01
It does the work, and writes the document
Research memos, assessments, summaries, chronologies, comparisons, proofreads, drafts and memos to the file, in Word or text, with spreadsheets where the work has rows.
Pick a task, say what you need and drop in the matter's files: research a question, assess a position, summarize a record, build a chronology, compare two versions, proofread a filing, draft a document or check the citations in a draft you already wrote. It comes back in Word, text or a spreadsheet, and a lawyer reviews, verifies and signs it.
Our system produces better results because it does not take its own work on trust. The cases it cites are looked up in the published case law and checked by a separate stage whose only job is checking. Anything flagged goes back to be fixed and is checked again, and whatever is still open is marked for your lawyer.
The trade, stated first
HiCourt does the checking before the draft reaches your lawyer, not after. That takes longer than a single pass of AI, and the lawyer still assesses the work.
Checked on your side
Your hours
Whatever a tool did before handing the draft over, the verification that counts is the one the signing lawyer does: run down each authority, read the pin cites, work out which sentences are load-bearing and which are decoration. The question is how much of that is still waiting when the draft lands.
Checked before
HiCourt's hours
Before the draft reaches you, HiCourt looks up the cases it cites, has a separate stage test them against the claims they support, and sends anything flagged back to be fixed. What is still open is marked, so your review starts where the work needs it.
The question for a pilot is how much useful, supported work reaches the lawyer and how much correction remains. That benefit has to be measured on finished work.
If turnaround is the priority on a piece of work, use something faster. We would tell you that. HiCourt is for the matter where you would rather be right.
Bring this to every demo, including ours
Your firm is not choosing between HiCourt and nothing. It is sitting through demos of serious products built by large teams, and most of them will do a great deal that is genuinely useful. So we are not going to characterise them for you. Here is the checklist we would want a firm to use on this whole category — the left column is a question to put to any vendor, and the right column is what HiCourt answers. Ask us the same ones.
The input
Can I hand it the matter, or do I have to break the matter into questions myself?
Hand it the matter. Pick the task, say what you need in your own words and drop in the matter's files; each file is read and indexed before a run uses it.
Your own material
Does it work from our files and our past work, or only from what I type?
From both. Files you add are read and indexed, and for research and analysis the firm's past work is recalled, labelled as reviewed or not and as clean or with known problems.
Unfinished work
If a run cannot finish, do I get a half-done document dressed up as finished?
No. A run that stops keeps what it has, labelled incomplete in every file and on the page, and says why. You can ask it again with the same inputs.
Uncertainty
When the tool is not sure, do I ever find out?
Yes, at the sentence it affects. Uncertain and missing results stay visible, and each claim in the finished document carries a status, including “not checked”. Nothing uncertain is smoothed over.
Who checks
Is the thing that checks the work the same thing that wrote it?
No. A separate checking stage, which wrote none of the work, reads the opinion and tests each case against the claim it is cited for. Quotations, pinpoints and whether a case is still good law are not checked yet.
A defect channel
When something comes back wrong, does it get fixed, or does it get a label?
It gets fixed. A flagged citation goes back to be corrected or withdrawn, and the fix is checked again. A check of a draft you pasted in reports on it and never rewrites it.
When it stops
What ends that loop, and does the output tell me which ending it hit?
The loop is bounded, and the run records why it stopped. Anything still flagged stays flagged for your lawyer, and a final finding never triggers a silent rewrite.
What gets checked last
Does verification run on the draft, or on the document I actually receive?
On the document you receive. Its cited claims go back through the checks, and a separate check asks whether it does what you asked. Findings are reported to your lawyer, never used to rewrite it quietly.
Provenance
Can I map a sentence in the output back to the authority it rests on?
Largely. Flags sit against the sentences they affect, and each claim's status is recorded beside the document. A full sentence-to-source map for every claim is still to come.
What a green tick means
What settled this citation, and as of when?
The record says whether the case was found in the published case law and what the check found against the claim. A run date is not a good-law date: whether a case is still good law is not checked yet.
Screens and placement
Does the search respect our ethical walls, and where does our material rest?
Yes. Ethical walls and matter membership are applied before anything is searched, so a run draws only on what the person asking may see. Each firm runs in its own dedicated hosted environment.
Where something is not built yet, our answer says so. Ask every vendor to do the same.
The whole thing, on one picture
The rest of this deck takes this picture one piece at a time. If you look at one slide in this deck, look at this one.
What it does
01
Research memos, assessments, summaries, chronologies, comparisons, proofreads, drafts and memos to the file, in Word or text, with spreadsheets where the work has rows.
02
The cases the work relies on are looked up in the published case law, then a separate checking stage reads each opinion and tests it against the claim it is cited for. Uncertain results stay visible. Quotations, pinpoints and whether a case is still good law are not checked yet.
03
The document comes back with the status of each authority and each claim, and what is still open, carried into the files you download.
Why the second step exists. AI writes citations that look perfect for cases that were never decided, and it will happily tell you an invented case is real if you ask it. So HiCourt doesn't ask. It looks the case up.
Reason one
The cases the work cites are looked up in the published case law. A case that cannot be found is flagged, not assumed. Then a separate checking stage, which wrote none of the work, reads the opinion itself and tests it against the claim it is cited for.
A found case and a clean check tell your lawyer where to look first. They do not establish that the law is right; that judgement stays with your lawyer.
