How is this different from asking a chatbot to review a contract?

A chatbot compares your document against its general impression of contracts. This compares it against the specific agreement your organization would have signed, clause by clause, and classifies each difference against policy you wrote. The output is a decision list, not an essay.

What if we do not have a clause library yet?

Most organizations have one implicitly — in their last twenty agreements and in one person's head. The practical path is to designate a current agreement as the standard and let the deviations that keep escalating tell you which clauses deserve a library entry.

Does it work on agreements signed before we used Paqt?

Yes. Extraction on an executed agreement is the usual first step for a back catalogue, and it is often how teams discover the renewal windows and obligations nobody was tracking.

How do we check a finding we disagree with?

Each finding links to the clause in both documents. You read the two texts side by side rather than arguing with a summary, and correcting the standard it compared against fixes every future comparison.

Reading the document is not the same as summarising it

A summary tells you what a contract says. Paqt's document layer tells you how it differs from the one you would have signed, what it commits you to, and which of your own approved clauses to use instead — because it works against your library, not against a general impression of contracts.

Four jobs on the document itself

Each one is grounded in something you own — your standard, your library, the executed text — rather than in a model's general sense of what contracts usually say.

Comparison against your standard — Inbound paper is diffed clause by clause against the agreement you would have signed. The output is a list of deviations with what each one changes, not a paragraph describing the contract.
Deviation classified by effect — Each difference is sorted by what it does to your position — acceptable variation, tradeable, or material — against the policy you defined rather than a generic risk score.
Drafting from your clause library — Language comes from clauses your team authored and approved. The assistant assembles and adapts them; it does not invent house style it has never seen.
Extraction from executed paper — Terms, notice windows, renewal triggers, caps and conditions are pulled out of the signed document as structured data, each linked back to its clause.

Why your library beats a bigger model

A general-purpose model produces plausible contract language. Plausible is the wrong target — you need the language your organization has already decided it will defend.

Clauses are versioned objects — Each clause in your library has a version history and an owner, so an approved change propagates to the next draft instead of living in one lawyer's copy.
Templates assemble from clauses — A template is a composition, not a file. Updating a clause updates every template that uses it, which is what keeps a large template set honest.
Findings cite the text — Every deviation and extracted obligation links to the clause it came from, so you verify against the source rather than trusting a restatement.
Attribution on everything — Which clause version a draft used, and which model action produced a finding, are both recorded — so the work is reviewable rather than magical.

What happens when paper arrives from outside

This is the sequence behind every inbound agreement, whether a human or the assistant acts on the result.

Ingest and structure — The document is brought in as a clause-addressable instrument rather than a flat file, which is what makes everything downstream possible.
Compare and classify — Clause by clause against your standard, with each deviation labelled by its effect on your position and linked to both texts.
Summarize by exposure — The brief leads with what is material and what it costs, so a reviewer starts at the risk rather than at clause 1.1.
Hand to the assistant or to a human — The same analysis feeds either path: a reviewer decides, or the assistant negotiates the deviations your playbook already answers.

What this layer does not do

Being precise about the limits is what makes the capabilities believable. These are deliberate design boundaries, not gaps.

Not advice: It does not tell you whether a term is enforceable
Not advice: It does not practise law or replace a qualified reviewer
Not advice: It compares against your standard, not against case law
Not advice: Judgement calls escalate rather than resolve
Not a black box: Every finding links to the clause text it came from
Not a black box: Drafts record which clause versions they used
Not a black box: Model actions are attributed in the timeline
Not a black box: Nothing is generated outside the library without being marked
Not your owner: Your clause library and templates remain yours
Not your owner: Exportable at any time, including the version history
Not your owner: Only marketplace listings you publish leave your workspace
Not your owner: Shadow mode available before you rely on any of it

What buyers ask about the document layer

How is this different from asking a chatbot to review a contract? A chatbot compares your document against its general impression of contracts. This compares it against the specific agreement your organization would have signed, clause by clause, and classifies each difference against policy you wrote. The output is a decision list, not an essay.
What if we do not have a clause library yet? Most organizations have one implicitly — in their last twenty agreements and in one person's head. The practical path is to designate a current agreement as the standard and let the deviations that keep escalating tell you which clauses deserve a library entry.
Does it work on agreements signed before we used Paqt? Yes. Extraction on an executed agreement is the usual first step for a back catalogue, and it is often how teams discover the renewal windows and obligations nobody was tracking.
How do we check a finding we disagree with? Each finding links to the clause in both documents. You read the two texts side by side rather than arguing with a summary, and correcting the standard it compared against fixes every future comparison.