TL;DR
Microsoft Rayfin is an open source tool that builds an app’s backend and deploys it straight onto Microsoft Fabric. One command sets up the database, login system, and API together. The app’s data lands in OneLake right away, with no separate step to move it into analytics. Rayfin is still in preview, so Microsoft has not set a launch date yet. It only makes sense for teams already using Fabric as their data platform.
Microsoft introduced Rayfin at Build 2026 to close a specific gap in AI-assisted development. Coding agents had become capable of building frontends quickly, but the backend, the database, the authentication layer, the access policies, still required manual assembly and a separate review against governance standards. Rayfin is an open-source SDK and CLI that lets a developer, or a coding agent operating under their direction, define a complete application backend in code and deploy it directly onto Microsoft Fabric.
The structural change is what happens after deployment. A Rayfin application writes directly to OneLake, so its data is available to Power BI and Fabric’s AI agents without a separate integration pipeline. Rayfin remains in public preview rather than general availability, and that status affects every major decision this guide addresses, including cost, its comparison against Supabase and Firebase, and which types of applications are ready to run on it today.
Key Takeaways
- Microsoft Rayfin is an open-source SDK and CLI, announced at Microsoft Build 2026, for deploying application backends directly onto Microsoft Fabric.
- One command provisions a TypeScript-defined database, authentication, GraphQL API, and hosting, and app writes land in OneLake immediately.
- Rayfin entered public preview on June 2, 2026, and Microsoft has not announced a general availability date.
- Rayfin has no separate pricing tier. It consumes Capacity Units from whatever Fabric capacity the workspace already has.
- Rayfin apps inherit Fabric’s existing role-based access control, sensitivity labels, and Purview policies automatically.
- Kanerika’s Microsoft Fabric practice is evaluating Rayfin for clients building new applications on an existing, governed Fabric estate.
Watch on YouTube
Kanerika on Microsoft Fabric: AI Insights with Karl + Faster Azure Migration
Kanerika’s Karl agent and the Azure to Fabric accelerator now run as native Microsoft Fabric workloads. A short look at what shipping an application inside Fabric looks like in practice.
What Is Microsoft Rayfin?
Microsoft Rayfin is an open-source SDK and command-line interface for building an application backend and deploying it directly to Microsoft Fabric. A developer defines the data model, business logic, and access policies in TypeScript, then runs the Rayfin CLI to provision that backend on Fabric in one step.
The core capabilities don’t vary by project. A SQL database, authentication, a generated API layer, storage, and hosting all provision together as one unit, a deliberate departure from the assemble-it-yourself model most backend platforms still expect. Microsoft’s own GitHub repository puts it plainly: backend-as-a-service with built-in auth and data.
Rayfin is still in public preview, announced at Microsoft Build on June 2, 2026. That status is worth holding onto through the rest of this guide.
One naming point first. Rayfin is the open-source SDK and CLI a developer runs locally to define and deploy an app. Fabric Apps is the managed environment inside a Fabric tenant where that app runs, and it is also in public preview. Neither one replaces Power BI or Power Apps, and how Rayfin, Fabric Apps, Power BI and Power Apps compare comes down to who builds the app and where new data lands.
Kanerika Service
Building on Microsoft Fabric in 2026?
Kanerika is a Microsoft Fabric Featured Partner that covers implementation, migration, and governance across the full Fabric stack.
See Fabric Services →
How Microsoft Rayfin Works
A Rayfin project moves through a fixed sequence, from code to a running backend on Fabric.
1. Defining the App in Code
A Rayfin project starts with a data model written in TypeScript, using decorators to define entities, relationships, and access rules. Business logic and API behavior live in the same codebase, version-controlled through the GitHub-based workflow Microsoft built the launch around.
2. Deploying With the Rayfin CLI
Scaffolding a new project runs through npm create @microsoft/rayfin@latest, and deployment runs through npx rayfin up, the same two commands Microsoft’s CLI reference walks through end to end. That single command also handles every later schema or frontend update, so shipping a change doesn’t require a separate release process from building it.
3. What Microsoft Fabric Provisions
A deployed Rayfin app gets a Fabric SQL database, Microsoft Entra ID authentication, a generated GraphQL API, and static content hosting, all inside the customer’s own Fabric workspace. These generate automatically from the TypeScript data model, not configured as separate services one at a time.
