An AI-powered incident response engineer. It takes a problem description, analyzes logs, metrics, traces and code. Finds the root cause and suggests fixes. It helps prevent regressions and speeds up debugging end to end.

Features

• ReAct agent loop with optional multi-agent pipeline
• MCP tool protocol over stdio
• Semantic memory with pgvector similarity search
• Incident deduplication, severity classification (P1-P4)
• Slack notifications, alerting poller
• Prometheus and AWS CloudWatch integration
• Claude Desktop integration via MCP
• JWT auth with per-user incident tracking
• Next.js dashboard, CLI, REST API

Stack

TypeScript, Node.js, Express, Next.js, Google Gemini API, PostgreSQL + pgvector, MCP, Prometheus, AWS SDK v3, Docker

Quick start

npm install && cd dashboard && npm install && cd ..

Create .env:

GOOGLE_GEMINI_API_KEY=your_key
API_URL=http://localhost:3002
JWT_SECRET=your_random_secret

Start Postgres and run:

docker compose up -d
npx tsx services/checkout.ts    # terminal 1
npx tsx src/api.ts              # terminal 2
cd dashboard && npm run dev     # terminal 3

Generate traffic and open http://localhost:3000:

for i in {1..20}; do curl -s http://localhost:3001/checkout > /dev/null; done

Tests

npm test

Docs

See docs/ for detailed documentation:

• Architecture
• API reference
• CLI usage
• Claude Desktop setup
• AWS integration
• Environment variables