Trust
Security and data handling
Project documents often contain confidential commercial, technical and operational information. This page describes exactly how Uozan handles them. It makes no claim of certification.
Last updated 16 October 2026
Access control and isolation
- Every API request carries a verified sign-in token. Identity is taken from the token, never from the request.
- Every project, document, assessment and report is checked against its owner on the server before it is read, changed, exported or deleted. Another account’s project behaves exactly as if it did not exist.
- The database has no client-side access rules: browsers cannot read or write it directly.
- Subscription status is verified on the server from Stripe; usage limits are enforced in database transactions, so neither can be bypassed from the browser.
Storage
- Original files are kept in private cloud storage (Google Cloud, via Firebase) and are reachable only through signed links that expire after five minutes, issued to the owner.
- Extracted text, items and reports are stored in Google Cloud Firestore, encrypted at rest by the provider.
- Files are validated by content before processing; legacy binary Office formats and executables are rejected; archives are size-checked before they are expanded.
AI processing and prompt-injection safeguards
- Tables and registers are processed by Uozan’s own code. Only narrative text is sent to an AI model, through Vercel’s AI Gateway, to a model provider (currently Google Gemini for extraction). The executive summary is drafted by a second model (currently Anthropic Claude) from the findings text, which can include item names quoted from your documents.
- Your documents are not used to train models.
- Document text is passed to models as data, with instructions to ignore any instructions it contains. Models have no tools and cannot take actions. Everything a model returns is checked against the source text, and anything that does not appear there is discarded — so text hidden in a document cannot inject findings, change other results or reach other customers’ data.
- Usage records hold token counts, timings and identifiers — never document content.
Processors
- Google Cloud / Firebase — authentication, database, file storage.
- Vercel — hosting and AI Gateway.
- Google and Anthropic — AI model processing as described above.
- Stripe — payments. Card details are entered on Stripe’s page and never reach Uozan.
- PostHog — cookieless product analytics on page and feature use. No document content.
Retention and deletion
- Delete a document or a whole project at any time. Deleting a project removes its original files, extracted text, items, assessments, reports and review history.
- When a subscription ends, project data remains available to export for 30 days and is then deleted automatically.
- Cost and usage records (identifiers and token counts only) are kept for accounting.
Reporting a vulnerability
Please report security issues through conjora.ie/support. We will acknowledge and investigate every report.