Choose what you share

Browser checks

The homepage’s default checker runs in your browser. It does not send prompt text to the API or research database, store a prompt history, or save drafts in browser storage. Clearing or reloading the page discards that draft.

API and MCP checks

These calls send the prompt and context you select to the Worker for processing. Legacy calls store categories, repair identifiers, word counts, algorithm version, timestamp and a pseudonymous case identifier. Version 2.0 defaults research off and also supports privacy: {research: false}. Enabled research stores bounded action, coverage, inventory and question counts, without resource identifiers, facts or excerpts. It does not store your prompt or answer. Use research: false to opt out. The IP address is used transiently for rate limiting, not saved in the research tables.

Outcome reports and optional examples

Outcome reporting is optional. Reports include your stated criterion, evidence source, public model identifier if supplied, paired results and any measured token or latency counts. Do not put private deployment names or secrets in metadata.

Donations require separate explicit human consent for each redacted example and confirmation that identifying information was removed. They are stored privately in quarantine. Automated checks are limited. Donation does not authorize public quotation or model training.

Retention and withdrawal

The weekly retention job removes donated text older than 90 days and individual cases and outcomes older than 180 days. Because cleanup runs weekly, deletion can occur up to seven days after those thresholds. Use the case ID and version 2 withdrawal token to delete a case, its report history and donation earlier; legacy cases use their reporting token. Minimal pseudonymous deletion markers are retained for backup recovery. Keep these tokens private; without it we cannot associate a pseudonymous case with you. Previously generated aggregate snapshots remain.

Operational logs

Application request logs contain generated request identifiers, timing, HTTP status, API/MCP surface and result category. They exclude prompt bodies, responses, request headers and URLs. Automatic Worker invocation logs are disabled in configuration. The hosting service still processes ordinary connection information under its own policies. There are no advertising or third-party analytics scripts.

Optional AI reviewer

If configured, the separate AI-review button sends your prompt and context to the server and Cloudflare Workers AI using Llama 3.3 70B Instruct FP8 Fast. This does not enable research or donate examples. Provider processing and retention follow its terms; this site does not promise zero provider retention. Avoid private text in URLs, which can remain in browser history or infrastructure logs.