Steward Inbox

The Steward Inbox is where Steward surfaces issues and improvements it discovers in your semantic layer. Like application monitoring tracks uptime and errors, Steward continuously monitors the health of your semantic layer by analyzing your AI Analyst conversations — so you don't have to review them manually.

Steward is proactive: instead of waiting for you to find problems, it comes to you with what it found and a plan to fix it.

How It Works

Steward runs automatically in the background — no setup required. It monitors conversations your AI Analysts have with users and picks up on signals: issues flagged by the AI Analyst itself, user thumbs-downs, query errors, and moments where the analyst had to ask the user for clarification. These signals are grouped and analyzed to filter out noise, and what comes out are actionable inbox items — each with context and a proposed plan.

Items fall into three categories:

• Alerts — critical issues that are actively breaking things, like a column that was renamed in your warehouse

• Suggestions — actionable improvements with a clear next step, like adding a metric the analyst had to calculate manually

• Insights — observed patterns worth knowing about, like users frequently asking about data you don't have modeled yet

Items that go untouched expire after a period of time and are removed from your inbox automatically.

Working with Inbox Items

Each inbox item has a title, a description of what Steward found, and a proposed plan — a summary of the steps Steward will take to address it.

You can approve the plan to let Steward execute it right away, or reply to discuss it first. If you want Steward to adjust its approach, just tell it what to change — Steward updates the plan and presents it for your approval before executing.

If an item isn't relevant, dismiss it. Steward will ask for a reason: already fixed, not relevant, incorrect, or other. These reasons help Steward improve its future suggestions.