Architecture Overview

Platform architecture

Stackless is a customer-specific analytics environment that combines managed ingestion, a cloud warehouse, managed transformation models, semantic models, Stackless dashboards, and the Agent.

Business Systems
  -> Source Connectors
  -> Customer Warehouse
  -> Transformation Models
  -> Semantic Models
  -> Dashboards, Agent Artifacts, Monitors, Exports

The Catalog and Operations sit across the whole stack. The Catalog gives the Agent and users a searchable map of assets and lineage. Operations runs the scheduler jobs that refresh models and shows their run history.

Customer environment

Stackless deployments are isolated by customer and environment. A production customer environment has its own application endpoint, authentication configuration, data access settings, warehouse configuration, and operational runtime.

This matters for users because:

• Your URL, users, roles, dashboards, and data access are scoped to your organization.
• The Agent only sees data that your role can access.
• Source configuration and warehouse schemas are managed per environment.
• Operational actions such as refreshes, publishes, and scheduled exports run in that environment.

Application layer

The Stackless application provides:

• The Agent conversation UI.
• Dashboards and the dashboard builder.
• Data pages for Sources, Catalog, Transformation Models, and Semantic Models.
• Monitors for scheduled analyses and checks, and Operations for scheduler jobs and runs.
• Support requests and feedback.
• Settings for users, roles, Slack, and Agent data access.

Data layer

The data layer includes:

• Source connector metadata and sync state.
• Warehouse tables and views.
• Managed Transformation Model definitions and materialized relations.
• Semantic Model definitions.
• Dashboard specifications, versions, published state, and access metadata.
• Monitor definitions, revisions, and run history.
• Agent conversations, artifacts, uploads, and run events.

Transformation layer

Transformation Models are managed dbt models organized in staging, intermediate, mart, and report layers. Most use a structured specification that supports filters, joins, aggregations, unions, flattening, dedupe, windows, pivots, unpivots, tests, and a refresh policy. When logic does not fit the spec, the Agent can write a versioned SQL model instead.

Semantic layer

Semantic Models define measures, dimensions, joins, and business names. They are the layer dashboards and Agent charting use to make a metric mean the same thing everywhere.

Semantic Models should usually reference published Transformation Models, especially when the business logic includes cleaned fields, enriched joins, dedupe, attribution, or custom metric rules.

Dashboard layer

Stackless dashboards query Semantic Models live. They start as private drafts, can be shared privately, and are published to the roles assigned to them. The dashboard system supports version history, replacement forks for editing published dashboards, filters, tabs, charts, KPI cards, tables, metric definitions, exports, scheduled email exports, and Agent-assisted edits.

Agent layer

The Agent can:

• Search the Catalog.
• Inspect dashboards, models, metrics, and lineage.
• Run read-only warehouse analysis within access limits.
• Create briefs, findings, table artifacts, and chart cards.
• Build or edit private dashboards.
• Draft and publish Transformation Models and Semantic Models after your confirmation.
• Set up and manage source connectors.
• Help prepare publish, refresh, or downstream-impact workflows.

The Agent is built to produce durable outputs, not just temporary chat answers. Monitors reuse the same Agent to run governed analyses on a schedule.

Operations and monitoring layer

Operations runs the scheduler jobs that refresh Transformation Models, keeps their run history and logs, and sends failure alerts to Slack, email, or a webhook. Monitors watch governed metrics and analyses for the business, and deliver results with a traceable run history. Source sync status lives on Data > Sources, and scheduled export history on each dashboard.

Security model

Stackless uses authentication, role-based permissions, per-role schema and dashboard access, and a workspace-wide list of indexed schemas to control what users, the Agent, Monitors, and MCP clients can see or change. See Security and Access for details.