BI as code
BI as code is the practice of defining business intelligence assets, such as dashboards, charts, metric definitions, and data connections, as versioned text documents instead of configuration saved inside a GUI.
It borrows the habits software teams already rely on: changes are diffs, every change has an author, a reviewer can approve or reject it, and a bad change can be reverted. Data teams already work this way for transformation code and pipelines; BI as code extends the same discipline to the layer people actually look at.
How it works in dvt
In dvt, each BI asset has a text form. A dashboard is a JSON dashboard spec. A data source is a named connection that panels reference by sourceId, with its credentials stored encrypted and never written into a spec. A visual identity is a set of design tokens in the spec’s theme. Every one of these is something a person can read in a review and an agent can write over MCP.
What dvt adds beyond “store it as text” is the review loop around the text. Every write is validated against the schema before it saves and can be previewed before it applies. Every save becomes an attributed revision with a visual diff. A reviewer can pin a comment to a single panel, and an agent can read the thread, make the edit, and resolve it. BI as code then means more than files in a repository: it means every change to what people see is checkable.