Autonomous debugging and troubleshooting agent for production-grade systems from DevOps, SRE, and Operations perspective.

There are two separate architecture options and execution this application supports:

• OPTION A - Claude Code Mode for System Admins: Project-specific Claude Subagents and Agent SKILLS along with Default Claude Tools for SKILLS. These are organized under .claude folder (as per default project-specific agents).
• OPTION B - LangGraph-based Application: Structured Custom Developed LangGraph-framework based Agents with below Multi-agent architecture pattern with an Orchestrator Agent coordinating specialized sub-agents:
  • Observability Agent: Queries Grafana for metrics, dashboards, and alerts
  • Infrastructure Agent: Queries Kubernetes for cluster state, pods, services, and logs
  • Knowledge Search Agent: Queries Stackoverflow for troubleshooting knowledge
  • Incident Management: ServiceNow integration for incident tracking
  • Code Management: GitHub integration for issues and PRs

• LangGraph DeepAgent: Multi-agent orchestration, Deep Agent is n Agent harness built using LangChain as framework (for tools, model access) and LangGraph as Runtime (for checkpoints, memory)
• LiteLLM: Unified LLM provider access
• LangSmith: Observability and tracing
• MCP Servers (via Docker Hub):
  • Kubernetes MCP Server
  • Grafana MCP Server
  • Stackoverflow MCP Server
  • ServiceNow MCP Server
  • GitHub MCP Server

1. Install dependencies: uv sync
2. Configure environment variables (see .env.example)
3. Deploy MCP servers separately
4. Run agent: python main.py

See IMPLEMENTATION_PLAN.md for complete implementation details.

• LangGraph DeepAgent Quickstarts