Co-authored by

What People Mean by "Microsoft Fabric Apps"

Quick answer: "Fabric apps" currently refers to at least four distinct capabilities: the new Fabric Apps item type (public preview, June 2026), translytical task flows for write-back inside Power BI reports (GA April 2026), Power Apps on OneLake via Dataverse virtual tables (GA November 2024), and the Extensibility Toolkit for ISV-built workloads (GA March 2026). This post separates all four and adds Rayfin, the open-source SDK that ships with the new item type.

The confusion is understandable. Microsoft has released multiple app-layer capabilities in Fabric over the past 18 months, each with a different scope, audience, and maturity level. Most articles about "building apps with Fabric" mean one of these four things — often without specifying which.

Worth stating clearly at the start: the Fabric Apps item type announced at Build 2026 is genuinely new. The other three have been in production for months or years. The sections below cover each in sequence, then offer a practical decision table.

Fabric Apps: The New Item Type

Quick answer: Fabric Apps is a managed backend-as-a-service environment that Microsoft released to public preview on June 2, 2026 at Microsoft Build. It provisions SQL storage, a GraphQL API, Entra ID authentication, and static hosting as a single first-class Fabric item, and connects to Power BI semantic models through DAX binding — without embed tokens.

Before Build 2026, building a custom web application on top of a Power BI semantic model meant either Power BI Embedded (with its token infrastructure and separate licensing overhead) or a standalone app calling Power BI REST APIs directly, managing its own token refresh. Neither is fast to scaffold. Fabric Apps changes that setup.

When you create a Fabric Apps item in a workspace, you get a managed execution environment: a SQL database in Fabric, Entra ID SSO authentication, a GraphQL API auto-generated from your data models via Microsoft's Data API Builder, and static hosting for frontend assets. The entire app appears as a single workspace item with child resources nested underneath it.

The data app template is the most useful starting point for analytics teams. It pre-wires connectivity to a Power BI semantic model, handling the authentication handshake that would otherwise require custom token management. Charts in a data app use DAX binding: each visual has a co-located .dax file containing the query with runtime placeholders (such as {{YEAR}} or {{FILTERS}}), a .json Vega-Lite spec defining the visual grammar, and a .tsx React component wiring them together. At runtime, the app calls the Semantic Model Execute Queries REST API — meaning row-level security defined on the semantic model is enforced at query time, exactly as in a native Power BI report. The tenant admin setting "Semantic Model Execute Queries REST API" must be enabled.

Authentication is Fabric SSO through Entra ID, and only that. There is no app-owns-data embed path, no guest user support, no anonymous access. For internal teams sharing a Microsoft 365 tenant, Fabric Apps eliminates the need for separate embedding infrastructure. For external-facing analytics — partner portals, customer dashboards — Power BI Embedded remains the supported option.

One open question as of June 2026: whether viewers need Power BI Pro licenses on smaller F SKUs is not yet documented by Microsoft. That answer will determine how much of a cost shift Fabric Apps actually represents for internal deployments compared to existing embedding patterns. Watch the Microsoft Fabric licensing documentation for updates before making architecture decisions based on assumed cost savings.

Rayfin: The Code-First Backend SDK

Quick answer: Rayfin is an MIT-licensed open-source CLI and SDK for building and deploying Fabric Apps. Running npx rayfin up provisions a SQL database in Fabric, a GraphQL API via Data API Builder, Entra ID authentication, and static hosting — all from TypeScript type definitions, with no manual infrastructure configuration.

Rayfin's model is straightforward: your TypeScript types define your data model, and the CLI handles the rest. You define entities as TypeScript classes with decorators, run npx rayfin up db apply to push the schema and generate GraphQL and REST endpoints through Data API Builder, then npx rayfin up staticapp deploy to build and ship the frontend. The CLI reference describes rayfin up db apply as generating and applying "DAB configuration to the remote Rayfin item workload endpoint."

The OneLake routing matters for teams already in Fabric: application data written through Rayfin lands in OneLake by design, immediately accessible to the rest of the Fabric data stack — Power BI reports, notebooks, data pipelines, real-time analytics — without a separate ETL step. If your team already uses Microsoft Fabric for real-time analytics or lakehouses, the application data that Rayfin writes becomes part of that same governed data environment automatically.