Anything flagged goes back to be corrected or withdrawn, and the fix is checked again. The loop is bounded, the run records why it stopped, and anything still flagged stays flagged for your lawyer.
Checking a draft you pasted in is different: it reports on your draft citation by citation and never rewrites it.
Checks cover case citations. Statutes, regulations, rules and treatises are not looked up, and quotations, pinpoints and whether a case is still good law are not checked yet. Your lawyer checks those where they matter.
And a lawyer at your firm still reads it and signs it. HiCourt tells your lawyer where to look and what it found; the judgement about what the law means, and the signature on the filing, stay with the person whose name is on it. We built it that way on purpose. That duty doesn't move to a vendor, and you wouldn't want it to.
Where all that time goes
Step 1 · Your request
Pick the task and add the matter's files. Each file is read and indexed, permissions are applied first, and for research and analysis the firm's permitted past work is recalled.
Step 2 · Drafting
Drafted from a close reading of your request, your files and the permitted past work, and written to find what hurts your position as well as what helps it.
Step 3 · Citation check first check
The cases the work relies on are looked up in the published case law, and a separate checking stage tests each one against the claim it is cited for. Uncertain and missing results stay visible.
Step 4 · Write-up
Everything that survived the checks goes into the document you asked for, and every flag still open goes with it, against the sentence it affects rather than buried at the end.
Step 5 · Final check final checks
The finished document's cited claims go back through the checks, and a separate check asks whether it does what you asked. Each claim gets a status, including “not checked”.
Step 6 · Delivery
In Word or text, with spreadsheets where the work has rows, and the checks recorded against it. A run that cannot finish keeps what it has, labelled incomplete.
Your lawyer reviews it, checks it and signs it. HiCourt files and sends nothing.Step 7 · Memory
Every finished run is filed automatically, and a lawyer's signed review makes a new version. Later work draws on it only where the person asking may see it.
The library gets more useful with each matter.Every step runs before your lawyer sees the work. What it could not settle is marked, not hidden.
What comes back
Sample output · a memo on whether a liability cap survives an allegation of gross negligence
Flagged
The case could not be found, the check found it does not support the claim, or the result was uncertain. Each flag says which.
Sent back & cleared
It was flagged, went back to be fixed, and the recheck found nothing against it. The record keeps what changed. A clear recheck is not a legal opinion.
Still open
What the system could not settle, with the reason, for your lawyer to decide. Quotations and whether a case is still good law are not checked yet.
Your lawyer's review starts with what is still open, instead of being spread evenly across everything the document cites.
Reason two
Most tools stop at the draft. HiCourt checks the cases the draft relies on, sends anything flagged back to be fixed, and checks the finished document again before your lawyer sees it.
Our system produces better results because the checking is built into the work, not left to the end.
Checked by a stage that did not write it
The checking stage wrote none of the work. Its only job is to test what the work cites.
Fixed, not just labelled
A flagged citation goes back to be corrected or withdrawn, and the fix is checked again.
All of it recorded
The run records what it did, what each check found and what changed on every repair.
A tool that writes once and hands it over leaves the checking to you. HiCourt does the checking first, and shows you what it found.
Reason three
Files you add are read and indexed, and for research and analysis the firm's permitted past work is recalled and labelled. Every finished run is filed back into the firm's memory, so the library gets more useful with each matter.
Today each firm runs as Option A, in its own dedicated hosted environment. Options B and C are not offered yet.
Option A
Option B
Option C
Nothing from your library is shared with another firm, pooled with anyone else's work, or used to improve anything outside your office.
Ask every vendor this one
A shared index is a confidentiality question inside your own firm before it is anything else. A retrieval index does not respect an ethical wall by nature: ask an ordinary question and it returns whatever is semantically closest, including from a matter the person asking is screened off from. They've done nothing wrong, which is exactly why the control has to sit in the software.
HiCourt applies ethical walls and matter membership before anything is searched. A run draws only on what the person asking may see, and what they may not see is not searched, recalled, shown or used in their work.
Deleting a file everywhere it has been copied, backups included, is not finished yet. Until it is, we will not tell you it is.
Each run keeps a protected record of what it did and what it found, readable only by people who are still allowed to see that matter.
Your call, and mine
Each firm runs in its own dedicated hosted environment. Keeping the library on a machine in your office or on each person's computer is not offered yet.
Yours to decide
Already decided
I'd rather show you this trade than hide it. Nobody can give your firm the most capable AI and nothing leaving your premises, and anyone who tells you otherwise hasn't thought about it. If your firm genuinely needs the second, that's a different conversation about buying hardware — and I'd rather have it than sell you this and have you discover the difference afterwards.
By design
These aren't limitations we ran into. They're the shape of the product, and the reason a firm can put it in front of a matter at all.
Cost
Every check, repair and final pass is work the system does before your lawyer sees the draft, so a checked run costs more than a single pass of AI. Each run records what it did and why it stopped.
How your firm gets billed is a conversation rather than a price list. A flat annual fee, a charge per lawyer, a charge per matter, or usage-based. Each suits a different size and shape of firm, and I'd rather fit it to how your firm actually buys things than hand you a number and defend it.
Each run records what it used. Billing a firm automatically from that record is not built yet, and a cost shown on screen is an estimate, not a bill.
Tell me how your firm would want to be billed, because that genuinely decides it.
What I'm asking
That's the whole question. You've seen what the extra time buys. What I want from you is the part I can't get from anywhere else: whether this fits how your firm actually works, and what you'd change.