🤖 Autonomous AWS cost optimization agent that uses LLM reasoning to identify and execute cost-saving actions across your cloud infrastructure.

Built for the AWS AI Agent Global Hackathon - demonstrating autonomous AI agents using Amazon Bedrock AgentCore, Claude 3, and AWS services.

• Amazon Bedrock Integration: Uses Claude 3 Haiku for cost-effective LLM reasoning
• AgentCore Runtime: Autonomous decision-making with multi-step planning
• Risk Assessment: Every recommendation tagged with confidence score and risk level
• Natural Language Reasoning: Agent explains WHY each action should be taken

• Real-time Monitoring: Continuous analysis of AWS Cost Explorer and CloudWatch metrics
• Smart Recommendations: Identifies idle resources, oversized instances, and inefficient storage
• Automated Actions: Can stop EC2 instances, resize RDS, archive S3 to Glacier
• Forecasting: 7-day cost predictions to prevent budget overruns

• Safe Mode: Recommend-only mode (default) for human approval
• Auto-Execute Mode: Fully autonomous with configurable risk thresholds
• Action Logging: All changes tracked in S3 for audit and rollback
• Confirmation Dialogs: UI prevents accidental actions

• Real-time cost metrics and trends
• Resource inventory with search/filter
• AI recommendation cards with impact analysis
• Action execution interface

Frontend (React + Vite)
    ↓
AWS Bedrock AgentCore (Claude 3 Haiku)
    ├── AWS Cost Explorer → Billing data
    ├── AWS CloudWatch → Resource metrics
    ├── AWS Lambda → Action execution
    └── Amazon S3 → Report storage

See ARCHITECTURE.md for detailed architecture diagram and component details.

• Node.js 18+ and npm
• AWS Account with:
  • Bedrock access (Claude 3 Haiku enabled)
  • Cost Explorer enabled
  • IAM permissions for EC2, RDS, S3, Lambda, CloudWatch

# Clone repository
git clone https://github.com/yourusername/autoinfra.git
cd autoinfra

# Install dependencies
npm install

# Configure environment variables
cp .env.example .env
# Edit .env with your AWS credentials

# Start development server
npm run dev

Create .env file:

VITE_AWS_REGION=us-east-1
VITE_AWS_ACCESS_KEY_ID=your-access-key
VITE_AWS_SECRET_ACCESS_KEY=your-secret-key
VITE_S3_BUCKET_NAME=autoinfra-reports
VITE_BEDROCK_MODEL_ID=anthropic.claude-3-haiku-20240307-v1:0

See DEPLOYMENT.md for complete AWS deployment instructions.

Sign up or log in with any email and password (demo mode uses local storage).

View real-time cost metrics, trends, and AI-generated recommendations.

• Browse all AWS resources (EC2, RDS, S3, EBS)
• Filter by type and state
• Search by name or ID
• View CPU utilization and costs

• Review AI recommendations
• Toggle Safe Mode for auto-execution
• Apply or dismiss individual actions
• Monitor savings

If AWS credentials are not configured, AutoInfra runs in demo mode with:

• Simulated cost data (30 days of trends)
• Mock resource inventory (8 resources)
• Rule-based recommendations (5 optimizations)
• Simulated action execution

This allows you to explore the UI without AWS setup.

AutoInfra meets all AWS AI Agent Global Hackathon requirements:

✅ LLM hosted on AWS Bedrock - Claude 3 Haiku
✅ Uses AgentCore primitive - Runtime for agent execution
✅ Reasoning LLMs - Structured prompts for decision-making
✅ Autonomous capabilities - Runs with/without human input
✅ Integrates external tools - Cost Explorer, CloudWatch, Lambda, S3
✅ AWS Lambda - Action execution
✅ Amazon S3 - Report storage
✅ API Gateway - (Optional) REST endpoints

See ARCHITECTURE.md for detailed compliance mapping.