Nikolajunserrichter Fabric Apps Skill Fabric Apps

Microsoft Fabric Apps (Rayfin)

What this is

Fabric Apps (preview) is a Microsoft Fabric workload for building full-stack, data-driven web apps. You write TypeScript entity classes decorated with @entity(), @text(), @role(), etc., and the Rayfin CLI compiles them into a managed SQL database in Fabric, a GraphQL Data API, row-level authorization, and static hosting — all behind a single app endpoint with Microsoft Entra SSO.

The whole platform is the "Rayfin" toolchain. Three pieces show up constantly:

• @microsoft/rayfin-core — decorators you use to define data models (@entity, @text, @role, …).
• @microsoft/rayfin-client — the type-safe RayfinClient your frontend uses to read/write data.
• @microsoft/rayfin-cli (rayfin / npx rayfin) — scaffold, run locally, apply schema, deploy.

Mental model of the data flow — keep this in mind, because every "how do I…" question maps to one stage:

TypeScript entity classes (rayfin/data/*.ts)
  → decorators (@entity, @text, @one, @role …)
    → CLI compiles to → SQL schema + GraphQL API + auth policies
      → RayfinClient (generated, type-safe) → your frontend (React/Vue/Svelte/…)

When to reach for it (and when not)

Good fit: rapid prototypes, internal tools/dashboards, data-exploration UIs, and structured backends for AI/agent apps — anything where you want a live, authenticated, data-backed app without writing backend boilerplate.

Bad fit (tell the user up front, don't try to force it):

• Not Azure Service Fabric, Fabric notebooks/lakehouses/pipelines, or Power BI. If the user means one of those, this skill does not apply.
• No custom auth providers — only Entra SSO (deployed) and email/password (local dev only).
• No existing/external database — Fabric Apps owns the schema; you can't point it at one.
• No composite keys, no many-to-many (use a join entity), no stored procedures / multi-step transactions, TypeScript only for models.

Before writing code

This is a preview product and the API surface moves. Your training data may be stale or invented. Before generating non-trivial Rayfin code, verify decorator names, client method signatures, and CLI flags against the live docs using the microsoft_docs_search / microsoft_docs_fetch MCP tools (or microsoft-docs skill) and the reference files in this skill. Inventing a decorator or a RayfinClient method that doesn't exist is the most common failure mode here — a quick lookup prevents it.

If the user hasn't confirmed prerequisites, remind them once: the workspace needs Fabric capacity, and a tenant admin must enable the "Fabric Apps (preview)" workload before any deploy will work.

The core workflow

Follow these stages. Most requests are "do one stage" — identify which and go there.

1. Scaffold or initialize

New project from a template:

npm create @microsoft/rayfin@latest my-app   # add --workspace <name> to target a workspace
cd my-app

Add Rayfin to an existing/empty directory instead (interactive: pick services, dialect, hosting):

npx rayfin init .
# non-interactive example:
npx rayfin init my-app --services db,storage --auth-methods fabric --static-hosting --overwrite

This produces the standard layout (see references/cli-and-config.md for the full tree):

your-project/
├── rayfin/
│   ├── data/          # entity classes + schema.ts  ← you spend most time here
│   ├── rayfin.yml     # backend service config (auth/data/storage/staticHosting)
│   ├── .env           # local secrets/values for interpolation (gitignore it)
│   └── tsconfig.json  # generated; don't edit
├── src/               # frontend (React/Vue/… depending on template)
└── package.json

2. Define data models

Each entity is a class in rayfin/data/*.ts. This is the heart of the platform — get the decorators right and everything downstream (DB, API, types) follows. Read references/data-modeling.md for the full decorator reference, relationships, and the gotchas; the essentials:

import { entity, uuid, text, int, decimal, boolean, date, set, one, many } from '@microsoft/rayfin-core';

@entity()
export class Notebook {
  @uuid() id!: string;
  @text({ min: 1, max: 100 }) name!: string;
  @date() createdAt!: Date;
  @many(() => Note) notes?: Note[];        // one-to-many: @many on parent
}

