AgentOps
A self-hosted observability platform that instruments LLM agents, ingests trace events, and visualizes them in a React dashboard with span trees, cost analytics, and decision-quality evals.
Project Details
AgentOps tracks the full lifecycle of an agent run: trace creation, planner reasoning, LLM calls, tool invocations, errors, token usage, and cost. The system is built as a monorepo with a custom Python SDK, a FastAPI ingestion service, a Redis-backed worker, a PostgreSQL store, and a React/Vite dashboard.
I added an in-browser playground of lightweight agents (trip planner, researcher, trade signal, diff summarizer, JSON wrangler, share page) plus a chat UI that runs tasks and links each run to its full trace. The hosted version is exposed through a Cloudflare Tunnel for easy demos.
Tech: Python SDK, FastAPI, PostgreSQL, Redis, React, Vite, Docker Compose, Caddy, Cloudflare Tunnel.
Live: agentops.website
Live: agentops.website
