The review starts from
a structured file
Incoming contracts arrive with the dates, terms and clauses already extracted, each linked to the paragraph it came from, and with deviations from your own template listed. What the system cannot establish, it reports as missing. The assessment stays with your lawyer — this shortens the reading, not the responsibility.
What this changes in the day's work
Four situations that come up in every legal department with more incoming paper than reading time. None of them is solved by reading faster.
Incoming contracts pile up and the review happens under time pressure
A pre-review arrives with the contract: extracted dates, terms, liability and termination clauses, each linked to the paragraph it came from. The lawyer starts from a structured file instead of page one.
The other side sends a revised version and nobody knows what actually changed
Versions are compared by meaning rather than character by character — across formats, so a PDF against a Word file still works. Two changed words can be a different obligation.
Deviations from your own template are found late, if at all
The contract is checked against your standard and your own rules. What differs is listed with the clause; what is missing is reported as missing rather than quietly passed over.
Obligations and deadlines live in contracts nobody re-reads after signing
Dates and recurring obligations are extracted into a list that can be monitored. The work is finding them once — keeping them visible afterwards is cheap.
Four things it actually does
Named narrowly on purpose. A list of everything a language model can theoretically do is useless when the question is what will run on Monday.
Pre-review of incoming contracts
NDAs, framework agreements, licences: extracted terms, liability caps, termination and renewal clauses, governing law. Each finding carries its clause reference, so verifying takes seconds rather than a second reading.
Version comparison
What changed between two drafts, stated in terms of obligation rather than diff markers. Works across formats and over whole contract portfolios when a clause has to be traced through many documents.
Deviation from your standard
Your template and your rules are the reference. The system reports where the incoming document departs from them and where an expected clause is absent — the second case is the one that gets overlooked.
Search across your own portfolio
Questions answered from your own contracts and documents, with the source paragraph. This is your material only: we do not search case law or legal databases, and we do not pretend to.
How it is introduced
In three stages, and the first one is free. It answers the question that decides everything else: do your own contracts support this? That is a question about the documents, not about the technology.
One contract, one day
freeYou send a single contract; you get the extracted terms and the findings back, each with its clause reference. No access to your systems and no commitment — you are judging the output on your own material.
Pilot on real intake
2–3 weeksFifty to a hundred contracts and your own template as the reference. We measure two things: the share of findings confirmed by your lawyers, and the share of questions your contracts simply do not answer. The second number describes your documents, not the system.
In the workflow
4–6 weeksConnected to where contracts already arrive, with the permissions that already exist, plus a procedure for new versions. Without that procedure the system answers from superseded drafts within months.
What it does not do
Four limits, stated here rather than discovered later. For a legal audience this section is more informative than any feature list — it says where the responsibility stays.
FAQ
Does this replace the lawyer's review?+
No. It changes where the review starts: from a structured file with extracted terms and clause references instead of from page one. The assessment, the risk judgement and the signature stay with your lawyer — we shorten the reading, not the responsibility.
How do we verify a finding?+
Every finding carries the document, section and paragraph it came from, so checking one takes seconds. Where the system cannot establish something, it says so instead of producing a plausible sentence — that honest gap report is the reason the rest can be trusted.
Can it search case law or legal databases?+
No. The material is your own contracts and documents. Searching published case law would require licensed sources we do not have, and a system that guesses at precedent is worse than no system at all.
Does it work with contracts in other languages?+
Yes. You can ask in English about a German or French contract and get the answer in the language of your question, with the reference pointing at the original paragraph. In cross-border work this is the common case rather than the exception.
What about contracts that are only scans?+
They need a text layer. Text recognition can be part of the project, but it has to be agreed: recognition errors in a liability clause are worse than no automation, so we would rather name the condition than quietly accept poor input.
How are access rights handled?+
They are taken from your existing system. Someone who cannot open a contract in the document store does not get answers from it here either. Rights are read from the source rather than configured a second time, so a change takes effect immediately.
Where is the data processed?+
On our own infrastructure in the EEA. Your contracts are not used to train models, the processing locations are disclosed, and a data processing agreement is part of the contract. For privileged material this is usually the first question, which is why it has a verifiable answer.
What does it cost and how long does it take?+
Discovery on your own contracts starts at €5,000; an implementation project on existing infrastructure starts at €25,000. A single-contract test takes a day, a pilot two to three weeks, production another four to six depending on how deep the integration goes.
One contract is enough to judge
Send a contract you know well. You will see what the system extracts, what it flags as a deviation, and where it says it cannot establish something — which is the part worth reading first.
Send a contract