Caveat is an analysis service, not a tracker. We exist to help readers see manipulation patterns in the news they already read. This page explains, in concrete terms, what we do and do not keep about you and the articles you analyze.

What we don't store

We do not store the full text content of articles you analyze. When you use Caveat, your browser fetches the article directly from the publisher. The extracted text is sent to our analysis service over HTTPS, the analysis runs, and the article text is discarded once the response is returned.

We do not maintain a copy of the article on our servers, in S3, or in any log. We do not forward the article text to any third party. We retain only the analysis output and a short excerpt described below.

What we do store

About your account (in Amazon Cognito):

• Your email address
• Your invite code association
• Your running per-day usage count (a number, no URLs)

About each analysis (in our analyses database):

• The URL of the article
• The article title and publication name
• The analysis output (patterns, scores, loaded language, and claims)
• A 250-character excerpt of the article, retained for fair-use citation
• The timestamp the analysis was run

What we do NOT store: any field linking an analysis to the account that ran it. Your account holds counts; the analysis database holds analyses. Even with full administrative access to our systems, we cannot answer the question "which articles has account X analyzed."

We also do not show you a list of articles you have analyzed. This is intentional. Showing such a list — even one stored only on your own device — creates a real privacy risk if someone else uses your browser (a family member, a roommate, a public workstation). Each analysis appears once, in the side panel at the moment you run it. After that, the report exists in our database keyed by a random ID, but neither we nor your device retains a record connecting it to you.

How analysis works

Your browser fetches the article from the publisher's site, just as it would for any other reader. Caveat runs the publisher's page through a readability extractor inside your browser, then sends the extracted text to our analysis service. Our service runs the analysis against a large language model, returns a structured report, and persists only the metadata enumerated above.

We never crawl, scrape, or fetch articles server-side on your behalf. The paywall, the consent banner, and the publisher's analytics all see your browser, not ours.

Third parties

Our infrastructure runs on Amazon Web Services in the us-west-2 region (Oregon). When you run an analysis, the extracted article text is sent over HTTPS to Anthropic, our AI provider, for processing under their Commercial Terms; our service does not retain the text after the analysis completes.

Anthropic's Commercial Terms (Section B, current as of this writing) state that Anthropic does not train its models on Customer Content sent through the API. We have no control over Anthropic's policy and will update this notice if it changes.

We do not use third-party web analytics, advertising pixels, marketing-automation tags, or session-replay tools on this site.

Cookies and local storage

We use localStorage in your browser to keep you signed in (your refresh token and your account ID). We do not set tracking cookies. We do not share any browser storage with third parties.

Account deletion

You can request deletion of your account from the account settings page, or by emailing privacy@getcaveat.fyi. We honor account deletion requests within 30 days. Deletion removes your account record, your history, and your invite-code association. Aggregate, de-identified usage counts may be retained for operational reasons.

Beta data

Caveat is in invite-only beta. Data collected during the beta may be deleted in bulk at the end of the beta period. We will give registered users at least 14 days of notice before any such deletion, by email and on this page.

Health-adjacent content

Caveat is not designed to collect or process consumer health data as defined by the Washington My Health My Data Act (RCW 19.373). If you analyze an article on a health, mental-health, or medical-services topic, the resulting analysis is stored under your account on the same terms as any other analysis and is deleted when you delete your account.

Contact

Privacy questions, deletion requests, and disclosures go to privacy@getcaveat.fyi.