4. Connecting Applications to OneLake
Every write to the Fabric SQL database lands in OneLake without a separate pipeline. Power BI semantic models, Fabric notebooks, and Fabric data agents read that same data immediately, which is the actual structural difference from a standalone backend sitting outside the data estate.
That direct line into OneLake is also what makes the next question practical. What should a team build on it first?
Microsoft Rayfin Use Cases
Microsoft’s documentation names four scenarios for Fabric Apps, the environment Rayfin deploys to. They are rapid prototyping, internal tools and dashboards, custom front ends for exploring Fabric data, and backends for AI agents that need persistent state. The five use cases below are where that guidance turns into something a team would ship.
1. Write-Back Apps That Go Beyond the Dashboard
A Power BI report shows the number but rarely lets a manager change it. A Rayfin app can sit beside that report and capture an approval, a forecast override, or a corrected record. The write goes to a Fabric SQL database and lands in OneLake immediately.
Microsoft’s Rayfin AMA framed this as apps that capture decisions, trigger actions, and write results back into the business process. When the write-back is a single button inside a report, translytical task flows are the lighter option. Rayfin is the better fit when the workflow needs its own screens, roles, and history.
2. Governed Replacements for Shadow IT Apps
Shadow IT shows up as spreadsheets, Access databases, and agent-built side apps that IT has never reviewed. HFS Research’s Ashish Chaturvedi described Rayfin to InfoWorld as a governed on-ramp for that category. Rebuilding one of those apps on Rayfin gives it Entra sign-in, Fabric workspace roles, sensitivity labels, and Purview coverage without a separate security project.
IDC’s Stewart Bond told the same publication that governance and operational control make the stronger case for CIOs. The risk in a shadow app usually sits in its ungoverned data copy. Rayfin removes that copy by keeping the data inside Fabric, where OneLake security already applies.
3. Persistent State for AI Agents
An agent that triages alerts, drafts purchase orders, or files tickets needs somewhere to keep its task queue, its approvals, and its audit trail. Microsoft’s documentation positions Rayfin as the structured backend for agents that need that persistent state. The state lands in the same governed estate the agent reads from.
In practice this means a Fabric data agent and a Rayfin backend can share one security model and one copy of the data. Functions, the planned way for an app to trigger notebooks and pipelines, were still on the roadmap in Microsoft’s June 2026 AMA. Agent workflows that must start Fabric jobs should confirm current status first.
Datasheet
Karl, the Data Insights AI Agent for Microsoft Fabric
Kanerika’s Karl runs as a native Microsoft Fabric workload and answers business questions over governed OneLake data. The datasheet shows what it does and how it fits inside a Fabric tenant.
View the Datasheet →
4. Custom Front Ends and Admin Tooling on Fabric Data
Field teams, data stewards, and tenant admins often need a search box, a form, and a list rather than a report. Rayfin’s connectors give an app typed access to a lakehouse, warehouse, SQL database, or semantic model. The hosted front end can sit behind Fabric item permissions, and users sign in with the Entra identity they already have.
A Microsoft solution engineer’s published walkthrough builds a Fabric estate optimizer that inventories reports, scores usage, and flags reports nobody has opened in three months. Capacity and cost tooling of that kind is a sensible first app. The data is already in Fabric and the users are already in Entra.
5. Where Rayfin Is the Wrong Tool Today
Microsoft lists two cases Fabric Apps does not suit. The first is any application that needs complex multi-step transactions or stored procedures. The second is any application that needs an authentication provider other than Fabric SSO or email and password.
Deployed apps authenticate through Fabric SSO only, which rules out customer-facing apps for users without an Entra identity for now. Region coverage is also still partial. The June 2026 AMA recommended provisioning capacity in West US 2 or North Central US while support expands.
A team whose capacity sits elsewhere should budget for a second capacity or wait for its region. Either way, that decision belongs in the architecture review before any code is written.
