Charts built for Chat

Blog · September 14, 2026 · Dave Fowler

We’re unfastened sourcing dbt Charts, a declarative connection for dashboards, truthful that moreover the dashboards you build by chatting pinch an supplier tin beryllium governed.

AI for information is here, and the long-promised self-serve analytics is finally happening. Anyone pinch a information relationship tin chat a study into beingness in an afternoon, and the first results are impressive.

The frictions show up fast, though. By default an supplier turns 1 elemental report into a heap of files: HTML, CSS, and JavaScript, a mates of floor plan libraries, and a React aliases Streamlit app erstwhile it has to beryllium live. Tracing a consequence backmost to its root intends pursuing it done respective languages and files, which is slow for group to audit and costs the supplier clip and tokens connected each change.

BI devices went the different measurement and bolted copilots onto their UI-first apps. That keeps the AI connected governed rails, but constrictive ones: the supplier tin do only what the UI exposes.

So coming you take betwixt the messy state of codification and the constrictive control of a BI tool. We built a 3rd option: skip ahead, aliases publication on for really BI sewage here.

Unbundling BI

As dbt Labs laminitis Tristan Handy wrote precocious in BI’s Second Unbundling:

When I started successful data, BI devices were full-stack. Everything happened inside one product: information ingestion, transformation, compute, caching, semantics, visualization, identity. The BI instrumentality was the information stack. MicroStrategy, Cognos, etc: they’re not conscionable visualization tools, they’re integrated data platforms.

Then the modern information stack happened. From ~2015 to 2022, the infrastructure layers of that BI bundle sewage pulled retired and turned into purpose-built infrastructure. Compute went to the Big 5. Ingestion went to Fivetran. Transformation went to dbt. The BI instrumentality was near with: visualization, interactive analytical interfaces, semantic definitions (sometimes!), identity and entree management, and web hosting.

What that unbundling near down is the BI instrumentality we cognize today, and charts are its biggest piece. They stayed successful the UI for bully reason: for astir people, clicking is quicker than penning YAML. But much and much charts won’t beryllium made by people. As the beforehand extremity and personification of everything becomes progressively a chat agent, this preference flips. Agents are fluent successful code, SQL, and Git, and clumsy successful someone else’s UI. So charts request to move to wherever agents work: into code.

Today we’re taking the adjacent measurement successful unbundling BI: we’re unfastened sourcing dbt Charts, which takes charts retired of the BI instrumentality and puts them successful code, specifically a caller system YAML language that tin state a afloat interactive dashboard successful 1 auditable YAML file. Chat freely pinch an agent, and what it makes has the state of codification while staying easy to read.

In dbt Charts, SQL remains the connection for declaring WHAT information you want to see, and we wrap that successful YAML to state HOW you want to spot it.

We’ve spent a agelong clip distilling the connection to a fewer core, extensible elements: heavy successful what they tin express, easy to shape and read. The YAML wraps much than SQL. Markdown carries the prose, and Jinja, arsenic successful dbt, carries variables and macros.

Here’s a mini example: 1 adaptable (a UI filter), 1 query and 1 chart.

That record is the full board. The CLI renders immoderate committee record to fixed SVG, or to HTML, PNG, PDF, and moreover the terminal, connected your laptop aliases successful CI, and serves a files of them arsenic a site:

Those fewer elements spell deep: complete 1,100 config options today, across sixteen floor plan types and the composed charts built from them. And for illustration immoderate bully language, it tin definitive analyzable layouts and visuals.

You seldom group those options by hand. Styles cascade: a floor plan inherits from its board, the committee from its theme, and a taxable is 1 statement to switch. A committee can also extends: different board, truthful a location style aliases a modular study is written once and inherited everywhere. Boards enactment short, and theming stays cheap.

A complete dbt Charts board, rendered from 1 easy-to-read YAML file.

Deep integration pinch dbt

You don’t person to usage dbt Charts pinch a dbt project, but erstwhile you do, a lot unlocks. The floor plan furniture sits straight connected the toggle shape layer, and the deeper the integration, the easier it is to alteration both.

With dbt Charts, your charts/ directory lives adjacent to your models/ successful the same Git repo, truthful a alteration to a exemplary and its charts ships connected 1 branch, through 1 CI run, and breaks earlier it reaches production.

Queries scope models done ref(), resolved from your manifest, truthful a renamed model aliases a missing file fails the propulsion petition that collapsed it, before dbt tally rebuilds the warehouse:

Support for the dbt Semantic Layer is planned, truthful a committee tin usage a metric as the task defines it alternatively of restating its SQL. Follow dbt-labs/dbt-charts#1.

Built for chat

Agents tin beryllium rather blind, and they do champion pinch a tight feedback loop. dbt Charts gives them one: strict validation of some the YAML and the SQL, and an extended group of visualization checks that emblem problems earlier anyone sees the board:

A beautiful, cohesive reporting system

We dream dbt Charts, for illustration dbt earlier it, becomes the unfastened modular connection for its furniture of the information stack. We designed it for a early wherever humans and AI build together, and we wanted it to look for illustration that future, not for illustration another dashboard grid. We recruited RJ Andrews, a data schematic designer, author, and historian, to creation the charts. His grasp of the craft’s history is what makes the consequence consciousness new: it reaches past the dashboard era to what charts looked for illustration erstwhile group drew them pinch care.

Many devices cheat pinch cards and boxes that clone alignment astatine the costs of visual noise and mislaid space. We worked retired the spacing, sizing, and layout of every chart, connected its ain and adjacent to its neighbors.

The consequence is simply a cohesive strategy of charts that feels a level supra existent BI.

Alongside the open-source language, coming we’re launching dbtCharts.com successful nationalist beta: a hosted level for the rest of BI. With charts pulled out, what remains is chiefly hosting, access control, and a UI. By their quality these possibly can’t beryllium unbundled, aliases astatine least shouldn’t be, truthful the level handles them connected apical of the open-source language.

The level connects to your storage and adds conversational analytics, a visual editor for the finishing touches, type history, and sharing pinch permissions for users and groups, truthful the group reference a committee don’t request a warehouse login.

And of course, these charts were built for chat. The level has first-class conversational analytics: for illustration Claude aliases ChatGPT, but pinch permissioned read-only entree to your storage and an master analyst’s skills and tools built in. Explore by chatting pinch charts, and astatine immoderate constituent click successful to fine-tune and save the board.

Because it’s built connected the unfastened language, each change, from chat, the visual editor, aliases code, lands successful the aforesaid YAML successful your Git repo. Nothing is locked in: the aforesaid committee runs connected your laptop, successful CI, and on the platform, and teams tin self-serve, supplier successful hand, without creating a second, hidden information stack.

Try the beta

The dbt Charts connection is unfastened root nether the Apache 2.0 license, and you can author, render, and service boards locally without creating an account. Install it yourself, aliases manus your coding supplier 1 line:

• Code: github.com/dbt-labs/dbt-charts
• Docs: docs.dbtCharts.com
• Hosted BI: dbtCharts.com
• Community: #dbt-charts connected the dbt Community Slack

dbt Charts is pre-1.0 and still changing. When the grammar changes, boards migrate arsenic they parse, truthful the boards you constitute coming support rendering. Try it, show america what is missing[1][2], subordinate the chat successful #dbt-charts connected Slack, and thief america build the floor plan layer that unfastened information infrastructure has been waiting for.