Launched this week
Under Seal: Legal Doc Reader on iphone
On-device AI for legal documents. No cloud, No Account
10 followers
On-device AI for legal documents. No cloud, No Account
10 followers
Under Seal pulls every date and deadline out of a legal document on your iPhone or iPad by pattern, not by asking a model. Redlines drafts, searches your whole library, and never uploads a page.
Hi all, hope you are doing well. I'm the maker.
I built Under Seal after watching a model read a passport and put the holder's mother's name in the holder's row. It was fluent, it was formatted correctly, and it was wrong. I'd never have caught it if I hadn't happened to know the family.
That's the whole design. In a legal file, a wrong answer that reads well is worse than no answer at all. So the parts that matter most use no model:
• Dates and periods : every one the document states, pulled out by pattern. A regex cannot drop a deadline silently and cannot invent one.
• Redlines : two drafts compared section by section from the words themselves. A summarising model reports what it noticed; a diff reports what changed.
• Clause search : 12 saved searches (indemnity, termination, governing law, restrictive covenants…) run across your whole library. Retrieval only.
• Passports the machine-readable lines parsed to ICAO 9303, with check digits.
The model does exactly one job: answering a question in plain English from passages retrieved out of your own documents, with the citation attached so you can check it. If the file doesn't address your question, it is allowed to say so rather than guess.
All of it runs on the device Apple Intelligence, or Llama 3.2 3B on older hardware. No account, no login, no uploads, works in airplane mode. A 10,000-page library indexes into about 35 MB.
On the price: $14.99/mo or $99.99/yr. That's high for an App Store app and I'd rather say why than let it sit there. It's built for people billing by the hour on files where a missed response window decides the matter not for reading a lease. The 7-day trial holds nothing back: the whole app, your own documents, at their real length. If a week on real material doesn't make the case, it isn't for you, and I'd rather you found that out for free.
It does not give legal advice and it is not a substitute for your own judgement. The citations exist so you can verify every answer, and you should.
I'd genuinely like to hear from anyone who reads long files for a living especially about what it fails to extract. That's the feedback I need most.