CloudOps Multi-Agent System

Important: This project is intended for educational and demonstration purposes only. It is not intended for use in production environments without additional hardening. See Security Hardening for Production below and the AWS Solutions Guidance Disclaimer.

A hierarchical multi-agent system for AWS cloud operations, built on Amazon Bedrock AgentCore with Strands Agents SDK and an AG-UI streaming Next.js frontend.

Ask it things like “What were my AWS costs last month?”, “Are any critical AWS Health events open?”, “How resilient is my Direct Connect topology?” — and get structured answers, interactive visualisations, and downloadable reports.

---

Architecture

Three layers:

• Supervisor receives the user request, loads memory, delegates to a domain agent.
• Mid-level agents (FinOps, Governance, Ops Excellence) route to exactly one specialised leaf per request.
• Leaf agents call AWS APIs via Lambda tools through the AgentCore Gateway.

All agents run on AgentCore Runtime containers. Inter-agent communication is SigV4-signed HTTP; the supervisor speaks AG-UI to the browser. All agent behaviour is defined in a single config file — src/agents/hierarchy.json — so adding or changing agents is a JSON edit, not a code change.

For system internals (memory, tool tracing, build hashing, gateway sync), see docs/architecture.md.

---

What’s included

• Hierarchical agent orchestration with memory, follow-up suggestions, and structured reports.
• 12 AWS tools wired through AgentCore Gateway: Cost Explorer, CUR (Athena), Cost Optimization Hub, Discounted Commitments (Savings Plan / RI sizing and expiry), Billing/Anomalies, Pricing, Health Events, Network Resilience, Tag Governance, Lambda Runtime, CloudWatch, and AWS DevOps Agent.
• AWS DevOps Agent investigations for explicitly requested Health events and CloudWatch alarm incidents, with durable conversation cards, a compact activity panel, and mitigation outcomes. The deployment flow asks for the target Agent Space ID; no account or Agent Space ID is embedded in source.
• Health events pipeline with rules-based risk scoring, Claude Haiku 4.5 narrative enrichment, and optional Organisation-wide view for 1000+ account orgs. See docs/agents/health-events.md.