Your data
Metadata only, masked before AI, purged on a schedule.
Written for the person who has to approve the tool, not for lawyers. The legal versions are in the privacy policy and the terms.
What is read
Object names, DAX expressions, Power Query scripts, relationships, descriptions, report pages, visuals, filters and bookmarks. That is the metadata of your model — and metadata can itself be sensitive: customer names in a measure, a server name in a query, a file path with a user name. We treat it as confidential.
What is never read
The rows of your data. Local cache files (cache.abf and other .pbi artifacts) are filtered out in your browser before the upload starts; they never reach our servers. No query is ever run against your sources.
What reaches the AI provider
Only when you tick “AI review”. Before sending, user names in Windows or Unix paths, server and host names, IP addresses and e-mail addresses are replaced by placeholders (<user_1>, <server_1>…) and restored only in the answer shown to you. The Gemini API is used under paid terms, under which Google does not use the content to train models.
Your own Gemini key
Pro and Max accounts can store their own Google AI Studio key. The AI review then runs on your Google project: your model metadata goes from our server to Google under your account and your terms, and Google bills you directly. The key is validated once, stored AES-256-GCM encrypted under a server secret, never logged and never displayed again (only its last four characters); you can remove it here or revoke it on Google's side at any time. Restrict the key to the Generative Language API and set a spending cap on the project.
What is stored, and for how long
Saved audits keep findings, scores and model metadata until you delete them. The model source needed by the fix workspace is stored encrypted (AES-256-GCM) for 30 days and then purged automatically. Private (unsaved) audits keep their result for at most 24 hours in a processing record.
Where
Database and authentication: Supabase, EU region. Application: Vercel. Payments: Stripe (we never see card numbers). AI: Google Gemini API (United States, EU-US Data Privacy Framework). The full list of sub-processors is in the privacy policy.
What we do not claim
The audit checks technical rules on metadata. It is not a Microsoft certification, not a GDPR certification and not a review of the permissions of your published environment. Findings are decision support; you validate before deploying.
Need a data processing agreement or a security questionnaire filled in? Write to [contact e-mail]. See also the privacy policy and the terms.