Skip to content

How Ask answered

Every Ask run is recorded on your machine. You can see, for any answer, which sources Ask looked at, what it skipped, the SQL that ran and the checks it passed.

From an answer

  1. Under the answer, click How it was answered.
  2. The panel opens on How it answered. It starts with the outcome, such as Answered from a certified block., and a line like Worked for 52 s · 6 steps · 1 AI call.

  3. Read the steps in order:

    • Read the question: how Ask understood it.
    • Certified blocks: whether a certified block answers it.
    • Semantic layer: whether metrics and dimensions could answer it.
    • Governed tables and AI wrote SQL: used only when nothing more trusted could answer.
    • Checked before running and Ran on the warehouse.
    • Answer, or No answer with the reason.

    Each step is marked done or skipped. Click a step to see its details, such as the AI calls and the SQL it wrote. 4. Use the other tabs for the detail: DQL, SQL, Data used (the certified block, metrics, dimensions, tables, joins and filters) and Checks. 5. Click Open full trace for the complete record.

Ask observability

In the left rail, under Govern, click Ask observability. It lists every local Ask run, newest first.

Filter the list by:

  • Outcome: completed, blocked, failed, cancelled and more.
  • Mode: Ask or Research.
  • Selected tier: Certified, Semantic, Governed relational, Exploratory SQL, Clarify or gap.
  • Trust: Certified, Governed, Review required, Blocked.

Click Open trace on a run. The trace page has three views: Trace Tree (the steps, nested), Agent Graph (how the steps depend on each other) and Timeline (where the time went).

The trace keeps what each AI call was for, how long it took and its reply, the checks, the SQL that ran and a sample of its rows. It does not keep full prompts, credentials or file paths.

From the terminal

zlens agent trace list
zlens agent trace show <trace-id>

To share a run with a maintainer without sharing your data, export a strict bundle. It replaces identifiers with stand-ins and refuses raw SQL, URLs, paths and secrets:

zlens agent trace export <trace-id> --out ./ask-trace-strict --profile strict
zlens agent trace validate ./ask-trace-strict

Compare two runs, for example before and after a fix:

zlens agent trace compare ./before ./after --mode compare

Command details: Ask traces reference.