Microsoft Rayfin Use Cases at a Glance
| Use case | What a team builds first | Why Rayfin fits | Watch for |
|---|
| Write-back app | Approval or forecast-override screen beside a Power BI report | Writes land in OneLake with no pipeline | Translytical task flows may be enough for a single button |
| Shadow IT replacement | Rebuild of a spreadsheet or Access workflow | Entra sign-in, workspace roles, labels, and Purview by default | Migrating data out of the old copy |
| AI agent backend | Task queue, approvals, and audit trail for an agent | Agent state and source data share one security model | Functions for triggering Fabric jobs still on the roadmap |
| Custom front end | Admin or steward tool over lakehouse or semantic model data | Typed connectors plus a hosted UI behind item permissions | Fabric SSO only, so internal users for now |
Governance and Data Management in Rayfin
A Rayfin app doesn’t get its own separate security model. It inherits whatever governance already exists on the Fabric tenant it deploys into.
1. Role-Based Access Control
Every child service Rayfin provisions, the database, the API, the hosted frontend, is a distinct artifact inside the Fabric workspace, subject to that workspace’s existing RBAC rules. That’s architectural, not something configured after the fact. Access follows the workspace, not the app, which one detailed technical breakdown confirms holds even at the child-service level.
2. Sensitivity Labels
Data written through a Rayfin app carries whatever sensitivity labeling the organization already applies across its Fabric estate. A Rayfin-built application handling regulated data doesn’t need a separate classification pass. It inherits the same labels Power BI reports and Lakehouses already use.
3. Microsoft Purview
Purview policies apply to Rayfin-provisioned data the same way they apply to any other Fabric-native asset, since the data never leaves the tenant. That inheritance is only as strong as the governance already in place before the app gets built, which is the part most “governed by default” pitches leave out.
Watch on YouTube
Why Most Fabric Deployments Fail at Scale
Amit Chandak, Microsoft MVP and Kanerika’s CAO, on why Fabric deployments break at scale and the governance decisions that prevent it. Worth watching before any app inherits your tenant’s policies.
What Does Microsoft Rayfin Cost?
Rayfin has no pricing tier of its own, which is easy to misread as free. A deployed app draws Capacity Units from whatever Fabric capacity is already assigned to the workspace, the same pool that runs Warehouses, Lakehouses, and Notebooks, and Microsoft’s pricing documentation is explicit there’s no separate billing line.
The conversion rate is fixed. One Fabric CU equals 0.383 SQL database vCores, and every GraphQL query or mutation the app makes bills at 10 CUs per hour of request and response processing time. Microsoft recommends F4 capacity for development and F32 or higher for production, a threshold independent cost analysis treats as a hard floor, not a suggestion.
As an illustrative example, an internal app running a modest 50 hours of combined GraphQL processing time in a month would draw roughly 500 CUs, before SQL storage and compute get counted separately. That’s additive to whatever the organization already spends on Fabric, not a new bill.
One licensing detail is still genuinely unsettled. Below F64 capacity, users viewing a deployed app typically still need Power BI Pro licenses, and at F64 and above that requirement may not apply. One technical analysis calls this exact behavior undocumented and worth testing against a live tenant, and that’s the right instinct until Microsoft closes the gap in writing.
Microsoft Rayfin vs the Alternatives
1. Rayfin vs a Traditional Custom Backend
A traditional backend means provisioning a database, building authentication, developing an API layer, and standing up hosting as separate, individually maintained pieces. Rayfin collapses that into one deployment, at the cost of being locked to Fabric as the runtime.
2. Rayfin vs Supabase and Firebase
Supabase runs on PostgreSQL, with Row Level Security enforced natively at the database layer, and it can be self-hosted. Firebase runs on Firestore, a schemaless NoSQL document store, and only added a SQL option, Firebase Data Connect, in 2024, years after Supabase built its entire pitch around SQL from day one. A detailed 2026 platform comparison still treats that gap as the defining split between the two.
Neither platform has a native path into a broader governed analytics estate. A Supabase or Firebase app’s data stays inside that platform unless a team builds its own pipeline out. That’s the actual gap Rayfin closes, not raw feature parity, but where the data ends up living.
Rayfin vs Custom Backend vs Supabase vs Firebase
| Area | Custom Backend | Supabase | Firebase | Rayfin on Fabric |
|---|
| Database | Self-managed | PostgreSQL | Firestore (NoSQL) | Fabric SQL database |
| Authentication | Built by hand | Supabase Auth | Firebase Auth | Microsoft Entra ID |
| API layer | Developed manually | Auto-generated REST/GraphQL | Client SDKs | Generated GraphQL |
| Runs on | Own infrastructure | Supabase’s platform (or self-hosted) | Google Cloud | The org’s own Fabric tenant |
| Native analytics access | None | Not built in | Not built in | Direct, via OneLake |
| Governance | Configured separately | Row Level Security (Postgres) | Firebase security rules | Inherited from the Fabric tenant |
Every alternative here manages authentication and a database competently. Only one of them puts that data inside a platform the organization already governs for analytics, and that’s the entire argument for choosing it.
