AgentOps
Observability for AI agents: session tracing, tool-call tracking, and cost analysis.
About
AgentOps is an observability platform built for AI agents, giving developers session-level tracing of agent runs, tool and LLM call recording, cost and latency analytics, and replay debugging. It is designed to sit alongside multi-agent frameworks such as CrewAI, AutoGen, and LangGraph, capturing what agents did, which tools they called, and where runs went wrong.
The platform grew out of the open-source AgentOps project and has become a widely used reference point in AI agent observability, with a community snapshot noting roughly 5,500 GitHub stars as of mid-2026. For teams shipping agentic workflows, it answers the basic operational questions: how much did this run cost, how long did it take, and exactly what the agent did step by step.
What it can do
- Session tracing for agent runs
- Tool and LLM call recording and replay
- Cost and latency analytics per run
- Integration with CrewAI, AutoGen, and LangGraph
Setup
Install the AgentOps SDK in your Python project and add your API key; it auto-instruments supported agent frameworks.
Prerequisites
A Python project using an agent framework such as CrewAI, AutoGen, or LangGraph.
Use cases
Works with
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 123; Trending score 72/100 from researcher assessment.
★ 5,541 GitHub stars
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.