Core Concepts

The short version

Stackless turns source data into trusted analytical outputs through a chain of governed objects:

Sources -> Warehouse Tables -> Transformation Models -> Semantic Models -> Dashboards, Monitors, and Agent Artifacts

The Agent can work across that chain, but each object has a different purpose.

Source

A source is an upstream system Stackless reads from. Examples include commerce platforms, CRM systems, accounting tools, ad platforms, databases, spreadsheets, files, and APIs.

Source data is usually raw or lightly normalized. It is valuable, but it is not always ready for reporting because names, grains, IDs, time zones, and business rules may differ by system.

Warehouse table

A warehouse table is where source or modeled data physically lives. Many warehouse tables are technical assets. You may see them in the Catalog, but most business dashboards should not point directly at raw warehouse tables.

Transformation Model

A Transformation Model is a managed dbt model. It defines how raw or intermediate data becomes a cleaner dataset, in a staging, intermediate, mart, or report layer. Most are structured specs; the Agent can also write SQL models for logic a spec cannot express. Transformations can:

• Select and rename fields.
• Cast data types.
• Join sources.
• Deduplicate records.
• Flatten nested JSON.
• Aggregate facts to a stable grain.
• Pivot or unpivot data into the shape reporting needs.
• Add tests such as uniqueness, not-null checks, and accepted values.

Published Transformation Models become reusable inputs for Semantic Models, dashboards, and future transformations.

Semantic Model

A Semantic Model gives business meaning to a table or transformation. It defines:

• Measures such as revenue, orders, margin, active customers, or conversion rate.
• Dimensions such as date, location, channel, product, customer segment, or campaign.
• Joins between models.
• Human-readable titles and descriptions.

The most reliable pattern is to build Semantic Models on top of published Transformation Models, because the transformation layer holds the cleaning and business logic.

Dashboard

A dashboard is a persistent analytical page that queries Semantic Models live. Dashboards can include KPIs, charts, tables, tabs, filters, exports, and scheduled email exports. They start as private drafts, can be shared privately, and are published to the roles assigned to them. Published dashboards are the recurring reporting layer your team returns to.

Artifact

An artifact is a saved output created or attached inside an Agent conversation. Stackless supports:

• Briefs.
• Finding cards.
• Table artifacts.
• Chart cards.
• Draft dashboards.

Artifacts are useful when a result should survive the chat response. They can be reviewed, exported, reused, and attached to other conversations, and chart cards can be copied into dashboards.

Catalog

The Catalog is the searchable inventory of visible Stackless assets. It helps users and the Agent answer questions like:

• What data exists?
• Which model defines this metric?
• What upstream sources feed this dashboard?
• What downstream dashboards depend on this transformation?

Monitor

A Monitor is a governed analysis or metric check that runs on a schedule or after a dataset refreshes, and delivers its result to Slack, email, or a webhook. Monitors pin the exact models they use and keep a traceable run history. See Monitors.

Operations

Operations is the technical scheduler: the jobs that refresh Transformation Models, their run history and logs, and failure alerts. Use it when a model did not refresh or a dashboard looks stale. See Operations.

Role and access

Stackless access is role-based. Your roles control which actions are available, which warehouse schemas you can see, and which published dashboards you can open. The Agent, your Monitors, and any MCP clients you connect act with your access. See Security and Access.