swapitsneil/agentops-center

AgentOps Center is an open-source operations platform built for teams running complex AI agent systems. As AI applications grow beyond simple chatbots into multi-agent workflows that call external tools, query databases, and pass tasks between specialized nodes, they become hard to debug. When a workflow takes 45 seconds or produces a bad result, engineers often cannot tell which agent node stalled, which tool call failed, or how much money each step consumed in token costs. This project replaces guesswork with recorded telemetry data, giving engineers real evidence to diagnose problems. The system instruments every LLM call, agent node transition, tool execution, token count, and runtime exception using OpenTelemetry GenAI semantic conventions. That telemetry data flows into SigNoz, an open-source observability platform backed by ClickHouse for storing traces, logs, and metrics. A component called the Evidence Engine correlates those traces and logs, then feeds findings to a Root Cause Copilot that provides diagnosis with confidence scoring rather than having a language model guess at answers without system context. The project also includes a chaos testing feature, letting teams deliberately inject failures into agent workflows to test resilience. A workflow replay capability allows engineers to re-run problematic executions. The README mentions a Model Context Protocol integration for querying SigNoz, and the system tracks AI cost intelligence, reporting per-agent token spending. The frontend is built with Next.js 15 and hosted on Vercel, offering a dashboard, a copilot interface, and a chaos testing UI. The backend runs on FastAPI with LangGraph for multi-agent orchestration, hosted on Render. A live demo and API documentation are publicly available, and the project was built for the Agents of SigNoz Hackathon 2026. The README provides a thorough problem statement and architecture diagram but the provided text cuts off during the section on Root Cause Copilot integration. The full README is longer than what was shown.