Glossary · Agentic AI
What is Evals? Definition and meaning
Test sets and graders that measure quality of a prompt, model or agent.
Definition
Evals: Test sets and graders that measure quality of a prompt, model or agent.
In short: Automated tests for model behaviour.
When it is used
Preventing regressions when prompts or models change.
Where it fits
Evals belongs to Operations in the Agentic AI topic. These are the other terms in the same group:
- Tracing: Captures prompts, tool calls, inputs and outputs of every step as a trace.
- Regression boards: A standing set of benchmark rows that is re-run in full after each change so that a new fix cannot silently break an old result.
- Cost and latency budgets: Explicit ceilings on tokens, tool calls and seconds for an agent run.
- Sandboxing: Executes code and tools inside a restricted container or virtual machine.
- Prompt-injection defense: Techniques that stop text in web pages, files or tool results from hijacking the agent.
Learn more
The full guide to Evals explains how it works, when to use it and how it compares with the alternatives, with examples. The glossary lists every term on the site from A to Z.
Previous term: EU AI Act · Next term: Evaluation metrics