All guides

Using the Query Playground

Test your AI agent, explore knowledge base citations, and rate responses in the Query playground.

3 min read425 words

The Query page is your interactive testing environment. Use it to ask questions exactly as a customer would and see how the AI agent responds — including which wiki entries it cited.

Opening the Query playground

Click Query in the left sidebar.

Asking a question

  1. Type your question in the text area at the bottom of the chat panel.
  2. Press Enter or Ctrl+Enter to submit.
  3. The AI responds with an answer and a list of source citations.

The response is generated by retrieving the most relevant wiki entries and synthesising them into a coherent answer. If no wiki entries match the query, the AI will say so rather than fabricating an answer.

Reading source citations

Below the AI's answer you'll see a list of cited wiki entries:

  • Path — The wiki entry identifier
  • Title — The entry's title
  • Confidence — How closely this entry matched the query (%)

You can click any citation to open the full wiki entry. If a citation looks wrong, edit the entry to improve it.

Session management

The sidebar on the left shows your past sessions (conversation threads). Each session is a separate conversation:

  • Click any past session to reload its history
  • Start a new session by simply typing a question when no session is active — a new session is created automatically

Sessions are stored indefinitely so you can revisit past conversations to diagnose issues.

Rating responses

After each AI response you'll see two reaction buttons:

  • 👍 Helpful — The answer was accurate and useful
  • 👎 Not helpful — The answer was wrong, incomplete, or confusing

These ratings feed directly into your Analytics — the helpfulness score on the Analytics page is calculated from these reactions. Rate responses honestly to get accurate performance data.

Tips for testing

  • Test edge cases — Ask questions the AI might struggle with: ambiguous phrasing, niche topics, questions about content you haven't ingested yet.
  • Check for hallucinations — If the AI gives a confident answer with no citations, or cites an unrelated entry, that's a red flag. Investigate the wiki.
  • Test your most common support questions — Copy your top 10 support tickets and run them through the playground. This quickly reveals knowledge gaps.
  • Iterate quickly — Playground → identify gap → edit wiki entry → re-run query. Repeat until satisfied.

Related guides