Using the Agent

Open the Agent

Go to Agent in the left navigation. Using the Agent requires the core/agents:chat permission; if you do not see it, ask an administrator to check your role.

The Agent page has three panels:

1. Conversations: your conversation history, grouped into Owned and Shared with you. Toggle it with the icon next to New Conversation.
2. Chat: the conversation and the message input.
3. Artifacts: the briefs, findings, tables, chart cards, and dashboard drafts in this conversation. Toggle it with the icon at the top right of the chat.

Drag the dividers between panels to resize them. A health banner appears above the chat when some Agent capabilities are temporarily limited.

Start a conversation

Click New Conversation, or type into the empty Agent page. The welcome screen suggests starter prompts:

• "What data do we have available? Show me the key tables and metrics."
• "Build me a dashboard to track our key performance metrics"
• "Query our data warehouse to analyze recent trends"
• "Show me all metric definitions and what they measure"

The first message becomes the conversation's title; rename it from the conversation's ⋯ menu.

Ask a focused question

The Agent works best when you include the metric, time period, breakdown, filters, and the output you want:

Create a table artifact showing net revenue, order count, and AOV by channel for the last 90 days. Sort by net revenue descending.

Follow-ups use the conversation's context, so you can continue with "Break that down by week" or "Turn this into a chart card." Start a new conversation when you switch to an unrelated question.

Reference dashboards and datasets

Where references are enabled for your workspace, type @ in the message box, or use + > Dashboards and datasets, to point the Agent at specific dashboards, Transformation Models, or datasets. You can reference up to 10 items in one message. Click a reference chip to open it read-only in the artifact workspace.

Read the answer

Each response can include:

• Reasoning & tool calls: expand it to see the steps the Agent took, such as Catalog searches, queries, and previews.
• Evidence: for analytical answers, the drawer lists the data sources, time range, filters, metric definitions, grain, freshness, query logic, assumptions, and caveats behind the result.
• Actions to Copy message, Regenerate response, or rate it Good response or Bad response, with optional written feedback for the Stackless team.

The Agent queries the warehouse read-only and returns at most 500 rows per query, within the schemas your role can access.

Approve actions

The Agent asks before doing anything consequential:

• Publishing a Transformation or Semantic Model: it shows a publish review and asks you to type the exact confirmation phrase.
• Publishing a dashboard: you confirm in the publish dialog.
• Applying a source schema change: it shows the plan and asks for confirmation.
• Support tickets: an approval card, such as Create this support ticket?, must be approved before the ticket is created or updated.

Work with artifacts

The artifacts in a conversation are part of the Agent's context. Select a table, chart card, brief, or dashboard draft in the workspace and ask about it:

• Select a table: "Explain the top five rows."
• Select a chart card: "Change this to the last 90 days."
• Select a dashboard draft: "Add a tab for customer retention."

Selecting a published dashboard you can edit creates an edit fork automatically, so your changes never touch the live dashboard until you publish them. See Artifacts.

Attach existing artifacts

Use Open existing artifacts (or the + tile in the artifact workspace) to attach artifacts from other conversations. Check the ones you need, preview them, then click Attach to this conversation. With no conversation open, the button becomes Start conversation and attach.

Use this to continue a prior analysis, modify an earlier dashboard draft, or combine chart cards into one dashboard.

Upload files

Click the paperclip (or + > Upload file) to attach a CSV or XLSX file. It is staged and sent with your next message. See Uploads.

Stop or leave a response

Use the stop button to end a response that is still running. Runs continue in the background if you navigate away, and the conversation reopens when you return. Switching to another conversation or starting a new one while a run is active asks you to confirm.

Share and duplicate conversations

From a conversation's ⋯ menu or the share icon, the owner can Share it with teammates by email. Shared conversations are read-only for recipients and appear under Shared with you. Anyone can Duplicate a conversation to get their own editable copy.

Sharing a conversation does not grant access to the data or dashboards in it. Recipients still see only what their roles allow.

Health and usage limits

• Some agent capabilities are limited appears when one of the Agent's services (the model provider, Catalog, warehouse, or semantic layer) is degraded. Answers that need that service may fail or be slower. The banner can be dismissed.
• If your workspace has a daily AI usage budget, an AI usage today bar appears as you approach it, and Daily AI usage limit reached when it is used up. The limit resets at midnight in your workspace's timezone.

When to ask for a durable output

Use chat for quick answers. Ask for an artifact or dashboard when the result will be inspected, exported, shared, or repeated:

• "Save this as a table artifact."
• "Create a chart card."
• "Write a brief with key metrics and supporting evidence."
• "Create a finding card with confidence and next action."
• "Build a dashboard from these results."

For recurring checks delivered to Slack or email, create a Monitor.