Tool-belt series · Microsoft Fabric · Application layer
Fabric Apps is a first-class workspace item for building real web applications on top of your Fabric data — with Entra sign-in, managed hosting and a backend provisioned for you. It closes the two gaps the BI stack never covered: writing data back, and interfaces that aren't a report. This is the short orientation to read before you get access.
Fabric Apps went to public preview on 2 June 2026, announced at Microsoft Build. Preview means the shape is real but the details will move: expect API changes, gaps in ALM, and licensing that isn't fully settled. Don't put a month-end process on it yet.
What preview is genuinely good for: learning the model now, so that when it reaches your tenant you're the person who already knows what it replaces and what it doesn't.
Not a report. Not a Power App. A hosted web application that is a Fabric item — living in a workspace, governed by workspace permissions, versioned like everything else.
You build a browser-based application. Fabric hosts it as a managed service, provisions its backend, wires up Entra ID authentication so users sign in with their work account, and gives it governed access to your data. No web server to run, no embed tokens to mint, no separate app registration to babysit, no App Service to pay for and patch. The app is an item in a workspace — so the same roles, the same Git integration, and the same governance that cover your lakehouses cover your app.
That last point is the one that matters strategically. Every previous "let's build a small app on our data" story meant leaving the data platform and taking on a second stack. This keeps it in one place, under one permission model.
Fabric Apps ships as templates. The two that matter are operational apps and data apps. Confusing them is the easiest way to pick the wrong tool.
The default template. A transactional, write-enabled application with its own managed SQL database provisioned as a child item, exposed through a GraphQL API.
A code-first, visualisation-focused application that queries published semantic models using DAX and renders whatever interface you want.
Worth carrying this picture in your head — it's what makes the licensing and governance questions answerable.
Note: drawn from Microsoft's Build 2026 announcement and public preview write-ups — an orientation diagram, not an official architecture reference.
Alongside Apps, Microsoft published Rayfin, an open-source SDK and CLI for defining and deploying app backends. It's the code-first path: developers — and AI coding agents — declare the backend, and Fabric provisions it with enterprise security and governance attached from the start.
The interesting implication is that Apps is deliberately designed to be built by agents, not only by hand. If you already drive Copilot or Claude over a codebase, that skill transfers directly here.
The value of knowing this early is being able to say "no, that's a report" or "no, that's a Power App" with confidence, instead of building the wrong thing enthusiastically.
| If you need… | Use | Because |
|---|---|---|
| A dashboard people read | Power BI report | Faster to build, easier to govern, and everyone already knows how to consume it. Don't build an app to redraw a bar chart. |
| An interactive experience a report can't express | Data app | Custom interaction and layout, over multiple semantic models, reusing your existing measures and RLS. |
| Users to enter or correct data | Operational app | Real write-back with its own SQL backend that mirrors into OneLake — no Dataverse, no second platform. |
| A simple form for a handful of people | Power Apps (for now) | Mature, low-code, well understood. Fabric Apps is preview and is code-first — use it where its strengths matter. |
| Analytics embedded in an existing product | Power BI Embedded | Fabric Apps assumes users sign in with Entra to a Fabric-hosted app; it isn't a drop-in embed story for external customers. |
| A question answered in English | Data agent | Cheaper and faster than any UI you'd build for the same purpose. |
All of this is useful whether Apps lands next month or next quarter — and it's what makes your first app take a day instead of a fortnight.
Every finance and operations team has one: a spreadsheet of adjustments, a forecast entry file, a mapping table someone maintains by hand and emails round. That's your first operational app, and you can identify it today.
A data app queries your model with DAX. If the measures are inconsistent or the model is a flat table, the app inherits every one of those problems. Gold-layer modelling is the prerequisite, not an afterthought.
It's already generally available independent of Apps, and it's the same access path an operational app uses. Building one API over a lakehouse this month means the app layer is the only new thing to learn later.
An app is software: it needs an owner, a support route and a lifecycle. The teams that struggle with Apps will be the ones who treated it like a report. Raise this before the first one ships, not after.
Preview features change fast. Fifteen minutes a month on the Fabric blog keeps you the best-informed person in the room on this — which is most of what being the expert consists of.