Skip to main content
AgentX is a platform for testing and observing AI agents. It lets you:
  • Record agent traces from any framework (LangChain, CrewAI, OpenAI Agents, Anthropic, plain Python) and see them in the Live Traces tab.
  • Evaluate agent quality against test datasets, scoring responses with LLM judges, vector similarity, and Jaccard similarity.
  • Gate releases with CI/CD: block merges or deploys when your agent’s pass rate drops below a threshold.

Quickstart

Send your first trace in under 5 minutes

Python SDK

Decorator and context manager tracing

CI/CD Evaluation

Gate releases with eval test sets

Agent Evaluation SDK

Score any agent with the Python SDK

​
How it works

AgentX is three independent capabilities built on one SDK and one dataset model. Use any one on its own, or combine them. Tracing: record what your agent did, viewable in the Live Traces tab.
Evaluation: score your agent’s outputs against a test dataset with an LLM judge and similarity metrics.
CI/CD: gate a merge or deploy on a pass rate threshold.
A trace and an evaluation result can also be linked, so a score comes with a full execution timeline instead of just a number. See linking a trace to an evaluation result. AgentX sits outside your agent in all three cases. There is no prompt injection and no changes to your agent’s logic. Tracing wraps your agent function with a decorator or context manager; evaluation and CI/CD call your agent function directly and score whatever it returns.

​
Base URL

All API paths in this documentation are relative to this base URL.