Test Cases

Test cases are the building blocks of your evaluation. Each defines one scenario.

Fields

• Input — the message sent to your agent
• Expected output — what a correct answer looks like
• Category — maps to compliance controls (e.g. access_control)
• Context — source docs for RAG faithfulness checking

Import

Import from DeepEval, Langfuse, or OpenAI Evals formats via the API or the UI. Format is auto-detected.

Safety probes

POST /api/v1/projects/:id/probes generates adversarial test cases (jailbreak, prompt injection, PII leak) automatically.