dbt Labs has open-sourced dbt Charts, a structured YAML language for declaring interactive dashboards in code, marking the latest step in what founder Tristan Handy calls “BI’s Second Unbundling.”
The shift addresses a practical friction: AI agents—increasingly the primary interface for data work—work fluently in code, SQL, and Git but struggle with traditional BI tool UIs. As Handy explained, early BI tools bundled everything: ingestion, transformation, compute, and visualization. Over the past decade, infrastructure layers split off to specialized tools (Fivetran for ingestion, dbt for transformation, cloud providers for compute). What remained in BI tools was primarily visualization and interactive interfaces. Now, as chat-based agents become more common, charts themselves need to move to where agents operate: into code.
Traditional code-generated dashboards create sprawling file structures (HTML, CSS, JavaScript, React apps) that are slow to audit and expensive to modify with each agent iteration. Conversely, BI tools that added copilots kept agents on narrow rails, limited to what the UI exposed. dbt Charts offers a third path: dashboards declared in a single, readable YAML file that agents can modify freely while remaining auditable and version-controlled.
In dbt Charts, SQL specifies what data to display, while YAML and Jinja define how to display it. The system includes over 1,100 config options across sixteen chart types, with cascading styles that inherit from boards and themes, keeping individual files concise. Charts are rendered to SVG, HTML, PNG, PDF, or even the terminal via the CLI.
When integrated with dbt projects, charts live in a charts/ directory alongside models/ in the same Git repository. A model change and its corresponding chart updates ship on one branch through one CI run, surfacing breaking changes before production deployment. The validation system flags problems early—layout warnings, overflow alerts—giving agents tight feedback loops for iteration.
dbt Labs is also launching dbtCharts.com, a hosted platform in public beta that layers hosting, access control, and conversational analytics on top of the open-source language. The platform includes a visual editor, version history, and permissioned sharing, allowing non-technical users to view dashboards without warehouse access. Every change—from chat, visual editor, or direct code—syncs to the same YAML in Git, preventing locked-in data stacks.
The open-source language is available under the Apache 2.0 license. Users can install locally without an account or provide it to coding agents via command line.
Key facts
- dbt Charts uses YAML to declare dashboards in code, designed for AI agents rather than UI-based BI tools
- The system integrates directly with dbt projects, allowing charts and models to version together in Git and fail on incompatible changes before deployment
- dbtCharts.com platform adds conversational analytics, visual editing, version control, and permissioned sharing on top of the open-source language
- The language supports 16 chart types with over 1,100 configuration options and cascading style inheritance to keep individual dashboard files readable