Your clients’ data doesn’t belong in ChatGPT.
Now it doesn’t have to.

Yes, completely. Stript runs multiple AI models directly on your hardware, not through a cloud API. You can verify this yourself: disconnect from the internet after the initial model download, and Stript works exactly the same. No document content ever touches a server. The only network requests are the one-time model download, a lightweight update check on launch (plus the update download itself, only if you choose to install one), the one-time email verification that activates the Free Evaluation, a periodic license check at most every 30 days, and (only if you explicitly enable them in the settings) optional crash reports. None of them ever contains document content.

Cloud-based tools, even those hosted in Germany, still send your document to a server to process it. With Stript, the document content never leaves your device: detection and anonymization run entirely on your own hardware, and you can verify it in your network traffic. That is the core difference, on-device instead of hosted.

Stript collects no document content and no usage telemetry. Activating the Free Evaluation uses your email address once for a verification link, and we keep a keyed hash rather than the address itself, as described in our Privacy Policy. Optional crash reports (off by default) send only technical error data, never document content and nothing that identifies you. Newsletter subscriptions are separate (double opt-in, unsubscribe anytime), and anonymization mappings stay encrypted on your local machine.

Across our German and English benchmark corpora of legal, financial and HR documents, our three-stage pipeline reaches an F1 of 95.8 (German) and 97.3 (English) with 95–98% recall. Checksum-validated identifiers like IBANs, credit cards, and bank accounts are caught at 100% recall, and structured data like emails is detected with 99%+ precision. You can review and adjust every detection before anonymization. Confirmed detections and everything above your confidence threshold are replaced.

Stript supports PDF, Word (DOCX), email (.eml and Outlook .msg, including supported attachments), CSV, RTF, ODT, XLSX, scanned or image-only PDFs (read on your device via OCR), and plain text. Legacy .doc and .xls files must first be saved as .docx / .xlsx.

Absolutely. Stript is designed for professional use in regulated industries, and everything runs on machines your IT department already controls. Docker self-hosting, team management, and SSO/SAML arrive with Enterprise, launching during Q3 2026.

Stript is built for GDPR-sensitive workflows. Document processing happens locally on your device. Your document content is not sent to Stript or to third-party AI providers by the app. Stript supports reversible pseudonymization and anonymization workflows. You remain responsible for reviewing the output and deciding whether it is anonymous enough for your use case. Website, purchase, licensing, and optional crash-report data are handled separately as described in our Privacy Policy.

Stript runs on macOS (Apple Silicon, M1 or newer), Windows 10/11 (64-bit), and Linux (64-bit x86_64, as an AppImage or .deb package). Minimum 8 GB RAM, 16 GB or more recommended. You need between 4 and 33 GB of free disk space for the local AI models depending on your hardware tier. They download once on first launch and then run fully offline. Stript adapts to your hardware. On 8 GB it runs a lighter mode with smaller models. This remains fully functional, but detection is slower and a little less accurate. With 16 GB or more it loads larger models and runs noticeably faster and more accurately. An optional GPU (Apple-Silicon GPU or NVIDIA CUDA) speeds detection up further. Either way, everything stays local on your machine.

That depends on the document and your hardware. Larger documents and scanned PDFs take longer to analyze than short letters or emails, especially on machines with 8 GB of RAM. Stript shows its progress while it works, and you can review the first detections while the deeper analysis is still running.

The Free Evaluation includes 5 successfully processed documents in total for one verified email identity on one active device, with no monthly reset. The bundled sample documents don’t consume the allowance, and neither do failed or cancelled attempts. Deleting a document doesn’t return it to the allowance, and documents you’ve already processed stay available.

Because the work doesn’t stop at the download. The detection models keep improving, the app receives regular updates, and the license and update services run continuously. A one-time price couldn’t fund that ongoing work. You can cancel anytime.

Purchases are processed by our payment provider acting as merchant of record. You can cancel a subscription at any time from the billing portal. It then runs until the end of the paid period. Details on withdrawal and refunds are set out in our Terms of Service. Read the Terms of Service →

Lemon Squeezy is our payment provider and acts as merchant of record. It handles the checkout, issues a VAT-compliant invoice, and remits the VAT for you. That is why its name appears on the invoice instead of ours. Your document data has nothing to do with this. Only the purchase itself runs through Lemon Squeezy, and you manage or cancel your subscription in its billing portal.

Email us at [email protected]. We usually respond within two business days, and requests from Pro users are handled with priority.