Last updated: March 2026
How Caraxe uses AI
Caraxe uses AI vision models to analyze screenshots captured during review sessions. This analysis helps detect design regressions, accessibility issues, layout problems, and other quality concerns that pixel-level comparison alone cannot catch.
What is sent to AI providers
- Screenshots of your app screens, captured during automated review sessions.
- UI structure metadata such as element types, positions, and accessibility labels.
Privacy redaction
Before any screenshot is sent to an AI provider, Caraxe applies a privacy redaction pipeline:
- Text input fields are detected and blurred.
- Password fields and editable regions are blurred.
- The blur radius defaults to a level that makes text unreadable.
This means the AI model sees the visual layout and design of your app, but not the actual text content typed into input fields.
AI provider
Caraxe uses a contracted third-party AI processing service for visual analysis. Customer artifacts are sent only after redaction and under the applicable API data-handling terms. The legal subprocessor is named in the DPA.
What AI does not do
- AI does not have access to your source code.
- AI does not have access to your APK binary contents beyond what is visible on screen.
- AI findings are not automatically acted upon. Findings are presented for human review.
- AI analysis is one input alongside local analysis. It does not replace deterministic checks.
Opting out
Vision analysis can be disabled per review configuration. When disabled, Caraxe still performs local analysis such as touch target checks, accessibility label detection, and layout heuristics.
Cost transparency
AI processing contributes to review costs. The internal ops dashboard tracks AI token usage and cost per review. This data informs plan pricing, but individual per-review AI costs are not currently exposed to end users.
Contact
For questions about AI processing, contact us at privacy@caraxe.app.