Free static audit · no account · nothing stored
Find what will break your Power BI model before your users do.
Drop a PBIP. In two seconds: a health score and every relationship, DAX and Power Query risk with the exact object. With the AI review: one verified, paste-ready fix per risky measure.
- 363
- checks proven from the model
- 2 s
- for the static audit
- 6
- verified fixes on the sample, score 68
Northwind Retail.pbip
cache.abf removed in the browser · 412 KB analysed
parsing · 138 checks · lineage
Health score
0/ 100
Not ready for production
50 findings · 7 critical · 6 verified fixes
Proposed rewrite · Share of Country reviewer: intended change
currentDIVIDE(
[Total Sales],
CALCULATE(
[Total Sales],
ALL( Sales[CustomerCode] )
)
)proposed · paste-readyDIVIDE(
[Total Sales],
CALCULATE(
[Total Sales],
ALL( Customer )
)
)The invisible risk
A report can look right today and still be a liability.
The demo works. The numbers look plausible. Then a scheduled refresh fails on a laptop path, a share shows 100 % in every row because it divides by itself, and a year-over-year mixes two exchange rates. None of it is visible from the report — all of it is visible from the model definition.
- · Local paths, hard-coded servers and secrets in Power Query
- · Denominators that keep the filter they should clear
- · Time intelligence on an unmarked or non-contiguous date table
- · Bidirectional filters that double counts, iterators that freeze pages
Sales performance · FY2026
Total sales
€ 48.2M
+12.4 % YoY
Margin
31.8 %
+1.2 pts
Customers
12 480
+640
Monthly sales vs last year
How it works
From upload to a fix you can paste, without a black box.
1. Upload the PBIP
Folder or zip from Power BI Desktop. Cache files are stripped before anything leaves the browser.
2. Static analysis
Parsing, 363 checks, relationship graph, model ↔ report lineage. Always free, always instant.
3. AI architecture review
One model-wide pass finds root causes across measure families and plans which objects deserve a rewrite.
4. Verified fixes
One focused rewrite per object, checked for syntax and references, then challenged by an independent reviewer for semantic equivalence.
Every AI rewrite carries the reviewer's verdict (equivalent, intended change, needs business validation) so you know how much to trust it before pasting.
Verified fixes
Not a suggestion. A rewrite you can paste, checked twice.
Every AI rewrite is verified for syntax and references against your model, then challenged by an independent reviewer for semantic equivalence. The verdict travels with the code.
Share of Country· Northwind Retail sample reviewer: intended change
+1 · −1 tokens
DIVIDE( [Total Sales], CALCULATE( [Total Sales], ALL(Sales[CustomerCode]Customer ) ) )
- 1
Root cause, not symptom
The denominator cleared the fact-table column, so every customer showed 100 %. The rewrite clears the dimension the slicer filters.
- 2
Verified against your model
Syntax, every referenced table and column, M formatting: checked before you see it.
- 3
Reviewed independently
A second model reads the original and the rewrite and rules: equivalent, intended change, or needs business validation.
- Object names are masked before the model sees them and restored only in what you read.
What is checked
Model, DAX, Power Query and the report, in the same audit.
Static checks are proven from the code and never guess. The AI review adds what static analysis cannot see: root causes across measures and paste-ready fixes.
Model
- · Bidirectional filters on facts
- · Missing or ambiguous cardinality
- · Snowflakes that should be one dimension
- · Numeric columns stored as text
- · Calculated columns typed implicitly
DAX
- · Iterators over the fact table
- · FILTER(ALL(table)) instead of column filters
- · Hardcoded years in time intelligence
- · Duplicated business logic
- · Wrong denominators in shares and ratios
- · Context-transition mistakes
Power Query
- · Local file paths and hardcoded servers
- · Steps that break query folding
- · Repeated source-acquisition patterns
Report & governance
- · Pages with dozens of visuals and interactions
- · Visual-level filters overriding page filters
- · Bookmarks and hidden-state navigation
- · Custom visuals inventory
- · Unused measures, missing descriptions
- · RLS roles with permissive filters
What you hand over
The audit is the start. The deliverables are why professionals pay.
