Reviewing contracts before you sign with AI

How the bot works

Nine steps. Pulling the text, laying terms beside your playbook, finding what is absent and checking the document against itself are ordinary code and cost almost nothing to run; the model is only asked to turn a clause into a sentence a reviewer can read.

1. It reads the document you were actually sent. The PDF, the Word file, the scan, the exhibit that arrived separately. A scanned page is read before it is used, and a page too poor to read comes back as a page to rescan.

2. Every term it pulls names the clause it came from. Parties, dates, term, renewal and notice, fees and escalation, liability, indemnity, insurance, data, termination, governing law — each one with the section and page behind it, so a reviewer can check a line without reading the whole document first.

3. It lays each clause beside your own written position. Your playbook says what you ask for, what you will accept instead, and what you will not sign. The bot reports where this document sits against that, clause by clause. The positions are yours; the comparison is the work.

4. It names what is not in there at all. No cap. No notice period before a renewal. No cure period. No limit on a price increase. A clause that is not there cannot be measured against your position, and it is easy to miss because there is nothing on the page to catch your eye.

5. It checks the document against itself. The exhibit that prices the work differently from the body. A term that starts on one date in the signature block and another in the schedule. A defined term used before it is defined. Two clauses that cannot both be followed.

6. It writes questions, not a verdict. The output is a numbered list for the person who decides: this clause is outside your position, here is the quote, here is what you usually ask for, here is the question to put to them. It does not tell you whether to sign.

7. Where your playbook has your own language, it offers your language. A suggested edit is a paste of the fallback wording you already approved, marked as a suggestion and attached to the clause it replaces. It drafts no clause of its own.

8. Anything your rules send to counsel goes to counsel. Above the value you set, past the term length you set, anything touching the clause types you marked, any document type you named — routed with the questions already written down, so the review a lawyer does starts further along.

9. It tells you what it could not check. The clauses your playbook has no position on, the pages it could not read, the exhibit that was never attached, the document type you keep out of scope. That list is the useful output in the first week, because it is the part of your paper nobody has decided yet.

What you configure

It can be configured to run in any business. This is what that actually means — the list of things that are yours, not ours.

Your playbook

For each clause type: what you ask for, what you will accept instead, and what you will not sign. Written in your words, versioned, and the only standard the bot measures against.

Your clause types

Which clauses matter on which kind of paper — a lease and a software order form do not get read for the same things.

Your document types

What it reads at all: vendor terms, master services agreements, order forms, statements of work, leases, renewals, non-disclosure agreements. And what it never opens.

Your thresholds

The contract value, the term length, the liability number and the data scope above which a human reviewer or counsel is required rather than optional.

Your routing

Who reads what first, who may approve, which clause types always go to a lawyer, and who is told when a document has been sitting unread.

Your sources of truth

Where the playbook lives, where past agreements and redlines live, which version wins when two folders disagree, and what counts as the executed copy.

Your summary format

What the one-page read looks like: the order of the terms, the fields your reviewers expect, and whether the questions come first.

Your record

What gets stored with the reviewed document — the questions, the playbook version it was checked against, who read it, what they decided — and for how long.

Your release switches

Which document types it may read on arrival, which wait for somebody to ask, and whether suggested wording is shown at all. Nothing is ever sent to the other side.

Transparent, and configurable by you

The dark band below is what the system will not do. This is what you can see and change — and it is how we build everything, not something invented for this page.

You see what it did, and why

Every term in the read names the clause and page it came from, every flag names the playbook position it was measured against, and every run records the playbook version and who reviewed it. A read can be checked against the document rather than taken on trust.

Versioned prompts you can tune yourself

The playbook, the clause types, the thresholds, the routing rules and the prompts behind turning a clause into a readable sentence are yours to tune. A change is a reviewable diff, replayed against agreements you have already signed so you can see what it would have asked about, and rolled back the same way.

A human in the loop wherever judgment actually matters

Reading the document, pulling the terms, comparing them with your playbook and writing the questions are automatic. Deciding what is acceptable, negotiating, taking legal advice and signing are a person's — that boundary is a property of the system, not a promise about how carefully we'll use it.

Every run logged, along with what it cost

One row per document: what arrived, what was extracted with its citation, every clause measured against your position, what was missing, what contradicted what, the questions raised, where it was routed and who read it — plus token usage and cost for that run.

It can be configured to run in any business. The standard is yours: your playbook, your clause types, your thresholds, your reviewers, your counsel. The agreements on this page are one configuration of it, not the limit of it.

Works with your tools

This reads the agreements where they already are and writes its read back beside them. We're not selling you a contract platform, and your files stay yours.

