agentops

Why we included this project

Teams shipping LLM-powered agents to production rarely have a clear view of what their agents actually did and what those calls cost. AgentOps closes that gap with a lightweight Python SDK that drops into most agent frameworks, including CrewAI, LangChain, AutoGen/AG2, and OpenAI Agents, and starts recording sessions after a couple of lines of setup. Once running, you get step-by-step execution traces for debugging and per-provider spend tracking across foundation models, plus the ability to replay full sessions when a run goes wrong. That makes it useful in two situations: during development, when you are trying to reproduce a bad agent run, and after launch, when a cost spike or a silent failure needs an explanation. The dashboard and API backend are also self-hostable, which matters for teams that cannot send traces to a cloud service. If you manage agents at any real scale and currently rely on scattered logs, this is a practical observability layer rather than a monitoring bolt-on.