Case Study
90% Data Accuracy with Microsoft Fabric and Power BI at SSMH
Kanerika built a Fabric lakehouse that unified SQL Server and SharePoint data for Southern States Material Handling, with role-specific Power BI reporting. The client reports 90% data accuracy and KPI reliability and 85% higher operational visibility.
Read the Case Study →
Where Microsoft Rayfin Fits in the Application Stack
1. AI-Native Applications
An AI agent reading from OneLake gets the current state of the business, not a snapshot from an overnight pipeline run. That immediacy is the reason Rayfin was built with coding agents as a primary user from the start, not a capability bolted on afterward.
2. Data-Driven Business Applications
Applications that genuinely need operational and historical data in the same view benefit most, since that’s the exact seam Rayfin removes. A dashboard combining live transaction data with a year of historical trends no longer needs two separate systems reconciled after the fact.
3. Internal Enterprise Applications
Workflow apps, operations dashboards, and employee-facing tools are the cleanest fit right now. The payoff of skipping an ETL pipeline is immediate, and the audience is already inside the organization’s existing Entra tenant.
4. Rapid Prototyping to Production
A prototype built through Replit or GitHub Copilot can move to a governed production environment through the same deployment that built it, no separate hardening project required. Replit’s own framing at launch was blunt: agents write the code, Fabric ships it. That’s a real division of labor, since the backend Rayfin generates has to be production-grade the moment an agent deploys it, with no human reviewing the auth config in between.
Before You Build With Microsoft Rayfin
1. Preview Status and Product Maturity
Rayfin and Fabric Apps are both in public preview, not generally available, and multiple independent technical reviews are blunt that mission-critical workloads shouldn’t run on it yet. CLI commands, TypeScript decorators, and pricing are all still subject to change before GA. That’s not a reason to ignore Rayfin. It’s a reason to build the pilot somewhere that can tolerate breakage.
2. Fabric Dependencies
Rayfin only makes sense for organizations already committed to Fabric as their data platform. A team without that foundation would be adopting an entire platform dependency just to get a faster backend, which inverts the actual value proposition.
3. Technical Requirements and Learning Curve
A backend defined through TypeScript decorators is fast to provision but assumes real comfort with that pattern. A team standardized on a different stack will spend genuine time on the learning curve before Rayfin’s speed advantage shows up in practice.
4. Who Owns Security and Governance Here
Inherited governance is not the same as automatic governance. Someone still has to own whether the workspace’s existing RBAC and sensitivity labels are actually right for what a new application exposes, since Rayfin extends whatever policy already exists. It doesn’t audit it first.
Checklist
Microsoft Fabric Readiness Checklist
A planning guide for teams adopting Microsoft Fabric, covering data architecture, migration, governance, security, and workload optimization. Run it before an app inherits the tenant’s policies.
Get the Checklist →
How Kanerika Evaluates Rayfin for Clients Already Running Fabric
Kanerika holds Microsoft Fabric Featured Partner status and has been one of the earliest Microsoft Purview implementers globally, depth that carries real weight once an application’s governance has to hold up to the same scrutiny as its analytics. That combination is the lens Kanerika’s Fabric practice applies to Rayfin.
The evaluation starts with a basic filter. Does the client already run Fabric as its data platform, or would adopting Rayfin mean adopting an entire new platform dependency just to get a faster backend? For clients past that filter, the assessment turns on whether the application’s data needs and access policies genuinely benefit from sitting inside the same OneLake estate as everything else, or whether a standalone backend is still the more direct fit.
Given Rayfin’s current preview status, Kanerika isn’t recommending it yet for mission-critical production workloads. It’s being evaluated now, against real client Fabric estates, so the assessment isn’t starting from zero once general availability lands.
