Skip to main content
Tuner’s eval library has 27 predefined evals in four categories. Each eval runs an LLM judge against the call to check a specific behavior. Add any of them to an agent from the library, or write your own for behavior specific to your use case. New agents start with three evals from this library: Hallucination, Prompt injection resistance and Politeness.
The Add evals library with search, Create custom eval, category filters and predefined eval cards marked PASS/FAIL

​
Accuracy

Checks that the agent captured and confirmed information correctly.

​
Actions & Tools

Checks that tools and workflows were invoked correctly.

​
Conversation Quality

Measures how well the agent communicates throughout the call.

​
Safety & Compliance

Checks that the agent stays within required guardrails.

​
Add predefined evals to your agent

In Agent Settings > Evaluation Rules, click Add in the Evals section to open the library. Filter by category or search, select the evals you want, and click Add Eval. See Create evals for the full steps.
Predefined evals run on every production call and every simulated call. The same criteria score both.

​
Next steps

Evaluate agents with per-call instructions

Give evals the runtime instructions each call was made with.

Red flags and labels

Turn eval results into tags you can filter and alert on.