Branded report
A client-ready PDF: verdict, score, fix-first list, findings by area, verified rewrites with the reviewer's verdict, accepted risks. Your logo, your colours.
Compare
Re-upload after your changes: fixed, new and regressed findings per object, score trend.
Your standards
Switch checks off, change severities, add naming conventions and a quality gate. Every audit is judged against your standard.
Documentation
Data dictionary as Excel or Markdown, schema diagram as SVG or PNG, waivers for accepted risks.
Auto-documentation
Descriptions, a display-folder taxonomy and format strings for every measure and column, written by the AI and applied to your TMDL.
Impact & cleanup
What breaks if you change an object; unused measures and columns removed from the model in one download.
Performance
Upload the .vpax: the columns that are 80 % of the model, referential-integrity violations, what each fix saves.
Roadmap & test pack
A phased refactoring plan with effort, and DAX queries that prove each rewrite on your data.
Who it is for
Three jobs, one audit.
Consultant
Hand over a report in an hour, not a week.
Audit the client's model, export the branded PDF and the corrected TMDL, bill the day you saved. A single audit is 9 €; the pack of five covers a quarter.
Audit a client modelBI lead
A quality gate before anything is published.
Your standards (disabled checks, severities, naming rules), waivers for accepted risks, compare before/after, the score trend per model. Every release judged the same way.
See the deliverablesAnalyst
Learn from every fix on your own model.
Each finding says what is wrong, why, and shows the rewrite next to the original with the reviewer's verdict. The checks catalog explains all 138 patterns.
Browse the checks60-second tour
Upload, score, fix. That is the whole product.
Northwind Retail.pbip
cache.abf removed in the browser · 412 KB analysed
parsing · 138 checks · lineage
Health score
0/ 100
Not ready for production
50 findings · 7 critical · 6 verified fixes
Proposed rewrite · Share of Country reviewer: intended change
currentDIVIDE(
[Total Sales],
CALCULATE(
[Total Sales],
ALL( Sales[CustomerCode] )
)
)proposed · paste-readyDIVIDE(
[Total Sales],
CALCULATE(
[Total Sales],
ALL( Customer )
)
)Live walkthrough on the retail sample · a recorded tour replaces it at launch
Your data
Metadata only. Masked before any AI call. Deleted on a schedule.
The audit reads object names, DAX, Power Query and report definitions. It never reads your business rows.
Your PBIP
Unzipped in your browser. cache.abf and localSettings.json — the only files with data — never leave your machine.
Names masked
User names in paths, server names and e-mails become placeholders before any AI call. Restored only in what you read.
Gemini, paid tier
Metadata and DAX only, under Google's paid terms: not used for training, not retained.
Encrypted at rest
The model definition is stored AES-256-GCM so the fix workspace can show diffs. EU region.
Purged after 30 days
Automatically, every hour. Delete your account and everything goes with it, immediately.
Bring your own Gemini key on Pro and Max: your models never touch our key.
Pricing
Static audits are free. Pay for the AI and the deliverables.
Prices excl. VAT. Cancel any time from the billing portal; the plan stays active until the end of the paid period.
Free
Every static check, on every model, forever.
- Unlimited static audits (363 model and report checks)
- Schema diagram, relationships, data dictionary, JSON export
- Saved audits and model projects: reopen, triage, track the score
- Findings marked as reviewed, command palette, dark mode
Single audit
One client model, one deep dive. No subscription.
