Privacy policy

Data we process

For a link, the server downloads the public page and extracts its main text. For a screenshot, the image blob and its technical metadata are held only while Tesseract OCR runs; the original image is not stored and is discarded when OCR ends, whether it succeeds or fails.

We persist the analyzed extract (up to 2,000 words), minimal claim context, claims, verdicts, evidence fragments, source links, timestamps and technical audit records. Reports created from links after public listing was enabled are automatically listed and indexed when they finish conclusively with at least 60% evidence coverage and a usable title. Older reports and screenshot reports remain accessible only to people who have their URL. The full analyzed extract is never public.

A signed anonymous browser identifier supports a one-year cookie. The server stores a pseudonymous derivative for daily limits. A date-bound pseudonymous network signal also limits abuse; it is not used to identify a person or follow them across sites.

Purposes and legal basis

We process the submitted material to provide the report you request, preserve an auditable snapshot, protect free capacity and prevent abuse. Depending on applicable law, this is necessary to provide the requested service and supports the operator’s legitimate interests in security, reliability and accountability.

Do not submit confidential material or personal data that you are not entitled to process. The service does not proactively detect sensitive data before processing.

Providers and transfers

The server uses Vercel for hosting, durable workflows and AI Gateway; Neon for Postgres storage; Exa for evidence search when needed; and the selected free model providers routed by AI Gateway. Fetching a link also contacts that source website. These providers may process technical request data under their own terms and in other countries.

The browser sends the source to our server; model and search requests are made server-side. We do not sell personal data.

Data retention

Screenshot blobs and their original metadata are discarded at the end of OCR. They are never part of the report record.

The analyzed extract, report, claims, evidence records, model audit data and the pseudonymous usage records have no automatic expiry while the service remains operational. Reports are immutable snapshots and their URLs are intended to remain stable. An eligible link report remains indexed after it leaves the recent-reports section unless the operator removes it from public listings.

The browser cookie expires after one year. The network signal changes every calendar day, although operational and audit records containing pseudonymous derivatives may be retained with the related investigation. The operator can withdraw a report for legal, privacy or abuse reasons without exposing an administrative control publicly.

Automated analysis and limitations

AI systems identify claims, search for evidence and propose verdicts under deterministic evidence rules. They can misunderstand context, miss sources or produce errors. Every report is dated, shows uncertainty and is not a declaration of absolute truth.

The service does not replace medical, legal, financial or other professional advice and does not make decisions about a person with legal or similarly significant effects.

Your choices and contact

Depending on your jurisdiction, you may have rights to access, correct, object to, restrict or request deletion of personal data. There is no self-service deletion in this MVP. Use the removal link on an indexed report or send its URL to news-verifier.securely376@silomails.com. An approved listing removal is performed manually and leaves the report accessible to people who retain its URL; complete withdrawal remains available for legal, privacy or abuse cases. Do not include additional sensitive data.

The operator is Julian David. This policy may change as the experimental service evolves; material changes will be reflected here with a new effective date.