Test
Preview
Version 1.0 · Last updated 2026-06-25
Pushing a change to a live agent without checking it first is the fastest way to embarrass yourself. Preview lets you replicate any customer situation in a controlled space, ask the questions you care about, and see exactly what the agent does.
Key concepts
Conditions
The setup that shapes the Preview session: Audience, topic, Living Context state, and surface (marketing site, product, helpdesk). Same conditions a real customer would arrive with.
Conversation panel
Where you type messages and see the agent respond. Identical to what a customer sees.
Inspection panel
Side panel that shows what the agent is seeing internally: Sources used, Procedures available, Living Context values. Use this to diagnose unexpected answers.
Session log
Every Preview session is saved automatically. Share a link with a colleague, or come back later to review.
What you can do here
- Type questions and watch the agent respond
- Trigger Procedures to see them run end to end
- Switch Audience, topic, or surface to test different paths
- Inspect which Sources the agent used
- Share session URLs with colleagues for review
When to use it
- After updating a Source, to confirm the change shows up correctly
- Before publishing a new component or Moment configuration
- Before any pilot review meeting, as a quick sanity check
- When the agent is giving a confusing answer in production and you want to reproduce the issue
When not to use it
- You want to test at scale. Use Simulations instead. Preview is one human, one session.
- You want to score the agent against expected answers. Use Quality reports.
How it works
Preview runs against the same Engine the live agent uses. Living Knowledge, Memory, Context, and Procedures all behave the same way. The only difference is that Preview sessions do not affect Analyze metrics for the live agent.