What is FabricSensei? A toolkit and a channel for Fabric and Power BI with AI agents
A command line that connects any AI agent to every Fabric tenant you work in, free browser tools like the Theme Generator, and the videos that show the method. Built by a consultant who runs client work this way.
FabricSensei is a toolkit and a channel for people who build on Microsoft Fabric and Power BI with AI agents. Three parts: a command line that connects any agent to every tenant you work in, free browser tools like the Theme Generator, and the videos that show the whole method.
It is built by a consultant who runs client work this way every day. Everything on this page exists today or has a version number on it. The CLI is pre-release and read-only, and the page says so.
The series
Three short videos, one playlist.
| # | Video | Question |
|---|---|---|
| 1 | The brand | What is FabricSensei? |
| 2 | The CLI | What does the FabricSensei CLI do? |
| 3 | The toolkit | What tools does FabricSensei offer? |
Who is FabricSensei for?
- You build on Microsoft Fabric or Power BI, for one company or for many clients.
- You want an AI agent to do the repetitive work, and you want a gate between the agent and production.
- You do not want a hosted service holding your tokens or reading your tenant.
- You want the method, shown on real work, not a course.
What does the FabricSensei CLI do?
fs answers one question for every client in one place: which semantic models and jobs refreshed, which failed, when, and how long each run took. A person gets screens. An agent gets the same commands with --json. It is local-first: no hosted service, no telemetry, credentials in the OS keychain.
| Command | What it does |
|---|---|
fs |
The start screen: clients, workspaces, percent healthy, a menu. No tenant call. |
fs radar |
Every watched model, failed first. Checks each client every 30 minutes, a running model every minute. |
fs blip |
One fresh read, printed once: the 5 newest failures, the 5 newest good runs, one summary line. Models and run items: notebooks, pipelines, Dataflows Gen2, Spark job definitions, copy jobs. |
fs logs |
The local activity journal. --sweep reads the refresh and job history into it. |
fs connect |
Add a client: name, work email, domain. The tenant id comes from Microsoft’s public lookup. One sign-in, token to the keychain. |
fs clients, fs auth, fs selftest, fs ws |
Clients and watched workspaces, sign-in, machine check, workspace inventory and audit. |
Rules that hold in every version:
- Screens for people, print for scripts. With
--json, a flag, or no terminal, nothing opens a screen and nothing asks a question. - An agent drives
fsand never holds a token. One file is the whole agent contract: every command, the--jsonshapes, the exit codes, a “never do” list. - 180 tenant calls per client per hour. A read that would go over stops and says so.
- Exit codes: 0 clean, 1 findings, 2 the question could not be answered.
Where it is today
Version 0.3, pre-release, read-only. Nothing it does changes a tenant. No package release yet. The road to 1.0:
| Version | Adds |
|---|---|
| 0.3 | the screens, read |
| 0.4 | run items: notebooks, pipelines, dataflows, jobs |
| 0.5 | write safety, then refresh, run, cancel, SQL endpoint |
| 0.6 | deploy, parameters, rebind, Git sync, deployment pipelines |
| 0.7 | schedules, permissions, the agent kit |
| 0.8 | dogfood, docs, demo |
| 1.0 | package release and site launch |
Deploy and lint wrap Microsoft’s own tools rather than replace them. Every write goes through a gate and a per-workspace switch.
What tools does FabricSensei offer?
| Tool | State | What it is |
|---|---|---|
| Theme Generator | live, free | Pick a preset, set your brand color, a real report page restyles live. Download the theme JSON or the template report with the theme inside. |
| Template report | live, free | One page, fictional data, every color from the theme. The download behind the generator and the demo model in every video. |
| Blueprint | next | One zip: a starter Power BI Project, the rules pack an agent reads, the guide with the Power BI standard and the audit checklist, light and dark themes. The kit used on client work, packaged. |
| The CLI | pre-release | Above. |
What is the method?
The videos, in the order to watch them:
- Git is step 0. Save as a Power BI Project, put it in Git, no capacity, no terminal. 3 videos.
- AI on the model. What an agent can do, the loop, what goes wrong, the stack. 4 videos.
- Power BI on a Mac. Text, an agent, publish from code. Desktop for design only. 3 videos.
- The Theme Generator. The 60-second tour, and how it was built with an agent. 2 videos.
Every video answers one question in about 80 seconds. Every page has the prompts and the numbers.
Result
What the method runs on, as of this page:
| Measure | Value |
|---|---|
| Client engagements on the method | 4 |
| Power BI Projects in Git | 99 |
| Merged pull requests that changed a model | 282 |
| CLI commands, version 0.3 | 12 |
| Run item types the CLI reads | 5 |
| Videos written, 3 playlists | 12 |
Videos
Videos 1 to 3 and their transcripts land here after the recording.