At launch, Replit partnered with Microsoft as the primary development environment. The combination targets AI-assisted development workflows: Replit's coding agent scaffolds the frontend, Rayfin handles the Fabric deployment, and the result is an application with enterprise authentication and OneLake governance already in place.

The GitHub repository at github.com/microsoft/rayfin is MIT-licensed and supports experimental local development without requiring cloud resources. For practitioners who want to evaluate before committing Fabric capacity, the local mode is the sensible starting point.

Install: npm create @microsoft/rayfin@latest

For teams comfortable with TypeScript and existing Fabric workloads, Rayfin reduces what was previously a multi-week custom integration effort — API layer, token management, data access layer, hosting — to a CLI workflow that can produce a working deployment in a day.

The Apps Already on Fabric

Quick answer: Three app capabilities predated Build 2026 and are already generally available: translytical task flows for write-back in Power BI reports (GA April 2026), User Data Functions for serverless Python (GA September 2025), and Power Apps on OneLake via Dataverse virtual tables (GA November 2024). The Extensibility Toolkit for ISV workloads reached GA in March 2026.

Translytical Task Flows

Translytical task flows let report users insert, update, or delete records in Fabric SQL databases, lakehouses, and warehouses — or call external APIs — without leaving the Power BI report. The "translytical" name signals the combination of transactional and analytical: the report both reads and writes.

The feature entered preview on May 19, 2025 and reached GA on April 20, 2026. Each task flow calls a User Data Function (serverless Python) that handles write-back logic, validation, and any downstream calls — Teams messages, Azure OpenAI, third-party systems. Filter context from the visual passes to the function as parameters, so the action is scoped to what the user is currently looking at.

This covers the write-back scenario that previously required a separate Power App, a custom API, or a direct SQL connection from the report. It does not require Rayfin or the Fabric Apps item type.

User Data Functions

Serverless Python hosting inside Fabric. Functions are callable from Pipelines, Notebooks, Activator alert rules, translytical task flows, and external systems via REST endpoints. They support Python 3.11, PyPI packages, pandas DataFrames, and async functions.

User Data Functions reached GA on September 24, 2025. They became available to all tenants without an admin toggle at that point. They are the execution engine underneath translytical task flows and can be used independently for any serverless computation workload that needs to run inside Fabric's governance boundary.

Power Apps on OneLake via Dataverse Virtual Tables

GA since November 28, 2024. When you link a Power Apps environment to Microsoft Fabric, Dataverse tables sync to OneLake through shortcuts — no data copy, no ETL pipeline. You can then join that Dataverse data with other OneLake content (lakehouse tables, external sources), and surface the combined result back into Power Apps as virtual tables.

One important constraint: virtual tables sourced from Fabric via OneLake are read-only in Power Apps. Users can view and act on Fabric data inside a Power App, but they cannot write back through this path. For write-back, translytical task flows via User Data Functions is the supported pattern.

Extensibility Toolkit

For software vendors, not internal teams. The Extensibility Toolkit lets ISVs build workload items that appear native inside a Fabric workspace, with their own OneLake folder per item, following the same CRUD patterns as built-in Fabric items. Introduced September 22, 2025; GA March 19, 2026. ISVs distribute through the Fabric Workload Hub. If your goal is building an internal analytics app, this is not the right tool.

Which One Should You Reach For?

Quick answer: If you need write-back inside an existing Power BI report, use translytical task flows. If you need a standalone web application against a semantic model, Fabric Apps and Rayfin are the right direction — but treat them as preview-stage. For Power Platform teams already in Dataverse, the OneLake virtual table pattern has been production-ready since late 2024.

On preview maturity: Fabric Apps and Rayfin entered public preview at Build 2026. The data app template produces a working application quickly, and the DAX binding approach is well-defined and consistent with how Power BI reports query semantic models. But the feature set will change between now and GA — particularly around licensing, multi-tenant support, and connector coverage.