• Microsoft 365 and SharePoint
• Google Workspace
• A shared drive folder
• A contract management system
• A document management system
• An e-signature tool
• Scanned PDFs
• A spreadsheet of agreements

Works well for

• Law firms — A first pass against the firm's own playbook, so the read starts at the clauses that are not standard.
• Manufacturers — Supply agreements arrive on the supplier's paper, with the price, the term and the remedy already chosen.
• Companies that lease their locations — A lease is long, the renewal and the operating costs sit deep in it, and it gets signed once.
• IT services and managed service providers — Vendor terms and pass-through clauses stack up behind every product they resell for somebody else.
• Medical and dental practices — Payer agreements, equipment leases and service contracts all land on somebody else's template.

How a rollout works

Nothing is signed on day one, and nothing is sent to anybody. The first milestone is a set of agreements you have already signed, read again against your own playbook, so you can compare the questions it raises with the ones you actually asked.

1. 01Discovery

   The quote call, then a working session. We go through the kinds of agreements that arrive, where your standard positions are written down today (and which ones are only in somebody's head), your thresholds, who reviews, when counsel is required, and what must never be read automatically. You get the playbook structure, the rules and a written scope.

2. 02Dry run

   Agreements you have already executed, read again. Nothing is written back and nothing is sent. You compare the questions with your own negotiation history — and the clauses it had no position for are where the work is, because that is your playbook telling you what it is missing.

3. 03Assisted mode

   Live on one document type. The bot reads what arrives and writes the questions beside it; your reviewer decides and your counsel still sees whatever your rules send them. Who signs does not change.

4. 04Optional limited automation

   Optional, and narrower than it sounds. A short list of document types you name, read on arrival with the summary written back automatically, still with every hard stop in force, nothing suggested to the other side and the kill switch. Plenty of teams stop at step 03, and that's a good place to stop.

What we don't claim

• No numbers we haven't earned. No accuracy percentage, no turnaround figure, no clauses-per-hour claim, no risk score. We have not read your paper. What a read takes and what it costs come out of the dry run, and they will be your numbers.
• No customer claims on this page. No names, no case studies, no logos.
• It is not a lawyer and it does not replace one. There is no configuration in which this gives legal advice. On anything with real exposure the design routes to counsel instead of answering, and the read exists to make that review start further along rather than to avoid it.
• It cannot check a position you have not written down. If the playbook says nothing about limitation of liability, the bot has nothing to measure that clause against and will say so. Writing the positions down is work, it is yours, and it is worth doing whether or not a bot ever reads them.
• A document it found nothing in is still a document somebody has to read. Nothing here produces an approval, and the person who signs is the person who decides. Anything of consequence should be read by a person as well.

Questions people ask

Can AI review a contract before we sign it?

Yes, as a first pass with a person at the end of it. An agent can read the document you were sent, pull every term out with the clause and page behind it, lay each clause beside the position you have written down, name what is missing and what contradicts something else in the same document, and hand a reviewer a numbered list of questions. What it will not do is tell you whether to sign.

Is any of this legal advice?

No. It is not a lawyer and it says nothing about whether a clause is enforceable, how a court would read it or which law should govern. It reports what the document says and how that compares with your own written positions, and your rules decide what goes to counsel.

What does it check the contract against?

Your playbook: for each clause type, what you ask for, what you will accept instead, and what you will not sign. If that lives in a document today, we start from the document. If it lives in the head of the person who always negotiates, the discovery session is where it gets written down.

What happens when a clause is something our playbook never covered?

It comes back as a question with the clause quoted, not as a recommendation. A position nobody has taken is not something a model should take on your behalf, and the list of those clauses is how a playbook grows.

Does it redline the document?

It can attach your own approved fallback wording to the clause it would replace, marked as a suggestion for a person to accept or discard. It composes no new clause and it sends nothing to the other side, so nothing it suggests can reach the counterparty by itself.

Can it read a contract that was scanned?

Yes, and it reads the scan before it uses it. A page that is too poor to read comes back as a page to rescan rather than as guessed text, because a term read wrong is worse than a term marked unreadable.

How is this different from tracking the dates in a contract we already signed?

That is a different bot with a different boundary, and it has its own page. This one ends at signature: it reads the draft and raises questions. Tracking the renewal, the notice window and the obligations afterwards starts where this stops.

Will it slow down a deal we want to sign quickly?

The read happens when the document arrives rather than when somebody gets to it, and the questions are written before the first meeting about them. The decision still takes exactly as long as your people need, and your rules decide what cannot move without a lawyer.

Is there a record of what it checked?

Yes, per document: what arrived, every term with its citation, every clause measured against your position, what was missing, what contradicted what, the questions it raised, where it routed, who read it and what they decided - plus the token usage and cost for that run.