Charts Built for Chat
dbt Charts
dbt Labs open-sourced dbt Charts, a YAML language for dashboards. It keeps AI-made charts auditable instead of a mess of files.
Based on reporting by dbt Charts — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
dbt Labs has open sourced dbt Charts, a declarative language meant to bring dashboards back under control when they’re built with chat. The pitch is simple: let agents make reports, but keep the result in one readable YAML file instead of a sprawl of HTML, CSS, JavaScript and app code.
That matters because the current choices are all awkward. AI can sketch a report quickly, but once it grows into something real, tracing it gets messy. BI tools, meanwhile, keep copilots inside a closed interface, which is tidy but limiting. dbt Charts is aiming for a third path: code-level freedom with a format that people can still audit.
The language uses SQL for the data and YAML for the presentation. Markdown handles prose, and Jinja handles variables and macros. A single board file can declare variables, queries, charts and rows, and the CLI can render it to SVG, HTML, PNG, PDF or even the terminal. It can also serve a folder of boards as a site.
dbt Labs says the language already covers more than 1,100 config options, across sixteen chart types and composed charts built from them. Styles cascade from chart to board to theme, and one board can extend another, so a house style or standard report can be reused instead of rebuilt. The company also says support for the dbt Semantic Layer is planned.
The deeper hook is the dbt integration. Charts can live next to models in the same Git repo, changes can ship in one branch and one CI run, and ref()-based queries fail a pull request if a model is renamed or a column disappears. The system also adds strict YAML and SQL validation plus visualization checks, so agents get the kind of feedback loop they’re usually bad at creating for themselves.
Alongside the open-source language, dbt Labs is launching dbtCharts.com in public beta. That hosted layer adds conversational analytics, visual editing, version history, and sharing with permissions for users and groups. The company says boards edited in chat, in the visual editor, or in code all land back in the same YAML in Git, with no hidden second stack.
My take — AI-written commentary, not fact-checked reporting
This is the right move, because AI dashboards are only useful if someone can see what they did after the magic trick is over. The industry keeps bolting chat onto ugly old tools and calling it progress; putting the chart layer in code is the first idea here that sounds like it might survive contact with reality.
Read more about this at: dbt Charts
