An AI-powered incident response agent that automates on-call duties using AWS Bedrock, Temporal, and sponsor integrations.

FlowPilot is a SaaS agent that "stands in" for your on-call engineer. It:

1. Senses CloudWatch alarms via webhook
2. Reasons via AWS Bedrock (Claude or Titan)
3. Executes remediation actions (pod restarts via kubectl or Block/Goose)
4. Communicates with multi-language reports (DeepL), diagrams, and voice alerts (Rime)

Built on Temporal for durable workflows and wrapped in MCP so the LLM can directly call out to sponsor tools.

• 🧠 AI-Driven Decision Making: Uses AWS Bedrock (Claude 3 Sonnet) to decide whether to reboot or ignore alerts
• 🔄 Automated Remediation: Executes kubernetes commands to restart failing services
• 🌎 Multi-Language Reports: Automatically translates incident reports to French and German using DeepL
• 📊 Visual Incident Diagrams: Generates system diagrams of affected components
• 🔊 Voice Alerts: Converts incident reports to voice alerts using Rime
• ⏱️ Durable Workflows: Built on Temporal to ensure reliable execution even during failures
• 🖥️ Interactive Dashboard: Web UI to monitor incidents, view diagrams, and hear voice alerts

• Temporal: Core workflow orchestration with durable execution and retries
• AWS Bedrock: AI decisioning using Claude 3 Sonnet model
• DeepL: Multi-language translation of incident reports
• Block/Goose: Secure command execution (simulated in development)
• Rime: Voice alert generation for audible notifications
• MCP/guMCP Protocol: Standardized agent-to-agent communication

• Node.js 18+ and npm
• Temporal server (local dev or cloud)
• AWS account with Bedrock access
• Sponsor API keys (see .env configuration)

1. Clone the repository

git clone https://github.com/your-username/flowpilot.git
cd flowpilot

2. Install dependencies

npm install

3. Create a .env file based on required variables:

# AWS Bedrock Configuration
AWS_REGION=us-east-1
AWS_ACCESS_KEY_ID=your_aws_access_key
AWS_SECRET_ACCESS_KEY=your_aws_secret_key
BEDROCK_MODEL_ID=anthropic.claude-3-sonnet-20240229-v1:0

# DeepL API for multi-language translation
DEEPL_API_KEY=your_deepl_api_key

# Rime for voice alerts
RIME_KEY=your_rime_api_key

# Block/Goose (if using)
BLOCK_GOOSE_API_KEY=your_goose_api_key

1. Use our all-in-one demo script

chmod +x demo.sh
./demo.sh

Or run services individually:

1. Start Temporal server (in a separate terminal)

npm run dev:temporal

2. Start the worker (in a separate terminal)

npm run dev:worker

3. Start the API server (in a separate terminal)

npm run dev:api

4. Access the Web UI

open http://localhost:3000

5. Trigger a test incident

curl -X POST http://localhost:3000/alarm -H "Content-Type: application/json" -d '{"AlarmName": "high-cpu-usage"}'

The FlowPilot web dashboard provides a real-time view of your incident response system:

• Incident Monitoring: View all current and past incidents
• AI Decision Tracking: See what decisions were made for each alert
• Visualization: View system diagrams of affected components
• Voice Playback: Listen to voice alerts directly in the browser
• Quick Actions: Trigger test alerts to see the system in action
• Integration Status: See which sponsor integrations are active

Access the dashboard at http://localhost:3000 when the API server is running.

To test FlowPilot without external dependencies, you can:

1. Use the test script

npm run test

2. Or directly call the workflow

npx ts-node src/test.ts

For hackathon demonstrations, I've prepared several scenarios:

1. High CPU Usage: Simulates a service with excessive CPU consumption
2. Memory Leak: Demonstrates the system detecting and remedying a memory leak
3. API Failure: Shows how the system handles connectivity issues
4. MCP/A2A Protocol Demo: Demonstrates agent-to-agent communication

Run a full demo with ./demo.sh and select option 5.

• /src - Main source code
  • /activities.ts - Core activities with sponsor integrations
  • /agent.ts - Express API for receiving alerts
  • /worker.ts - Temporal worker configuration
  • /mcp.ts - MCP wrapper for activities
  • /workflows/diagnose.ts - Main workflow definition
  • /protocols/mcpAdapter.ts - MCP/A2A protocol implementation
  • /mcp-integration.ts - MCP demo functionality
• /public - Web UI dashboard
• /diagrams - Generated system visualizations (created at runtime)
• /audio - Generated voice alerts (created at runtime)

MIT