Everything you need to evaluate AI agents
25+ Built-in Metrics
Response quality, RAG evaluation, agent tool use, multi-turn conversations. Every metric returns a 0–1 score with a human-readable reason.
JUnit 5 Native
@AgentTest, @Metric, @DatasetSource annotations. Evaluations run as ordinary test methods — no new tooling to learn.
LLM-as-Judge
OpenAI, Anthropic, Google, Ollama, Azure, Bedrock — plug in any provider. Research-backed G-Eval prompts, fully inspectable.
Framework Agnostic
Evaluates any Java AI agent. Optional auto-capture for Spring AI, LangChain4j, LangGraph4j, and MCP.
Local-First
No cloud, no SaaS, no data leaves your machine. Pair with Ollama for fully private evaluations.
CI/CD Ready
Standard JUnit XML output works with Jenkins, GitHub Actions, and GitLab CI. Maven and Gradle plugin included.
Install
Maven
<dependency>
<groupId>org.byteveda.agenteval</groupId>
<artifactId>agenteval-junit5</artifactId>
<version>0.1.0</version>
<scope>test</scope>
</dependency>
Gradle (Kotlin DSL)
testImplementation("org.byteveda.agenteval:agenteval-junit5:0.1.0")