Legal document redaction API for court decisions, contracts and briefs
14-day money-back guarantee · Cancel monthly · Invoice with VAT
Court decisions, briefs, contracts and case files name parties, witnesses, counsel, addresses and case numbers. ShinrAI detects these details and replaces them with realistic stand-ins, labels or masks before a document is published or sent to an AI model. ShinrAI 1.6 has been the managed API model since 6 October 2026; its training adds court judgments and EU legal text.
What ShinrAI detects in legal text
The model detects 21 categories of personal data in 15 locales. For judgments, briefs and contracts these matter most:
- People. Parties, witnesses, representatives and other participants, replaced consistently across the whole document.
- Places and organisations. Cities, street addresses, postal codes, companies and other organisations.
- Dates and ages. Detected by ShinrAI 1.6; ShinrAI 1.3 has no date or age category. Dates of birth stay a separate category.
- Identifiers. Customer and case IDs, national identifiers, bank accounts, payment cards and vehicle plates.
- Contact details. Email addresses, phone numbers, web and network addresses.
On request, ShinrAI returns legal references, years and amounts as annotations. Protection never changes them, so the legal reasoning stays readable. The full list is on the languages and data categories page.
Redact a court decision through the API
- Choose a policy. The preset
pseudonymizewrites realistic, restorable stand-ins.labelwrites numbered labels such as [PERSON_1], andmaskmasks characters. Rules per type replace, remove or keep single categories. The actionpartial, for example, keeps only the year of a date. - Set your own values. Tell ShinrAI never to report values such as the name of your court or firm. Give values of your own, such as internal matter numbers, a type.
- Send the document. Upload a PDF or DOCX file of up to 10 MB to
POST /v2/uploadsand start a job of kinddocument. The finished job returns the redacted PDF, the protected text and the entities, and the mapping on request. - Review the result. Have a person review every document before publication. The linkage-risk option estimates how strongly a text still singles out a person. It is a heuristic, not a count.
- Work with AI. The pseudonymised text goes to ChatGPT, Claude, Gemini or your own model. Your application restores the original values locally. Restoration uses no records.
The document guide and the native PII API v2 guide show every call with curl.
ShinrAI 1.6 on court rulings, measured and published
ShinrAI 1.6 detects dates and ages, which ShinrAI 1.3 does not. Its training data includes court judgments, sentences from EU law and generated legal documents in three languages. Innovius measured the release against ShinrAI 1.3 on 127 judgments of the European Court of Human Rights in English, about 5,000 characters each.
| What is scored | ShinrAI 1.6 | ShinrAI 1.3 |
|---|---|---|
| Person names | 80.7 out of 100 | 9,4 |
| Courts, institutions and other organisations | 70.9 out of 100 | 43,9 |
| Dates | 90.4 out of 100 | No date category |
The score runs from 0 to 100. It drops with every missed detail and with every false finding, and a finding counts only with exact boundaries. Specialists call it the F1 score. ShinrAI 1.3 leaves «Mr» and «Ms» outside a name, while this test counts them as part of it; that explains most of its low person score. On German sentences from EU law (MAPA, EUR-Lex), ShinrAI 1.6 reads 39.4 for person names and 57.5 for organisations (ShinrAI 1.3: 23.0 and 53.0), and 77.6 for dates.
The limits are published too. Names of courts and institutions follow other conventions in these judgments and score lower than person names. Sentences from EU law stay hard for every measured system, and on French EU law person names read 41.6, below ShinrAI 1.3 at 44.8. All numbers, methods and raw run files are on the benchmark board. Pilot on your own documents before you decide.
Long briefs in one pass
The model reads 1,024 tokens at once and slides over longer text with overlapping windows. A 30-page file processes as one document, and the same person gets the same stand-in throughout. A session keeps one mapping across many requests, by default for 24 hours.
Use in law firms, courts and legal departments
- Publish decisions. ShinrAI proposes the redactions and returns the redacted PDF. The final review and release stay with you.
- Draft with an AI assistant. Case files and correspondence reach the assistant pseudonymised. The answer comes back to your application with the real names.
- Review contracts. Parties, signatories, addresses and bank details are replaced before a model checks clauses and deadlines.
- Search case law. Judgment collections are protected before indexing, so names stay out of search indexes and embeddings.
ShinrAI is a technical tool, not legal advice. Whether a protected document is anonymous in the legal sense and may be published is your decision under the rules that apply to you.
Hosted in Germany or in your data centre
The managed API runs on STACKIT in Germany and is operated by Innovius UG from Teltow. Synchronous requests are processed in memory, without a content archive and without model training. Asynchronous job content is encrypted and expires 24 hours after submission. For isolated environments, offline packs with signed activation keep all processing on your own hardware. A data processing agreement under GDPR Art. 28 is published.
Test with your own documents
Starter costs €39 per month for 50,000 records and includes standard and Batch processing. A record is 1,000 Unicode code points; Batch processing counts half. Larger workloads move to Team, Business or Platform, or to a non-expiring pack.
Frequently asked questions
Can ShinrAI redact court decisions for publication?
ShinrAI detects people, places, organisations, dates and identifiers and replaces, labels or masks them. For PDF and DOCX files, a document job returns the redacted PDF. Review before publication and the legal assessment stay with you; ShinrAI is not legal advice.
Does ShinrAI keep statute references and amounts?
Yes. On request, ShinrAI returns legal references, years and amounts as annotations, and protection never changes them. Rules per type also decide which categories are replaced, removed or kept.
Which languages does it support for legal documents?
The model covers 15 locales, including English, German, French, Spanish and Italian; Hebrew is beta. The legal training data of ShinrAI 1.6 covers court judgments, EU legal text and generated legal documents in three languages. Published legal measurements cover English judgments and EU legal text in English, French and German.
Where does ShinrAI process my documents?
The managed API runs on STACKIT in Germany. Synchronous requests are processed in memory without a request-content archive or model training. Job data and API sessions stay only as long as their job or session needs them, by default at most 24 hours. You can also run ShinrAI on your own infrastructure with signed offline activation. The page Data handling and retention lists every retention period.
What does legal document redaction cost?
A record is 1,000 Unicode code points; Batch processing counts half. Starter costs €39 per month for 50,000 records. Non-expiring packs start at 5 million records for €1,590.
Can I get my money back if ShinrAI does not fit?
Not the right fit? Write to support within 14 days of your first plan payment and we refund it in full. Once per company, if you used no more than 10 % of the plan's monthly records.