `
  -d "{\"user_id\":\"u1\",\"prompt\":\"Draft a friendly email (no secrets).\",\"sensitivity\":\"Public\",\"labels\":[\"Public\"]}"

# Denied path (policy hit)
curl.exe -X POST http://localhost:8080/v1/route `
  -H "Content-Type: application/json" `
  -d "{\"user_id\":\"u1\",\"prompt\":\"my password is 123\",\"sensitivity\":\"Confidential\",\"labels\":[\"Confidential\"]}"

# Evidence log (AIMS)
curl.exe http://localhost:8090/v1/evidence

☸️ Kubernetes Deployment (images in GHCR)

# login to GHCR
docker login ghcr.io -u <USERNAME> -p <TOKEN>

# build & push all services (example loop)
$services = @("gateway","pep","aims","rag","tools","models")
foreach ($s in $services) {
  docker build -t ghcr.io/<org>/<repo>/$s:dev ".\services\$s"
  docker push ghcr.io/<org>/<repo>/$s:dev
}

🧭 Helm deploy (any K8s cluster/context)

# namespace + chart install/upgrade
helm upgrade --install platform .\charts\platform `
  -n prod --create-namespace `
  --set global.registry=ghcr.io `
  --set global.owner=<org>/<repo> `
  --set global.tag=dev

# verify rollout and service endpoint
kubectl -n prod rollout status deploy/sentinelflow-gateway
kubectl -n prod get svc sentinelflow-gateway

# (optional) quick access via port-forward
kubectl -n prod port-forward svc/sentinelflow-gateway 8080:80
curl.exe http://localhost:8080/healthz