Unlock Power BI and Fabric development: Git is step 0
Save as a Power BI Project and the model becomes text. Git then gives you history, undo, review, and a model an AI agent can work on safely. No Fabric capacity, no terminal.
Save your report as a Power BI Project (PBIP) and the model becomes text files, so Git can track every change. That one step unlocks history, undo, review, and a model an AI agent can work on safely. It needs no Fabric capacity and no terminal.
This is step 0 for modern Fabric development. Before you add an agent, before you add a pipeline, put the work in Git.
The series
Seven short videos, one playlist. Each answers one question.
| # | Unlock | Question |
|---|---|---|
| 1 | The file | How do I version control Power BI? |
| 2 | The history | Do I need Fabric capacity to use Git with Power BI? |
| 3 | The agent | How can an AI agent work on my Power BI model safely? |
| 4 | The cool things | How can I use AI to work on Power BI models? |
| 5 | The workflow | What is the workflow for AI in Power BI? |
| 6 | The gotchas | What can go wrong with AI in Power BI? |
| 7 | The stack | How can I make AI better at Power BI and Fabric? |
Videos 1 to 3 are on this page. Videos 4 to 7 have their own page and land one week later.
Why Git is step 0
- A number on the dashboard moved last week. Nobody knows which measure changed, or who changed it.
- Your folder has
report_final.pbix,report_final_v2.pbix, andreport_final_v3_USE_THIS.pbix. - Power BI Desktop saved the file and the model no longer opens. There is no way back.
- Two people edited the same .pbix. One of them lost a day of work.
- You let an AI agent edit the model. It changed something. You cannot see what.
One cause for all five: a .pbix is one binary file. Git cannot read it, and neither can an agent. A Power BI Project is text. Both can.
The same is true across Fabric. Notebooks, pipelines, and semantic models all have a text form. Once the work is text, every tool that reads text can help you: Git, a reviewer, an agent.
Unlock 1, the file: how do I version control Power BI?
In Power BI Desktop, choose File, Save as, and pick Power BI Project. You get a folder instead of one file:
Sales.pbip
Sales.SemanticModel/
definition/
model.tmdl
relationships.tmdl
tables/
Sales.tmdl
Date.tmdl
Sales.Report/
definition/
report.json
pages/
- The model is TMDL (Tabular Model Definition Language). Every table, column, measure, and relationship is a block of text.
- The report is JSON. Every page and visual is a block of text.
- Git tracks text. A change to one measure is one changed line in one file.
That is the whole trick. Everything below follows from it.
Unlock 2, the history: GitHub without a terminal
You do not need the command line. The GitHub website does all of this.
- Save the report as a Power BI Project. Close Desktop.
- On GitHub: New repository, then Upload files. Drag the project folder in. Leave out the
.pbixand the.pbicache folder. - Open
tables/Sales.tmdlin the browser, click the pencil, add or change one measure, and commit. - Open the commit. The diff shows the lines you changed and nothing else.
- Open History. Every change, who made it, and when.
What you get once it is there:
- History. One measure file, every change, who, when.
- Undo. A Desktop save normalized a model and it no longer loaded. One revert and it was back, same day.
- Review. A pull request shows the diff before it reaches the report. One review caught a column block nested inside another column, which had broken the import to the Power BI service.
- Team work. Two people, two branches, one merge. No more lost day.
When you want a .gitignore, these five lines keep the binary and the Desktop cache out:
*.pbix
**/.pbi/cache.abf
**/.pbi/localSettings.json
**/.pbi/editorSettings.json
**/diagramLayout.json
Do I need Fabric capacity to use Git with Power BI?
No. Plain Git works on any Power BI licence, with GitHub or Azure DevOps, on your machine or in the browser. That is what this page shows, and it is how we run every client engagement.
Microsoft Fabric also has a built-in Git integration. It connects one workspace to one branch and syncs from the service. It needs a Fabric capacity (a trial counts) or a Power BI Premium capacity, two tenant switches, and a workspace admin. Report and semantic model items were still marked preview on 2026-09-01 (Microsoft Learn, “Get started with Git integration”). Use it when you have capacity and want the workspace itself to sync. Use plain Git when you work local.
Unlock 3, the agent: how can an AI agent work on my Power BI model safely?
An agent can only edit what it can read. A .pbix, it cannot read. A Power BI Project in Git, it can: a model to read, a branch to work on, and a diff you check before anything ships. The gate is the same for a person and for an agent: nothing merges without a diff someone read.
Paste this into your AI coding agent (any agent that can run commands) from the folder that holds your .pbix files:
I have Power BI files in this folder. Set them up for Git.
1. For each .pbix, tell me to open it in Power BI Desktop and save it as a Power BI Project (PBIP) into this folder. Wait for me.
2. Create a .gitignore that excludes *.pbix, **/.pbi/cache.abf, **/.pbi/localSettings.json, **/.pbi/editorSettings.json, and **/diagramLayout.json.
3. Run git init, add everything, and make a first commit.
4. Show me the folder layout and explain, in 5 lines, what TMDL is and where my measures live.
5. From now on, before any change: make a branch. After any change I save in Desktop: run git status, keep only the files I meant to change, restore the rest, show me the diff, and open a pull request. Never merge without me.
Unlocks 4 to 7, AI on the model
Videos 4 to 7 show what an AI agent can do on the model once it is text, the loop it runs in (issue, branch, edit, gate, pull request, you merge), what goes wrong and what catches it, and the stack we run to make the agent good. They get their own page.
Result
Our own engagements since January 2026, all on plain Git:
| Measure | Value |
|---|---|
| Power BI Projects in Git | 99 |
| Merged pull requests that changed a model | 282 |
| Median lines changed per pull request | 3 to 110 |
| Smallest change | 1 line |
| Models restored from history after a bad save | 3 |
Videos
Videos 1 to 3 and their transcripts land here after the recording. Videos 4 to 7 link here when they ship.
