FabricSensei
Call
← Dojo

September 27, 2026 · Guide

Deploy your own Fabric demo for free

How I put a full Microsoft Fabric platform into a tenant in one afternoon with an open-source demo gallery, a trial capacity and an AI agent. The steps I ran, nothing else.

I needed a working Fabric platform to show a prospect. Client work is confidential, and the prospect had no Fabric yet. So I deployed an open-source demo into a tenant I administer and showed that instead. It took one afternoon and cost nothing. Here is how.

1. Where the demos are

The Fabric Demo Gallery is an open-source repo (MIT, with a Microsoft copyright; the GitHub org is not an official Microsoft org). Each demo is a complete platform: a lakehouse with sample data, bronze, silver and gold notebooks, a Direct Lake semantic model, a Power BI report and a pipeline. Twelve industries: retail, manufacturing, healthcare, financial services, energy, construction, education, hospitality, media, professional services, technology, transportation.

The repo has two ways in:

  • A hosted site, fabricdemogallery.com: sign in, pick a demo, deploy.
  • The same backend run on your own machine. This is the one I used, for the reason in step 4.

2. Pick a demo

Pick the one closest to the data your audience knows. I showed orders and inventory, so I picked Retail “Sales & Inventory”: 100,000 sales lines, 500 products, 30 stores, 15,000 inventory rows, three report pages (Sales Overview, Inventory & Risk, Margin & Basket Analysis). Each demo’s manifest.json in demos/<name>/ lists its items, sample data and an estimated deploy time of 8 to 12 minutes.

Pick a “Standard” scenario. The AI agent and Copilot scenarios need features a trial does not have.

3. Get a capacity

A Fabric trial is free for 60 days and runs the standard demos on its smallest size (4 capacity units). Start one from the Fabric portal as a tenant admin. Items are deleted 7 days after the trial ends, so plan the demo inside that window.

You also need the Azure CLI signed in to that tenant. No Azure subscription is required:

az login --tenant <your-tenant> --allow-no-subscriptions

4. Deploy

I tried the hosted site first. It asks for consent to six Fabric scopes from an unverified app, and the tenant answered “Need admin approval”. So I ran the gallery’s own backend locally. It then takes its tokens from the Azure CLI, and no third-party site ever holds them.

git clone https://github.com/Fabric-Demo-Gallery/fabric-demo-gallery
cd fabric-demo-gallery/backend
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
uvicorn app.main:app --port 8000

With no WEBSITE_SITE_NAME set, the backend runs in dev mode and uses your az sign-in. Then one request starts the deploy and streams its progress:

curl -N -X POST http://localhost:8000/api/deploy/retail-sales \
  -H 'Content-Type: application/json' \
  -d '{"demo_id":"retail-sales","capacity_id":"<capacity id>","workspace_name":"Fabric Demo"}'

The capacity id comes from the Fabric admin portal, or from GET https://api.fabric.microsoft.com/v1/capacities. The deploy creates the workspace, uploads the data, runs the three notebooks in order, refreshes the model and creates the pipeline. On 4 capacity units the notebooks took about 6 minutes. Spark can return a “capacity busy” error on the smallest size. The deployer retries on its own.

I wanted the demo inside an existing workspace, not a new one, so I created a folder there and moved the items in with the Fabric REST API afterwards. Skip that if a fresh workspace is fine.

5. Check it

Read-only, before you show anyone:

  • The workspace lists the lakehouse, three notebooks, the model, the report and the pipeline.
  • The gold tables have rows.
  • The three report pages render.
  • The pipeline has no schedule, unless you want a nightly load.

6. How I used it

A 15-minute walkthrough: the lakehouse layers, the gold notebook, the report, the pipeline, then a map from each layer to what the prospect’s own build would look like. I said it plainly: this is open-source sample work, similar to what we build, not our build.

Do it with an agent

Every step above is a command or an API call, so an AI agent can run them for you. I ran them from Claude Code. The agent cloned the repo, started the backend, sent the deploy request, watched the stream, and verified the items with read-only API calls. I signed in and gave the go on every write to the tenant. Keep that gate: the agent never creates, deletes or changes a tenant item without your word.

Want it deployed for you, or on your data? Book a call.