SpendOptimo

Amazon AgentCore powered Multi-Agent Autonomous FinOps Platform

Comment 1

SpendOptimo – Multi-Agent Autonomous FinOps Platform

Inspiration

Organizations continue to face the challenge of managing and optimizing their cloud costs effectively. Traditional FinOps tools primarily offer dashboards and reports, leaving engineers to interpret insights and act manually. The inspiration behind SpendOptimo was to create an intelligent system that could analyze, recommend, and even execute optimization actions autonomously.

The goal: transform FinOps from reporting to decision and action automation. SpendOptimo is built to serve as a cloud cost co-pilot that not only tells you what to fix but can also fix it—safely, transparently, and within organizational policies.

What It Does

SpendOptimo is a multi-agent system that helps organizations manage and optimize AWS cloud costs intelligently and autonomously. Key capabilities include:

  • Conversational Querying: Users can interact with the system using natural language to get cost summaries, trends, and optimization insights.
  • Policy-Driven Cost Governance: Cost rules and compliance policies are codified to instantly detect violations (e.g., disallowed instance families or underutilized storage).
  • Automated Recommendations: The system provides actionable recommendations for EC2 rightsizing, Lambda optimization, and S3 lifecycle improvements.
  • Approval-Based Execution: Once recommendations are approved, SpendOptimo can automatically execute them via a workflow agent.
  • Secure and Auditable Actions: All executions follow least-privilege IAM policies and are logged for auditability.
  • Extensible Design: Adding new services (such as RDS or DynamoDB) involves minimal code changes through modular tools and configurations.

SpendOptimo moves beyond analysis—it enables autonomous, safe optimization within enterprise FinOps environments.

How We Built It

SpendOptimo is designed as a distributed, multi-agent architecture with clear separation between intelligence and execution layers.

Core Components

  • Analysis Agent (Nova Pro):
    Performs deep reasoning, cost analysis, and generates policy-aligned optimization recommendations.
  • Workflow Agent (Nova Lite):
    Handles action execution—starting, stopping, resizing, or applying lifecycle policies to resources.
  • API Layer:
    Built with Python (Starlette + Mangum), deployed as AWS Lambda functions, exposing a secure interface between UI and agents.
  • Web App:
    A React + Vite frontend providing an intuitive chat-based interface to interact with the agents.
  • Infrastructure as Code:
    Provisioned using AWS CDK v2 in TypeScript, managing IAM roles, Bedrock AgentCore integrations, Lambdas, and frontend deployment.

AWS Services Utilized

  • Models: Amazon Nova Pro & Amazon Nova Lite
  • Automation & Orchestration: Bedrock AgentCore, Strands
  • Cost & Optimization APIs: AWS Cost Explorer, Compute Optimizer
  • Security & Governance: IAM, Cognito, CloudWatch Logs

Repository Structure

infra/                 # CDK infra code
agentcore_runtime/     # Analysis Agent code (Python)
workflow_runtime/      # Workflow Agent code (Python)
api/                   # API Lambda orchestration
webapp/                # Chat UI (React)

This modular design ensures each component can evolve independently while staying aligned through defined APIs and event-driven workflows.

Challenges We Ran Into

Across the various components of SpendOptimo, several challenges were encountered and solved:

  • Safe Automation: Executing cost optimization tasks automatically in a production cloud environment required strict safety controls, IAM least-privilege principles, and built-in approval checkpoints.
  • Latency and Timeouts: Some optimization workflows exceeded API Gateway’s timeout limit. The solution was to implement asynchronous task handling and Lambda self-invocation patterns for long-running jobs.
  • Balancing Intelligence vs. Cost: High-intelligence reasoning increases cost and response time. Splitting responsibilities between Nova Pro (analysis) and Nova Lite (execution) ensured both efficiency and scalability.
  • Policy Codification: Translating business rules and optimization guidelines into reusable, machine-interpretable policies was an early hurdle but became a core strength of the solution.
  • Extensibility: Maintaining modularity while introducing new service modules required consistent contracts between tools, prompts, and workflow handlers.

Accomplishments That We’re Proud Of

  • Delivered a fully functional multi-agent architecture for intelligent cost analysis and workflow automation.
  • Created a chat-driven FinOps assistant capable of natural-language interactions and contextual responses.
  • Implemented end-to-end automation from analysis to execution with user approvals and audit logging.
  • Designed the system to be policy-aware and extensible, enabling new services and rules to be added easily.
  • Achieved secure execution through fine-grained IAM and logging mechanisms.
  • Reduced manual FinOps effort from hours per week to seconds per query through automation and recommendations.

What We Learned

  • FinOps automation is achievable when policies, analysis, and execution are tightly coupled through AI agents.
  • Multi-agent design provides flexibility, cost control, and better scalability than monolithic systems.
  • Clear separation between reasoning (high-intelligence) and execution (low-latency) workflows yields better performance and reliability.
  • Safe automation depends as much on how actions are verified and logged as on what is automated.
  • Extensible architecture with well-defined service interfaces simplifies long-term maintenance and innovation.
  • Real value emerges when AI systems don’t just visualize data but autonomously act on insights.

What’s Next for SpendOptimo

The roadmap for SpendOptimo focuses on broadening capabilities and enterprise readiness:

  • Service Expansion: Add support for RDS, EBS, and DynamoDB optimizations.
  • Advanced Workflows: Implement scheduled actions, optimization history tracking, and cost-saving summaries.
  • User Experience Enhancements: Introduce notifications, real-time progress updates, and richer visualization of savings.
  • Enterprise Integrations: Add RBAC-based multi-user support, organization-level cost policies, and cross-account optimization.
  • Predictive Optimization: Incorporate anomaly detection and forecasting for proactive spend management.
  • Cross-Cloud Enablement: Extend beyond AWS to support multi-cloud FinOps use cases.

SpendOptimo aims to evolve from a powerful prototype to an enterprise-grade, autonomous FinOps platform that continuously learns, optimizes, and enforces best practices—without requiring constant human oversight.

Built With

  • agentcore
  • amazon
  • api
  • bedrock
  • cdk
  • cdn
  • gateway
  • lambda
  • llm
  • nova
  • sdk
  • strands
Share this project:

Updates