MSP AI Orchestrator - Autonomous Multi-Agent System

Fully autonomous AI-powered MSP management system using AWS Bedrock and the Strands Agents framework. This system operates 24/7 without human intervention, predicting IT problems 24-48 hours before they occur and automatically executing preventive actions.

🏗️ Architecture

Technology Stack

• Backend: Python FastAPI + Strands Agents (AWS official SDK)
• Frontend: React + TypeScript + Tailwind CSS
• AI Framework: Strands Agents SDK with AWS Bedrock integration
• Real-time: WebSocket connections for live updates
• Visualization: Recharts for analytics and performance tracking

8 Specialized AI Agents

1. Master Orchestrator Agent - Central command coordinating all sub-agents
2. Predictive Monitoring Agent - Predicts failures 24-48 hours in advance
3. Autonomous Decision Agent - Makes business decisions without human approval
4. Client Lifecycle Agent - Automates onboarding and client management
5. Resource Optimization Agent - Assigns technicians and optimizes schedules
6. Financial Intelligence Agent - Analyzes profitability and pricing
7. Security & Compliance Agent - Monitors security and remediates vulnerabilities
8. Learning & Adaptation Agent - Analyzes outcomes and improves models

🚀 Quick Start

Option 1: Simulation Mode (No AWS Required)

Run the system in simulation mode without AWS credentials:

# Start both frontend and backend
bash start.sh

The system will run in simulation mode with realistic agent orchestration.

• Frontend: http://localhost:5000
• Backend API: http://localhost:8000
• API Docs: http://localhost:8000/docs

Option 2: Full AWS Bedrock Integration

To use actual AWS Bedrock with Claude Sonnet:

Prerequisites

1. AWS Account with Bedrock access enabled
2. Claude Sonnet model access in AWS Bedrock console
3. AWS credentials configured

Setup AWS Credentials

**Method 1: Environment Variables **

Add these to your .env:

AWS_ACCESS_KEY_ID=your_access_key_here
AWS_SECRET_ACCESS_KEY=your_secret_key_here
AWS_DEFAULT_REGION=us-west-2

Method 2: AWS CLI Configuration

aws configure
# Enter your AWS Access Key ID, Secret Access Key, and Region

Enable Bedrock Model Access

1. Go to AWS Bedrock Console
2. Navigate to "Model access" in the left sidebar
3. Request access to Claude 3 Sonnet (us.anthropic.claude-sonnet-4-20250514-v1:0)
4. Wait for approval (usually instant)

Run with Bedrock

# Start with AWS Bedrock enabled
bash start.sh

The system will automatically detect AWS credentials and use Bedrock.

📊 Features

Real-Time Dashboard

• Live Agent Activity: Monitor all 8 agents with real-time status indicators
• Autonomous Decision Feed: Stream of auto-approved actions with ROI calculations
• Predictive Timeline: Visual timeline showing predicted issues 24-48 hours ahead
• Performance Analytics: Charts tracking accuracy, savings, and improvements
• Escalation Queue: Level 3 decisions requiring human approval

Autonomous Operation Levels

Level 1 - Full Autonomy (No human approval)

• Preventive maintenance <$2K
• Routine ticket routing
• Standard vulnerability remediation
• Client notifications

Level 2 - Conditional Autonomy (Auto-approve with notification)

• Actions costing $2K-$10K
• Service upgrades
• Security updates requiring downtime

Level 3 - Human-in-the-Loop (Requires approval)

• Actions >$10K
• Custom contract negotiations
• Major infrastructure changes

🛠️ Development

Backend (Python)

cd python_backend
python main.py

Strands Agents Tools: Each agent has specialized tools (AWS Bedrock Action Groups):

• monitoring_tools.py - analyze_system_metrics, predict_failure, calculate_business_impact
• decision_tools.py - evaluate_action_approval, calculate_roi, execute_approved_decision
• resource_tools.py - find_optimal_technician, optimize_maintenance_schedule
• security_tools.py - scan_vulnerabilities, auto_remediate_vulnerability

Frontend (React)

npm run dev

📦 Project Structure

├── python_backend/
│   ├── main.py                      # FastAPI server
│   ├── agents/
│   │   ├── strands_orchestrator.py  # Master orchestrator using Strands
│   │   ├── websocket_manager.py     # WebSocket connections
│   │   └── tools/                   # Agent tools (Bedrock Action Groups)
│   │       ├── monitoring_tools.py
│   │       ├── decision_tools.py
│   │       ├── resource_tools.py
│   │       └── security_tools.py
│   └── routes/                      # API endpoints
├── client/
│   └── src/
│       ├── components/              # React components
│       ├── pages/
│       │   └── Dashboard.tsx        # Main dashboard
│       └── lib/
│           └── mockData.ts          # Mock data for UI
└── start.sh                         # Startup script

🔐 Security & Compliance

• AWS Secrets Manager integration ready
• Audit trail for all autonomous decisions
• Rollback capability for failed actions
• Human override available for any autonomous decision
• Compliance monitoring (HIPAA, SOC2, GDPR)

📈 Success Metrics

• 95% decisions made without human intervention
• 80% problem prevention before client impact
• 24-48 hour prediction accuracy > 85%
• <5 minutes decision execution time
• <10% false positive rate
• 90%+ ROI positive on preventive actions

🌐 Production Deployment (Future)

Deploy to AWS Bedrock AgentCore for serverless, production-grade operation:

pip install bedrock-agentcore
agentcore configure --entrypoint python_backend/main.py
agentcore launch

📚 Resources

• Strands Agents Docs: https://strandsagents.com/latest/documentation/docs/
• AWS Bedrock: https://aws.amazon.com/bedrock/
• Strands GitHub: https://github.com/strands-agents/sdk-python