Preview & evals

Use preview to interact with an agent in the product, then evals to check repeatable scenarios.

Preview

On an agent detail screen, Preview opens a chat (and optional browser talk) against that agent’s prompt, model, and Knowledge Base. Responses stream from the preview chat API. Optional RAG pulls relevant Knowledge Base chunks into context.

Voice preview uses Web Speech in the browser over the same text stream. It is for product testing — not a stand-in for production telephony providers.

Evals

Evaluations define scenarios and observable checks. A run:

1. Walks the scenario turns against the agent
2. Grades outcomes with the configured LLM
3. Stores pass/fail, reasons, and transcripts you can inspect

Use evals to catch regressions when you change prompts, knowledge, or first-message behavior.

Suggested workflow

1. Preview happy-path and edge questions manually
2. Encode the important cases as eval scenarios
3. Re-run after prompt or Knowledge Base changes
4. Only then share a public demo or wider rollout

Related

• Agents
• Knowledge Base
• Getting started

Start buildingTry a public democorvialabs.com