Talk to Kanerika
Get Your Fabric Data Estate Production-Ready Before You Build On It
Kanerika’s FLIP accelerator compresses Fabric migrations to approximately 90 days, with governance and OneLake structure configured from day one.
Book a Meeting →
Unifying Six Operational Systems for FoodPharma on Microsoft Fabric
Challenge
FoodPharma, a functional food manufacturer in Santa Fe Springs, California, ran six disconnected systems, NetSuite, RedZone, Parity Factory, UpKeep, Paychex, and Outlook. Every cross-functional report meant reconciling all six by hand, and a plant manager chasing a yield issue could wait two days for an answer.
Solution
Kanerika moved the six systems into Microsoft Fabric, consolidating over 50 tables and roughly a terabyte of historical data into OneLake with automated daily refreshes. Power BI became the one place to build reports and query operational data directly.
“Kanerika helped us build and migrate to a Microsoft open Lakehouse. Their initiative and troubleshooting support made the entire experience smooth and productive.”
Thu Nguyen, VP, FP&A and BI, FoodPharma
Results
- Reporting time dropped from two business days to about 90 minutes, a 32x turnaround
- The BI team recovered roughly 15 hours a week of manual reconciliation
- A yield question that took two days now gets answered by mid-morning
Watch on YouTube
Microsoft Fabric Case Study: FoodPharma Cuts Reporting to 90 Minutes
How Kanerika unified six operational systems on Microsoft Fabric for FoodPharma and cut cross-functional reporting from two business days to 90 minutes.
Conclusion
Microsoft Rayfin closes a real gap between AI-generated frontends and the governed backend infrastructure enterprise applications need. Deploying directly onto Fabric removes the ETL pipeline that normally separates operational apps from analytics, and inherited RBAC, sensitivity labels, and Purview policies mean governance doesn’t get bolted on later. It’s still public preview, not GA, with real gaps in connectors and undocumented licensing edges. For organizations already running Fabric as their data platform, though, it changes what standing up a new internal application actually requires.
FAQs
Is Microsoft Rayfin generally available?
No. Rayfin and the Fabric Apps preview both launched at Microsoft Build 2026 on June 2, 2026, and remain in public preview. Microsoft hasn’t announced a GA date. CLI commands, TypeScript decorators, and pricing are all documented as subject to change before general availability, so production-critical workloads should wait.
What does Microsoft Rayfin cost to run?
Rayfin has no separate pricing tier. It draws Capacity Units from the Fabric capacity already assigned to the workspace, at a fixed rate of 10 CUs per hour of GraphQL processing time, with SQL storage and compute billed separately at 0.383 vCores per CU.
Is Fabric Apps the same thing as Rayfin?
Not exactly. Rayfin is the open-source SDK and CLI a developer runs to define and deploy an app. Fabric Apps is the managed environment inside a Fabric tenant where that app actually runs. The two names get used almost interchangeably, which causes real confusion.
What authentication does a Rayfin app use in production?
Microsoft Entra ID is the exclusive authentication method for a Rayfin app once deployed to Fabric. Local development supports an email and password fallback for testing without a live deployment, but production apps run entirely on the organization’s existing Entra tenant identity.
Can I try Rayfin without deploying to Microsoft Fabric?
Yes, in a limited way. Rayfin supports a local, offline development experience with no cloud resources required, which Microsoft describes as experimental. It’s useful for evaluating the developer workflow, but the governance and OneLake integration that make Rayfin distinct only apply once deployed.
Do Rayfin app users need Power BI Pro licenses?
Below F64 Fabric capacity, viewers typically still need Power BI Pro licenses. At F64 and above, that requirement may not apply, though this specific behavior isn’t fully documented by Microsoft yet, so it’s worth confirming directly against a tenant before budgeting around it.
How is Rayfin different from Supabase or Firebase?
Supabase and Firebase are strong, mature backend-as-a-service platforms, but both run on their own infrastructure, disconnected from any broader analytics platform. Rayfin’s entire premise is deploying inside a Fabric tenant an organization already governs, so app data reaches Power BI and AI agents without a separate pipeline.
Which regions support Microsoft Rayfin today?
Rayfin’s underlying Fabric Apps preview isn’t available in every Azure region yet. Microsoft currently recommends West US 2 or North Central US for new capacity, per the Fabric Community’s own Rayfin AMA, with broader regional support expected to expand as the preview matures.
