Find the tools to understand why your agent fails
Updated 2026-10-12
An agent can produce a convincing answer while retrieving the wrong material or calling the wrong tool. Start with the failure you need to observe, then find tools that expose the evidence or test the behavior.
Explore evaluation frameworks, tracing platforms, retrieval checks and security testing. Each tool card records a documented capability, an official link and a source date so you can verify the next step.
Browse by quality task
Compare releases, inspect execution, check retrieval support or exercise adversarial scenarios.
Read the evaluation vocabulary
Understand traces, assertions, datasets and groundedness before designing a test.
Submit an agent quality tool
Share an official source for a free editorial entry or update an existing capability note.
Discuss a disclosed partnership
Propose a separate product showcase or review with a clear scope and disclosure.
Three different places to start
- Release checks: DeepEval supports Python evaluations integrated with pytest and CI/CD. [deepeval.com]
- Failure investigation: Phoenix exposes agent execution through traces and supports experiments built from trace datasets. [arize.com]
- Adversarial testing: Giskard documents tests for prompt injection, harmful content and personal information disclosure. [giskard.ai]