Grade agent trajectories and tool-use decisions with AgentEvals

Score whether an agent took a sensible intermediate path, called tools correctly, and reached the outcome without relying only on final-answer checks.

Prerequisites

Python or TypeScript runtime, agent run outputs or trajectories, optional LLM judge provider

Installation

Use the upstream install or setup path that matches your environment:

• pip install agentevals
• npm install agentevals @langchain/core
• pip install openai
• npm install openai

Requirements and caveats from upstream:

• <summary>Python</summary>
• python
• Python Async Support

Basic usage or getting-started notes:

• To get started, install agentevals:

• <details open>

• bash

• Source: https://github.com/langchain-ai/agentevals

• Extracted from upstream docs: https://raw.githubusercontent.com/langchain-ai/agentevals/HEAD/README.md

Documentation

• https://github.com/langchain-ai/agentevals

Source

• Agent Skill Exchange