Your documents stay yours.
We treat legal documents with the same care as a law firm. Here is how.
Start for free
Legal documents are rarely harmless: a lease, a cease-and-desist letter, a termination notice. That is why we treat them with the same care as a law firm — technically and organisationally. Data and database sit in the EU, transport is secured with TLS, storage is encrypted with AES-256, and every database query is filtered to your account through row-level security.
Just as important is what does not happen: your texts are used only to answer your request, never for model training. Personal data patterns are replaced with placeholders before most AI calls and restored only in the response. And you stay in control: access, correction and deletion are self-service in your account settings at any time.
- The chat stays with your matter
- Re-check without uploading everything again
- You can export everything at any time
Where your data lives
Storage location, transport and separation are the three questions that come before any feature. We answer them here, before you upload your first document.
EU hosting
Data and database in the EU. No transfers to unsafe third countries.
Encryption
TLS in transit and AES-256 at rest for database and file storage.
Isolated documents
Documents live in per-user folders. No one else has access — including us in plaintext.
Who is allowed to see it
Access here is not a setting in the interface but a rule in the database — and it applies to every query, including our own. Which further rights you have, and how to exercise them, is set out in the privacy policy.
Row-level security
Every table is protected with RLS — each request is filtered to your user ID.
GDPR-compliant
Access, correction and deletion anytime — self-service in account settings.
This check never leaves your browser
Before a document is uploaded at all, the platform checks its format and size — locally, on your device. Pick any file: you get the same result you would see later in the product, without the file being transmitted anywhere.
It is the same validation code every upload path on the platform uses — not a simplified demo replica.
Upload check
Format, size and estimated text volume — instantly and locally.
The file stays on your device. It is not uploaded and its content is not read — only the file name, size and format are evaluated.
Nothing is sent: the check runs entirely in the browser, with no server request.
No file selected yet. The supported formats and limits are listed below.
Formats and limits
| Supported formats | PDF · DOCX · JPG · JPEG · PNG · WEBP · TXT · MD · MARKDOWN · XLSX · CSV · EML · MSG |
|---|---|
| Maximum file size | 25 MB |
| Not supported: .doc | Word 97–2003 (.doc) is a binary container the text extraction cannot read. We deliberately reject such files instead of charging for an analysis of unreadable content. Save the document as .docx or PDF instead. |
| Not supported: HEIC/HEIF | Apple's photo format is not read reliably everywhere. In the iPhone camera settings you can switch the format to most compatible, which produces JPG files. |
The check covers format and size. Whether a document can actually be evaluated — a scan with legible text, for instance — only becomes clear during text extraction.
How personal details are recognised before they reach a model
Further down you can see how the platform handles personal details. The rule-based core of this recognition — names, account data, contact and other details — you can try right here: it runs entirely in your browser, without the text being sent anywhere.
This preview deliberately shows only a slice: without an actual case file it lacks the names already known there, and the additional model cross-check on already-masked text does not run here.
Live pattern recognition
Type a text or use the example — detection runs instantly, locally.
The text stays on your device. There is no network call — not even when you click "Insert example text".
Recognised locally: 7
1×address1×date of birth1×email1×IBAN1×name1×phone1×zip/city
Sehr geehrter Herr Dr. [name], vielen Dank für Ihre Nachricht. Bitte kontaktieren Sie mich unter [email] oder telefonisch unter Telefon: [phone], gerne auch am Wochenende. Für die Rückerstattung nutzen Sie bitte diese IBAN: [IBAN]. Geburtsdatum: [date of birth] Meine Anschrift lautet: [address], [zip/city]
This preview uses the platform's rule-based core (names, account data, contact and ID details, among others) — without case-specific data and without the additional model cross-check of the full pipeline. The structural recognition is calibrated for German-language phrasing; it recognises IBANs, email addresses and internationally formatted phone numbers regardless of language. Use it for illustration, not to check real client documents.
Commitments
What we do not do
No AI training
Your texts are used only to answer your request, not for model training.
PII masking
Personal data patterns are replaced with placeholders at the edge and restored in the response. In the consumer area this is the default; in the lawyer portal a firm-level switch governs it (default “standard”), applying to the agents named for it — with documented exceptions where clear names are essential (details: privacy policy §4c, §4g, §4h).
AI model region
AI inference runs EU-resident (Frankfurt) with zero data retention and a contractual no-training clause — no transfer to unsafe third countries without EU Standard Contractual Clauses.
Professional secrecy
Client-matter data is handled in line with BRAO and § 203 StGB (Germany's professional confidentiality statute) — technically and organizationally.
Read the details
The underlying documents
Questions about security or DPA?
We answer technical and privacy questions personally.