@entity()
export class Note {
  @uuid() id!: string;
  @text() title!: string;
  @text({ optional: true }) content?: string;
  @boolean({ default: false }) isPinned!: boolean;
  @date() createdAt!: Date;
  @one(() => Notebook) notebook?: Notebook;  // many-to-one: @one on child, auto-creates notebook_id FK
}

Three rules that bite people constantly (full list in the reference):

• @entity() ≠ TypeScript optional. A trailing ? only changes the static type. To make a column nullable you must pass { optional: true } to the decorator. Use ! to tell TypeScript the framework initializes required fields.
• Register every entity in rayfin/data/schema.ts or the client can't see it.
• Use relative imports with .js extensions between entity files (from './Note.js') — the emitted ESM needs them.

3. Add authorization (@role)

Permissions live next to the data, as class-level @role(roleName, actions, options?) decorators. The only built-in role is authenticated. Policies are type-safe expressions over JWT claims and the item. Owner-only is the workhorse pattern:

@entity()
@role('authenticated', '*', {
  policy: (claims, item) => claims.sub.eq(item.user_id),
})
export class Task {
  @uuid() id!: string;
  @text() title!: string;
  @text() user_id!: string;   // populate from claims.sub — no @one() to the system USER entity
}

See references/data-modeling.md for field-level include/exclude, per-action rules, .and()/.or(), and admin-override patterns.

4. Use the data from the frontend

The generated RayfinClient gives a fluent, type-safe API — no raw GraphQL. Read references/client-api.md for the full CRUD/query/pagination/auth surface; the shape:

import { RayfinClient } from '@microsoft/rayfin-client';
import type { Note } from '../rayfin/data/Note';

type AppSchema = { Note: Note };

const client = new RayfinClient<AppSchema>({
  baseUrl: import.meta.env.VITE_RAYFIN_API_URL ?? 'http://localhost:5168',
  publishableKey: import.meta.env.VITE_RAYFIN_PUBLISHABLE_KEY,
});

const pinned = await client.data.Note
  .select(['id', 'title', 'notebook.name'])
  .where({ isPinned: { eq: true } })
  .orderBy({ createdAt: 'desc' })
  .execute();

await client.data.Note.create({ title: 'Standup', isPinned: false, createdAt: new Date() });
await client.data.Note.update({ id }, { title: 'Renamed' });
await client.data.Note.delete({ id });

Note the limitations you'll hit: no count() (use results.length), PagedResult.totalCount isn't populated, select only the fields you need.

5. Run locally, then deploy

npm run dev          # frontend dev server (Vite-based templates)
npx rayfin up        # deploy to Fabric: creates the item, applies schema, builds + uploads static content
npx rayfin up --dry-run --verbose   # preview without changing anything
npx rayfin up db apply              # apply just schema changes (add --force for destructive ops — data loss risk)

rayfin up requires sign-in (it'll prompt) and Fabric auth enabled in rayfin.yml (services.auth.fabric.enabled: true). Email/password works locally but never on a deployed app. Full command/flag reference and rayfin.yml schema are in references/cli-and-config.md.

Schema changes & caching gotcha

After editing entities, run npx rayfin up db apply. If the frontend keeps returning stale data shapes after a schema change, the generated config under rayfin/.temp/ is cached — stop the dev stack, rm -rf rayfin/.temp/, and restart. Never commit .temp/.

Security reminders to surface

• Static content is served from a public URL — keep secrets, API keys, and sensitive data out of frontend code and the repo. Use the publishable key (it's meant to be public); never embed real secrets.
• Sharing a deployed app needs the Run and interact item permission; deploying/editing needs Edit (Write). Workspace roles don't override item-level permissions.

Reference files

Load these as needed — they hold the depth that doesn't belong in this overview:

• references/data-modeling.md — every field decorator, type modifiers, relationships, foreign keys, the full @role/policy/field-permission system, and modeling gotchas.
• references/client-api.md — RayfinClient init, read/filter/sort/relationship-navigation, pagination, create/update/delete (incl. relationships), auth, and current limitations.
• references/cli-and-config.md — full Rayfin CLI command + flag reference, the rayfin.yml service schema, project structure, env vars, and the deployment sequence.

Source: NikolajUnserRichter/fabric-apps-skill — distributed by TomeVault.