NYU School of LawArtificial Intelligence and Corporate Deals and Advice
Learn from every redline.
Turn a reviewer's actual edits and feedback into evidence-based drafting rules you can check and share.
For the selected reviewer
By the numbers
Five steps
How it works
Rules come from what the reviewer actually changed, not from anyone's description of their preferences.
Compare and tag
1.1 Compare. Use your draft and the clean version that came back, one Word file with tracked changes, or your notes when the feedback was only oral. Every change is shown exactly.
1.2 Tag. Record why each change was made: a category, a one-line reason and a scope.
Ledger
Every tagged change in one searchable record: the evidence behind your rules.
Rules
3.1 Learn. Recurring tagged changes become candidate rules, each with its evidence attached.
3.2 Confirm. Accept or reject candidates. Only rules with enough evidence enter the skill.
Check
Test a new draft against the rules before it goes out: fixed text checks, plus an optional AI review of the rest.
Export
Compile the confirmed rules, each with an example, into a plain-text skill for any approved AI tool, or a Word or PDF rules sheet.
New here?
Try it in 2 minutes
Load the fictional demo (Partner A and Partner B) and follow a five-step tour: a redline becomes cards you tag, recurring edits become rules, a new draft is checked, and the rules are exported. Your own data stays as it is.
Optional AI
It suggests.
You decide.
Switch it on in the top bar. It suggests tags, groups reasons that mean the same thing, drafts rule wording and reviews a draft against the rules a pattern cannot check. Every finding has to quote your draft, and the app checks each quote.
Privacy
Local
by default.
Everything runs in this browser and is stored only here. AI stays off until you switch it on; then only the text an AI feature needs goes to OpenRouter.
Next step
Your reviewer's edits
are the rulebook.
Paste two versions of a draft to begin. Edits that recur become rules you confirm, check and export.
Step 1.1 · Compare
Compare drafts
Choose how the feedback came back. The comparison starts by itself as soon as both versions are in. Three ways in: your draft plus the reviewer's clean version (paste, .txt or .docx); one Word file with tracked changes, from which the app rebuilds both versions; or, when the reviewer only talked you through it, your notes of the oral feedback, one point per line. Text is compared paragraph by paragraph and word by word; spacing and blank lines are ignored. Each feedback point becomes a card you tag like an edit.
Step 1.2 · Tag
Tag the changes
Pick a category, write a one-line reason and press Enter. The next change opens by itself. Choose a category and write a one-line reason in your own words. Use "This matter only" for edits driven by the facts of one deal; those never count toward a rule. "Judgment call/one-off" is also excluded from rule counting.
Saved comparisons
Reopen a comparison to continue tagging. Showing the selected reviewer.
Step 2 · Ledger
The ledger
A searchable record of every tagged change across all reviewers. Select a row to open it. The ledger is the evidence base. Candidate rules are mined from rows whose scope is "Always". Search looks at reasons, changed text, matters and categories.
| Date | Profile | Type | Category | Change | Reason | Scope | Matter |
|---|
Step 3.1 · Learn
Your rules
Candidates appear when the same kind of change recurs for this reviewer. Edit the wording, then confirm or reject. Changes are grouped by reviewer and category, then by similar reasons (shared key words; with AI on, also reasons the AI found to mean the same thing). Changes tagged "This matter only" or "Judgment call/one-off" are ignored. A rule can only be confirmed once it has at least the minimum number of instances.
Candidate rules
Rejected candidates
Step 3.2 · Confirm
Confirmed rules
Each rule keeps its evidence. A rule is flagged "May be outdated" when newer evidence contradicts it.
Step 4 · Check
Check a draft
Paste a new draft. It is checked right away against the confirmed rules for this reviewer. Auto-checkable rules are tested with fixed text patterns, not AI, so read each result with care. Human-checked rules become a checklist; with AI on, you can also ask the AI to review them, and every quote it gives is checked against your draft. Only confirmed rules that meet the minimum instances are used.
The report
Step 5 · Export
Export the skill
A portable plain-text skill built from the confirmed rules for this reviewer. Paste it into any approved AI tool. The version number goes up each time you copy or download a skill whose rules have changed. Rules below the minimum number of instances are left out.
Back up your data
Everything is stored in this browser only. Download a JSON backup to move it or keep it safe.
Clear all data