- Everything in Free
- AI review: root causes across measures
- Verified, paste-ready DAX and M rewrites
- Private mode (nothing saved)
- Pack of 5 for 39 €, credits never expire
Pro
For the professional who ships models every month.
or 790 € per year
- 10 AI audits per month — or unlimited with your own Gemini key
- Branded PDF report with your logo, client-ready
- Compare: what changed since the last audit
- Your standards: custom rules, naming, severity, quality gate
- Waivers, documentation export, corrected model export
- 1-year history, sharing, private mode
Max
Hand over a finished, documented model.
or 1990 € per year
- 18 AI audits per month (unlimited with your key), everything in Pro
- Auto-documentation: descriptions, folders, formats for every measure, as TMDL
- Impact analysis and cleanup export (unused objects removed)
- Performance section from your VertiPaq stats
- Refactoring roadmap and regression test pack
- White-label report: no Auditabi mention
FAQ
What to know before your first audit.
General
It is a Power BI audit workspace. You upload a PBIP bundle or a raw TMDL export, inspect semantic-model and report findings, review recommendations, and reopen saved audits later.
The primary audience is BI analysts and Power BI developers. Consultants and other Power BI professionals can still use it, but the workflow is being designed around day-to-day model review.
The preferred input is a PBIP zip or parent folder containing the *.SemanticModel and optionally the *.Report folder. The fallback is a raw .tmdl or .txt export of the semantic model.
Audit engine
Static analysis does the structural work and is free: parsing, 363 checks, relationship graph, lineage between measures and visuals. The optional AI review runs in two passes: one model-wide architecture review that finds root causes across measures, then one focused call per risky object that writes a paste-ready rewrite. Every rewrite is verified against your model (balanced syntax, every referenced object exists, business semantics preserved) before you see it.
The current workspace focuses on DAX anti-patterns, model relationships, report complexity, naming and maintainability issues, and selected transformation hotspots that are likely to hurt refresh or usability.
Yes. The audit reconstructs the relationship graph so it can highlight risky bidirectional filters, ambiguous paths, and weak star-schema boundaries.
Privacy
The product is intended to work from semantic-model metadata and report-definition metadata such as object names, relationships, DAX, Power Query code, pages, visuals, and filters. It is not meant to ingest the imported business rows from local cache artifacts such as cache.abf.
A private audit record with findings, scores, object names and model metadata, so you can reopen it. The model source itself is stored encrypted (AES-256-GCM) for 30 days so the fix workspace can show diffs, then deleted. Before anything reaches the AI provider, user names in file paths, server names and e-mail addresses are masked. Local cache artifacts such as cache.abf never leave your machine.
Pricing
Unlimited static audits on any model, the full findings list, the schema diagram, relationships, the data dictionary, JSON export, and saved audits you can reopen and triage. The AI review (root causes and verified rewrites) is paid: 9 € for one audit, 39 € for five, or a Pro plan. The five sample audits show what it produces before you spend anything.
Buy a single audit (9 €) or a pack of 5 (39 €) when you review a client model occasionally. Pro (79 €/month) is for the professional who ships models every month: branded report, re-audit comparison, your own standards, waivers, documentation export. Max (199 €/month) adds the deep AI features: auto-documentation, impact analysis, cleanup export, performance and refactoring roadmap.
Pro includes 10 AI audits per rolling 30 days, Max 18 (at most 6 per day). Each AI audit runs several minutes of a frontier model on your whole semantic model. Static audits are never limited, and extra AI audits can always be bought as singles (9 €) or packs of 5 (39 €).
Pro and Max accounts can store their own Google AI Studio key. Every AI audit then runs on your Google project and Google bills you directly — about 1.5 € per audit at list price for a 100-measure model. In exchange the plan's monthly quota disappears (fair use 20 per day). You pay us for the 363 checks, the verification, the report and the workspace, not for tokens. The key is stored encrypted and can be removed at any time.
Yes. Upgrade, cancel or download invoices any time from the billing portal in the app. Cancelling keeps Pro active until the end of the paid period.
Your next audit takes two minutes.
Upload a PBIP, read the findings, decide. The static audit is free and unlimited; the AI review is there when you need the fix, not just the diagnosis.
Audit a model for free