Features · Comments & Copilot

The document you were sent, with an associate sitting next to it.

Drafts arrive from an AI, from co-counsel, from the client. JR3 opens beside them in Microsoft Word — reading the comments already in the file, answering the ones addressed to you, and turning the notes you leave into tracked changes.

Try Today →Schedule a Demo
ISO 27001 CertifiedSOC 2 Type IIZero Model Training
Microsoft Word — a patent application and a mutual NDA, start to finish
JR3 working alongside a patent application and a mutual NDA in Microsoft Word
The workflow

Six steps, none of which ask you to leave the document.

01
The draft arrives

Generated by an AI or attached to a client email. Open the .docx and JR3 is already on the right rail, with the document's sections listed.

02
Read what was flagged

The drafter's comments sit in Word's margin where they always have. Select the passage one points at and run Explain on the selection.

03
Answer in the file

Add comment posts the answer as a reply in the thread the drafter started, under your name, where the next reader will look for it.

04
Mark it up as usual

Select text, New Comment, type the note. Nothing about the way you read a draft has to change for JR3 to follow it.

05
Finalize from chat

Type / and call Study Comments and Finalize Document. JR3 reads every comment in the draft against your firm playbooks and proposes each change on its own card.

06
Accept, and it lands

Every accepted card writes its tracked change into the document and resolves the comment it came from. Rejected cards write nothing.

Explain, on the selection

Highlight a passage and the rail switches to skills for that selection — Rewrite, Explain, Cite, Comment, Review. The explanation names the risk in the language the file is written in.

Your comments, Word-native

Comments you leave are ordinary Word comments — visible to anyone who opens the file, with or without the add-in installed. They are also the instructions JR3 works from.

Injectable islands

Every proposed change arrives on its own card, with its reason.

Nothing is written into the document until you say so. Each card shows the tracked change it would make and the comment or playbook rule it came from, with accept and reject on the card itself. Rejected cards collapse and write nothing.

—Accept or reject per card, or Accept All when the set is right
—Locate jumps to the exact place in the draft the card touches
—Applied changes land as Word track changes, attributed and reversible
—The comment that prompted a change is resolved when the change is made
Two cards rejected, the rest accepted — each redline landing in turn.
Worked example · Mutual NDA

A client asks whether two clauses agree.

Nothing about the patent workflow was specific to patents. Here is the same rail on a three-page mutual NDA that arrived by email, with one comment in it from the other side.

01It arrives in Outlook

Sarah Li attaches a redline of the mutual NDA and asks whether Section 7 conflicts with another clause. Open the attachment and the rail is already there.

02Her comment is a Word comment

It came from her copy of Word, not from JR3, and it reads the same on both sides. JR3 treats it as the instruction for what to look at.

03Content Contradictions, run whole-document

A review skill reads the agreement end to end and returns three findings with their locations — the two-year term in §3 against five-year survival obligations in §7, and two more.

04Argue with it before it writes

Discuss in chat opens the finding as a conversation. Shorten survival, or extend the term? Choose the term, switch on an embedded comment explaining the change, and accept.

Finish

A redline, an explanation, and the email back.

Section 3 goes to five years as a tracked change. A comment beneath it says why and cites the playbook rule it followed. Sarah's original comment gets a reply rather than a silent edit — and /email drafts the cover note that carries the file back.

—Every edit is a Word track change, attributed and reversible
—Every answer is a Word comment the other side can read
—Patents, NDAs, licences, opinions — same five steps
See every review skill →
Your standard, not a general one

It searches your playbooks before it proposes anything.

The reasoning is visible while it works, so you can see which of your firm's materials the draft is being measured against. Playbooks, workflows, skills, sequences and rules travel with the document type — and stay yours.

How the document library works →
Thinking
Searching company playbooks…
Reading 5 comments and the claim set…
Firm playbook — Patent Prosecution: pair every method claim with a computer-readable-medium claim.
Firm playbook — Mutual NDAs: extend the term rather than shorten survival.
Any matter

Same rail. Different document.

The workflow does not change with the filing. What changes is the library behind it — the playbook, the rules and the review monitors that belong to that document type.

Patent applications
Specifications, claim sets, antecedent basis, support for every new claim.
Trademark filings
Refusal responses, identifications, specimens, TTAB briefs.
NDAs
Term against survival, forum against governing law, marking requirements.
Licensing agreements
Grant scope, field limits, royalty mechanics, defined terms that must agree.
Opinion letters
Freedom to operate, patentability, clearance — assumptions stated and held.
Complaints & briefs
Counts, elements and the record cites that have to support them.
Runs where you work
Microsoft Word add-in and Google Docs. No export, no second window.
Nothing trains on you
Documents and prompts never train shared models. Your library stays yours.
Model-agnostic
OpenAI, Anthropic, Gemini or privately hosted — the standard travels with you.
Built for procurement
ISO 27001, SOC 2 Type II, tenant isolation, private hosting available.

Read, write, and review like an expert in any field.

Send us a draft with comments in it — a patent application, an NDA, whatever is open on your desk — and we will run this workflow on your matter.