Introducing headless BI and the Fabi CLI

Dashboards aren't going away. The part where humans build them is.

Over the past year, something has shifted in how the best engineering and data teams work. They've stopped opening Tableau or Looker to drag and drop their way to a dashboard. Instead, they're opening Claude Code or Codex, describing what they want, and getting something back that looks less like a BI report and more like a custom-built app.

This isn't a niche workflow anymore. It's becoming the default. And it's why we're introducing headless BI and releasing the Fabi CLI.

What's actually changing

The traditional BI tool was a bundle: data connections, a query layer, a visualization layer, and a sharing mechanism, all wrapped up in one product. That made sense when building a dashboard required a specialized tool. It makes less sense when your AI agent can generate the visualization layer from a single prompt.

The visualization layer is moving into the agent. Teams are using Claude Code and Codex to generate fully custom dashboards that meet their exact needs, built in minutes rather than days.

For a lot of use cases, this works really well. Small, relatively clean datasets with minimal governance requirements? Your agent handles it fine.

The problems show up fast when you move beyond that.

Where agents break down

Anyone who has tried to build production-ready dashboards through an AI agent has run into the same set of issues.

Agents don't understand your data. They're starting from scratch every time. Without context about what your tables actually mean, your business definitions, or how your data sources relate to each other, the agent is guessing. You've probably seen it SELECT * across every table in your database just to get oriented. That's not a bug, it's a fundamental limitation of working without context.

Access control is an afterthought. Generating a dashboard in an agent is one thing. Deploying it in a way that respects who should and shouldn't see which data is another problem entirely. Most teams end up with a bespoke, fragile solution or skip it altogether.

Deploying and sharing is a massive hurdle. After all that, you still need to figure out how to get the dashboard in front of the people who need it, in a way that's maintainable.

All of these are blockers that prevent agent-built dashboards from being used seriously at scale.

What we built

Fabi is the data and context layer that sits underneath your agent. Today we're releasing the Fabi CLI so you can plug headless BI directly into Claude Code or Codex, right where you already work.

Fabi handles three things your agent can't do on its own.

Data source connections. Connect your databases, warehouses, and data sources once. Fabi manages the connections so your agent can query without setup overhead every time.

AI context management. Fabi crawls your data source metadata and ingests external context, so your agent actually understands your data before it starts working. Consistent definitions, meaningful schema context, no more SELECT * fishing expeditions.

Security and role-based access control. Define who can see what, once. Every dashboard built through Fabi respects those rules automatically, whether it was built by a human or an agent.

Getting started in under 5 minutes

Open Claude Code or Codex and tell it:

Install fabi <https://github.com/fabi-ai/fabi-cli>

Follow the prompts to connect your data sources, then start generating dashboards. When you're ready to share, ask the agent to deploy. Fabi handles access control automatically based on your configuration.

The whole setup takes under 5 minutes. The dashboards you get out look like something a frontend team spent days building.

Headless BI is the future

The visualization layer is moving into the agent. That's not a prediction, it's already happening. What we're building is the foundation that makes it work at scale: the data connections, the context, the governance.

Headless BI is finally here.

Try the Fabi CLI