Skip to content

Ask a question

Ask answers business questions in plain English. Every answer says what it is built on, so you know how far to trust it.

ZLens tries the most trusted source first:

  1. a certified block that answers the question,
  2. then semantic metrics from your semantic layer,
  3. then SQL written by AI from your tables, marked as needing review.

If none of these can answer safely, Ask refuses and says why.

Before you start

Ask needs an AI provider, even for certified answers. See Set up an AI provider.

A certified answer Asked how revenue changed by month, Ask answers from the certified monthly_revenue block: the answer is labelled Certified answer, names the block and its owner, and How it was answered shows each step.
  1. Open Ask from the left rail.
  2. Ask "How has revenue changed by month?".
  3. The answer is labelled Certified answer, from the monthly_revenue block, with its owner.
  4. The result table shows gross revenue and order count for each month.
  5. Switch the result to Visualization to see the monthly trend as a line chart.
  6. Open How it was answered to see the steps Ask took.

Ask

  1. In the left rail, click Ask.
  2. Type your question in Ask anything about your data…, for example Who are the top customers by lifetime spend?
  3. Press Enter. (Shift+Enter adds a new line.)

While Ask works, you see each step, such as Searched the project: 1 certified block, 24 metrics… and Asking the AI to read the question. Click Stop to cancel.

Read the answer

The answer starts with a trust line, then a short summary and the result table. For a question that a certified block answers:

  • The trust line reads Certified answer, followed by from and the block's name, and the block's Owner.
  • The result card says certified block.
  • Results shows the rows; Visualization shows a chart.

Other trust lines:

Trust line Means
Certified answer Ran a certified block exactly as reviewed.
Governed answer Built from semantic metrics and dimensions.
AI-generated answer AI wrote the SQL from your tables. Check it before relying on it. ZLens saves it as a draft block for review; see Answers that need review.
Needs clarification Ask needs one more detail. Answer the question it asks.
Couldn't run this query or No matching data Ask refused or found nothing, and says why. See When Ask refuses.

See how it answered

Under each answer, click How it was answered. A panel opens with tabs:

  • How it answered: the steps it took, such as Certified blocks, Semantic layer, Checked before running and Ran on the warehouse, each marked done or skipped.
  • DQL and SQL: exactly what ran.
  • Data used: the certified block, metrics, tables, joins and filters.
  • Checks: the checks run before the query.

See How Ask answered for the full trace.

Keep the answer

In the answer's panel:

  • Save as block saves the query as a draft block. See Save and certify.
  • Add to app adds the result to an App.

Narrow the question to one domain

Click Scope: All domains above the question box. Pick a Domain, and optionally a Subject area and a Purpose, then click Apply. Ask then searches only that domain's models, terms and skills.

Choose how hard Ask thinks

Click Answer settings (the sliders icon under the question box):

  • Thinking: Auto (default), Low, Medium or High. High cross-checks the number, so it is slower.
  • Research mode: for "why" questions that need several queries.
  • Model: shows the AI provider in use. Change it in Settings → AI provider.

Conversations

  • Each conversation is kept, so you can continue it later. They are listed under Recent; pin the ones you use often to Favourites.
  • New chat starts a fresh conversation.
  • Conversations are private to your machine and are not committed to git.

Ask from the terminal

zlens agent ask "Who are the top customers by lifetime spend?"

It prints the trust label (✓ Certified, ! AI-generated · review required or ✕ Blocked), the answer, and the thread id. Continue the conversation with --thread <id>. List conversations with zlens agent threads.

Your AI agent (Claude Code, Codex, Cursor) can use the same Ask through MCP. See Connect your AI agent.