Protecting your integration
Public access uses HTTPS. API keys are stored as hashes and belong on your backend. Model workers and the database are private. Request sizes, concurrency and account attempts are bounded. Account spending caps can stop further analyses.
Customers can add an authenticator app as a second sign-in step from Account. One-time recovery codes are shown at enrollment and stored only as hashes. Keep them somewhere separate from your password.
What we keep, and for how long
Uploaded images are processed in memory or temporary files and discarded after processing. API response bodies are retained for up to 30 days to support idempotent retries; some responses may contain extracted text or other personal data.
Request history includes source IP and limited input metadata. Playground calls also create zero-credit history entries with a limited outcome summary, without the submitted image or text. Moderation outcomes and source IP are cleared after 30 days by daily maintenance. Encrypted backups expire after seven days, so deleted records may remain in a backup for up to seven additional days. Accounting metadata is retained separately. Logs omit request bodies, authorization headers and query strings.
External fallback
Primary photo screening runs within our service. External fallback is off by default. It requires an enabled provider, approved processing terms and account authorization. A malformed image, a successful primary decision, or exhausted processing capacity does not trigger it.
Content screening signals
The authenticated English text beta processes submitted text locally. It does not store the text or matched contact values. Its generic scores and review flags use the existing 30-day replay retention. It never automatically rejects content or bans accounts.
AI-generated image detection returns a confidence percentage. Its score is not proof that an image is synthetic, and modern generators or edited photos may defeat it. The API returns only a confidence percentage: 90% or higher means most likely AI-generated. Your application decides how to use the score; the detector does not return review or rejection statuses.
Build a review path around the signal
Submit only content you have the right to process. Keep unavailable uploads pending review. Provide human review and appeals in your app. Do not use these tools as identity, age or legal-content verification, or ask support staff to inspect sensitive images over email.
Access & incidents
For security, support or data requests during the invited pilot, use the private operator contact shared during onboarding. Legal contracts and data-processing terms must be agreed before external production use. No enterprise availability or accuracy certification is claimed.