For a production deployment serving several hundred users in six months, translytical task flows on an existing Power BI report is a more stable foundation today. For a proof of concept or an internal tool with a small user group, Fabric Apps is worth evaluating now.

One GCC-specific note: Fabric Apps is deployable to Azure UAE North (Dubai) workspaces, which supports all Fabric workloads. If your organization already has Fabric capacity provisioned in UAE North, a Fabric App deployed to that workspace keeps data in-region. Verify current regional availability on Microsoft Learn before relying on this for regulated-sector deployments, since preview features sometimes carry additional regional restrictions.

What This Means for Your Power BI Investment

Quick answer: The semantic model becomes a shared layer across multiple surfaces — Power BI reports, Fabric Apps, translytical task flows, AI agents querying DAX — without duplicating data or business logic. Teams that built a well-governed semantic model are now able to expose it through more surfaces than before.

The consistent thread running through Fabric Apps, translytical task flows, and the broader direction Microsoft laid out at Build 2026 is that the semantic model is not just a data source for Power BI reports. It is a governed interface between raw data in OneLake and any surface that needs to consume or act on that data.

Investments in semantic model design — correct relationships, DAX measures, certified datasets, row-level security — now pay off more broadly. A sales team using a Power BI report and a custom Fabric App built on Rayfin can query the same semantic model, enforcing the same business logic and security, from different interfaces. An AI agent configured to query the model via the Execute Queries API gets the same governed result.

This also raises the stakes on semantic model quality. If four surfaces are pulling from the same model, a poorly written measure or a broken RLS rule propagates everywhere. The governance work that should have happened before is now more consequential.

For teams currently working through a Fabric rollout, the Microsoft Fabric enterprise adoption roadmap covers the sequencing decisions that affect where app-layer capabilities fit. The snapshot tables pattern for materializing point-in-time semantic model data into OneLake becomes relevant as more surfaces share the same model and need defensible historical records at a fixed point in time.

Beyond The Analytics works with UAE and Saudi enterprises on Fabric semantic model design, Power BI development, and capacity architecture. If your team is evaluating where Fabric Apps or translytical task flows fit in an existing Fabric investment, get in touch.

Frequently Asked Questions

Q: Is Fabric Apps the same as Power BI Embedded?

No. Power BI Embedded uses embed tokens and supports external users, anonymous access, and multi-tenant app-owns-data patterns. Fabric Apps uses Fabric SSO through Entra ID — there is no embed token path, no anonymous access, and no support for users outside your Microsoft 365 tenant. For internal teams on a shared tenant, Fabric Apps may eliminate the embedding infrastructure overhead. For external-facing analytics serving customers or partners outside your tenant, Power BI Embedded remains the supported option.

Q: When were Fabric Apps and Rayfin released?

Both entered public preview on June 2, 2026, announced at Microsoft Build 2026. At the time of writing, neither is generally available. The underlying components — Fabric's SQL database, GraphQL via Data API Builder, OneLake storage — are GA; Fabric Apps and Rayfin as a packaged developer experience are preview.

Q: Do I need to code to build a Fabric app?

For the Fabric Apps item type: yes, in practice. Rayfin targets TypeScript developers, and the data app template produces React components, Vega-Lite JSON specs, and DAX query files. Microsoft has positioned GitHub Copilot as the scaffolding tool, so the workflow is AI-assisted coding rather than visual no-code — but it is still code. For write-back inside an existing Power BI report, translytical task flows require a Python function via User Data Functions. For reading Fabric data inside a Power App without writing code, the Dataverse virtual table pattern (GA since November 2024) is the no-code option.

Q: How do Fabric Apps use my Power BI semantic model?

Through the Semantic Model Execute Queries REST API. The data app template scaffolds DAX queries as .dax files co-located with each chart component. At runtime, the app substitutes filter context values into the parameterized DAX query and calls the REST API, which executes the query against the live semantic model and returns the result. Row-level security on the semantic model is enforced at query time. The tenant admin setting "Semantic Model Execute Queries REST API" must be enabled. No data is duplicated — the app queries the semantic model directly, the same way a Power BI report does.