Stackless Data/Docs

AI Agent Overview

What the Stackless Agent can do, what it needs permission for, and how it creates durable analytical outputs.

What is the Stackless Agent?

The Stackless Agent is a conversational workspace for exploring data and building analytics assets. You ask in plain English; the Agent searches the Catalog, queries governed data, and produces durable outputs such as artifacts, models, and dashboards that persist beyond the conversation.

Agent workspace with conversation history, a chat about weekly revenue performance, and the brief it produced in the artifact panel

You can use the Agent from:

  • The Agent page, for everything below.
  • The Transformation Model Agent and Semantic Model Agent panels on the Data pages, which focus on modeling work for that page.
  • Monitors, where each run's analysis is an Agent conversation you can open and continue, including from replies in the Monitor's Slack thread.
  • External AI clients through the MCP server.

What the Agent can do

AreaThe Agent canNeeds
DiscoverySearch the Catalog, describe columns, trace lineage up and downstream, check freshness and connector status.core/agents:chat
AnalysisRun read-only warehouse queries (up to 500 rows) within your schema access, and analyze uploaded CSV or XLSX files.core/agents:chat
ArtifactsCreate briefs, finding cards, table artifacts, and chart cards.core/agents:chat
DashboardsBuild, edit, fork, preview, and publish dashboards, and restore earlier versions.core/dashboards:write
Scheduled exportsCreate and manage email schedules for dashboards.core/dashboards:schedule_exports
Transformation ModelsDraft, sample, publish, and refresh models.core/dbt:write
Semantic ModelsCreate, preview, and publish models.core/models:write
SourcesInspect connectors, start a new connection through the secure setup page, sync, pause, resume, change schedules, and apply schema changes.core/sources:read / :write
SupportCreate, reply to, and update your own support tickets, after you approve each change.core/agents:chat

Without the write permission for an area, the Agent can still read and explain that area but will not change it.

What the Agent does not do

  • Write to the warehouse. Every query is read-only.
  • Publish anything without your explicit confirmation, typed or clicked.
  • Accept passwords, API keys, or webhook URLs in the chat. Credentials go through the secure connector setup page.
  • Manage users, roles, or Settings.
  • Edit legacy read-only dashboards.
  • Read schemas or dashboards your role cannot see.

The Agent's context

In a conversation, the Agent knows the Catalog assets your role can see, your visible dashboards and models, the conversation so far, the artifacts in the conversation (especially the one you have selected), and any uploaded files.

The artifact workspace

The right side of the Agent page holds the conversation's durable outputs: briefs, finding cards, table artifacts, chart cards, and private dashboard drafts. A table can be downloaded, a chart card copied to a dashboard, and a dashboard published for recurring reporting. See Artifacts.

Artifact workspace holding a draft dashboard, a chart card, a table, and a brief, with the brief open

How a typical workflow goes

  1. You ask a question or request a build.
  2. The Agent searches the Catalog and existing models.
  3. It runs read-only analysis or drafts the object you asked for.
  4. It returns an answer with its evidence, an artifact, a model draft, or a dashboard draft.
  5. You review the result, the Reasoning & tool calls, and the Evidence.
  6. If the output should persist or be shared, you save, attach, publish, or schedule it.

Example workflows

Explore existing reporting

  • "What dashboards do I have access to?"
  • "Explain how this dashboard's revenue metric is calculated."
  • "Which upstream models feed this KPI?"

Answer a question with a table

  • "Show revenue by channel for the last 30 days as a table artifact."

Build a dashboard

  • "Build a dashboard for weekly revenue, orders, AOV, and active customers."
  • "Add channel and country filters."

Create model-backed reporting

  • "Create a Transformation Model for customer-level daily revenue."
  • "Run a sample and show me the rows."
  • "Create a Semantic Model on top of it, then build a dashboard from it."

Connect a new source

  • "Connect our HubSpot account." The Agent creates the connection and sends you to the secure setup page.

Review before you rely on it

The Agent can make mistakes. Review its evidence, and verify important results against the underlying model, dashboard, or table, especially when:

  • A change affects a published model or dashboard.
  • A metric is used in executive reporting.
  • The analysis uses uploaded files.
  • The result includes sensitive data.

Next steps

PreviousAdministration
NextUsing the Agent