AI legal assistant · Distinction Labs
You’re the lawyer.Stet remembers everything.
An AI legal assistant that remembers every matter, shows the source behind every citation and drafts the everyday work.
Request early accessPrivate preview
Morning brief · Tue 27 Oct
Three things need you today.
- Thursday 9:30. The hearing on the motion to dismiss. Your binder is ready.
- 11:42 last night. Doe’s opposition came in. It quotes Conley’s “no set of facts” test, retired in 2007 .
- Friday. You promised the client an update. A draft is waiting for you.
What standard will the court apply on Thursday?
Searched case law · 2 sources
Plausibility. The complaint needs enough facts, taken as true, to make the claim plausible on its face . Labels, conclusions and a formulaic list of the elements won’t do .
Tap a number to see its passage
What’s due this week?
From your calendar and notes
Three things this week:
- Thu 29 Oct, 9:30. The hearing on your motion to dismiss. The binder is ready. Calendar
- Fri 30 Oct. The client update you promised. A draft is waiting. Your promise
- Fri 30 Oct. The dispatcher’s deposition. Your brake-log question is first in the outline. Your note · Tue 20 Oct
Draft the client update.
Drafted · 2 sources
Here’s a draft. I left one point out: I found no source for when the court will rule.
Draft · to the client · not sent
Thursday at 9:30 the court hears our motion to dismiss. Doe’s opposition relies on a pleading test the Supreme Court retired in 2007 . The question for the court is whether Doe’s complaint states enough facts, taken as true, to make the claim plausible . We’ll make both points at the hearing.
Left out: when the court will rule. No source found.
I’ll call you after the hearing.
Prepare for Thursday.
Searched case law · 3 passages
Three points for the hearing:
- Doe’s opposition argues from Conley’s “no set of facts” test, which the Supreme Court retired in 2007 .
- The standard is plausibility: enough facts, taken as true, to make the claim plausible on its face .
- Labels, conclusions and a formulaic list of the elements won’t do .
Documents
Drop in a document.
A deposition, a contract, the other side’s brief: Stet reads it, sums it up and dates every event, with the page it came from.
Summarize this and give me the dates.
Lane deposition.pdf · 214 pages
Read 214 pages · 9 events
Coastline set the drivers’ schedules and disciplined them directly; the warehouse only posted the schedules. The first three events:
- 14 SepCoastline emails the week’s schedule to the warehousep. 41
- 21 SepThe warehouse asks to change two shifts; Coastline refusesp. 88
- 2 OctA Coastline manager disciplines a driver directlyp. 132
Before it goes out
Check a brief before you file it.
Paste a brief or drop the PDF. Every case citation comes back with a verdict and the record behind it, so you can verify each one fast, and the check keeps nothing.
In Stanford’s study, AI tools built for legal research gave false or invented answers 17–33% of the time.1
Real errors Found by Stet’s checker in petitions filed in the Supreme Court
District of Columbia v. Heller, 544 U.S. 570 (2008)
MismatchNo case in the record at 544 U.S. 570. The record spells this name exactly for one case, and holds it at 554 U.S. 570.
In a federal court of appeals opinion reproduced in a Supreme Court petition, and again in a district court order in the same petition.
Jones v. North Carolina Prisoners’ Labor Union, Inc., 433 U.S. 199 (1977)
Not foundNo case in the record at 433 U.S. 199, a volume Stet treats as complete.
In the body of a Supreme Court petition. Its own table of authorities gives the right page, 433 U.S. 119.
Memory
Ask what you did last week.
Every document, call and promise stays with its matter, behind its client’s wall, and Stet finds it when you ask.
What did I ask you about the dispatcher last week?
Whether he had seen the brake log before the crash. His deposition is on Friday, and that question is first in your outline.
From your note · Tue 20 OctWrong answers in the same 163 legal research questions.
3
Stet
123 right · 37 declined · 3 wrong
Where it couldn’t prove an answer, it said so, 37 times.
17
A general-purpose AI model, searching the same court records
145 right · 1 declined · 17 wrong
It answered more of them right: 145 to Stet’s 123.
76
The same model, answering alone
76 right · 11 declined · 76 wrong
Half the answers it gave were wrong.
Our own internal test; not a published benchmark.2
Private by design.
Each client’s material stays behind its wall. Stet reads across your matters only when you ask it to, and never while drafting. Nothing goes to an outside AI company.
Billed to the matter.
Every use is recorded against a client and matter and sent to your billing system. No seats, no minimums.
You make the calls.
Stet asks before every decision that’s yours, as Oregon’s bar advises for chatbots and AI agents.3
Behind every great lawyer,a perfect memory.
Early access opens soon, by invitation, to Stet’s first firms.
Sign inIf your firm has been invited
Made by Distinction Labs with CMYK, our model that shows where each answer came from