Privacy
Privacy, security and retention
APD Checker is operated by Lukshi, LLC. It keeps as little as possible, for as short a time as possible. Last updated November 14, 2026.
What we keep, and for how long
| Information | How long |
|---|---|
| Uploaded files | Encrypted (AES-256-GCM, a separate key per review session) and deleted as soon as text extraction finishes, normally within minutes. |
| Extracted text, findings and reports | Stored in our database so you can view results. Deleted 24 hours after the review session starts, or immediately when you choose “Delete everything now”. |
| Email address for the PDF | Encrypted under the session's key and deleted with the session (24 hours). Used only to send the verification code and the download link. |
| Verification codes and download links | Stored only as one-way hashes. Links expire with the session and stop working when the session is deleted. |
| “Request a review from Lukshi” | Your name, email, organization and message are emailed to Lukshi (cody@lukshi.com) and kept as business correspondence so we can respond. The checker itself keeps only your sharing choice and whether the request was sent, and deletes that with the session. If you choose to share the report, Lukshi gets a link that works only until the session expires. |
| Usage-limit counters | Daily counts keyed to a one-way hash of your network address or email. Deleted after 30 days. |
| Session records | Hashes and timestamps only, no content. Deleted 7 days after the session ends. |
| Application logs | Operational events (for example that a review finished and how long it took). Never document text, file names, findings, email addresses, codes or links. Kept by our hosting provider under its log retention. |
| Backups | We do not run our own database backups. Our hosting provider may keep infrastructure copies under its own terms; data in such a copy is not restored into the service after deletion. |
Review pages open only in the browser that started the review (an access cookie is required), and are marked not to be indexed by search engines. There are no analytics, advertising or third-party scripts on this site. No account is required.
Service providers
With your consent, the text extracted from your files is sent to these providers. Each provider handles data under its own terms and privacy policy, not ours, and we cannot delete what they hold.
- Anthropic (model claude-sonnet-5-5): receives the full extracted text of the files you confirm, to assess the APD and draft recommendations. Used through Anthropic's commercial API. Anthropic states that it does not train models on commercial API inputs or outputs by default; it does keep them for a limited period under its retention policy.
- TypeSafe AI (Jev): receives the file names and the opening pages of each file to identify the documents, and short passages for yes/no checks during the review (for example, whether a passage is template text). TypeSafe states that it does not train its models on customer data. You can opt out on the upload page.
- Resend: sends our emails. It receives your email address and the email itself (a code or a link, never findings).
- Railway: hosts the application, database and logs in the United States.
Zero data retention: we have not arranged zero data retention with any provider. “Not used for training” does not mean “not retained”.
Who can see your review
- Lukshi staff do not look at reviews. We see one only if you send it to us with “Request a review from Lukshi” and choose to share it.
- The operator can technically access the database to run, secure and fix the service. That access is limited to the operator's account and is not used to read reviews.
- We do not sell or share your information, and we do not use it for marketing.
What you should not upload
Do not upload documents that contain information about children, families or caseworkers, confidential case data, security-sensitive system details, or anything your agency has not approved for processing by an external AI service. This service has not been assessed under FedRAMP or any other government authorization program and makes no claim of suitability for restricted government data.
Security controls
- HTTPS only; strict content security policy; no external scripts.
- File type verified by content; macro-enabled and encrypted documents rejected; extraction runs in a resource-limited separate process.
- Document text is treated as untrusted: instructions inside documents are ignored and flagged, and AI output is checked before it is shown.
- Rate limits, daily usage limits and spending limits protect the free service from abuse.
Security issues and incidents
Report a security problem or a privacy concern to cody@lukshi.com. Please don't include document content. If we learn of an incident that affects the service, we will post a notice on this site. We can contact you directly only if you gave us your email address.
See also the terms of use.