The set is enormous
Base agreement, general conditions, exhibits, addenda — hundreds of pages that only make sense read together, often scanned, rotated, or marked up by hand.
Agent · Contract Review
Upload the base agreement, the general conditions, the exhibits and the addenda. Get back playbook-scored findings across seven review areas, a summary of the commercial terms, answers to anything you ask, and a redline your team can send — with every flag traced to the clause it came from.
Who it's for
Firms signing seven-figure agreements — AIA A104, A101, ConsensusDocs, or an owner's own paper — with no in-house legal team and a signing date that never moves. Outside counsel is too slow and too expensive for every contract, so most get a skim.
The agent doesn't replace your lawyer. It makes sure that when a lawyer or a principal does read it, they start from a scored, cited list of what actually matters — instead of page one. Advisory output; a qualified person reviews and approves. Not legal advice.
Structure, risk allocation, damages, insurance & bonding, notices, payment & termination, site conditions.
Severity scored against your firm's positions, cloneable per client or project — not a vendor's generic ruleset.
Base agreement plus exhibits and addenda, combined and reviewed together — or one merged PDF.
Page and quoted clause on every finding. Un-sourced findings are dropped before the report is assembled.
01 / The friction
The terms that cost you money are rarely hidden. They are standard, boring, and buried on page 140 of a document that has to be signed on Friday.
Base agreement, general conditions, exhibits, addenda — hundreds of pages that only make sense read together, often scanned, rotated, or marked up by hand.
A 200% punch-list holdback. A notice clause that voids your claim because you emailed it. Ordinary language, easy to skim past, expensive months later.
Two reviewers flag different things, and nothing records what was checked or cleared. When a claim lands, that record is exactly what you wish you had.
02 / What we built
The model detects. Code scores, verifies, and assembles.
The language model reads and detects. Everything consequential after that — severity scoring, citation verification, redline assembly — runs as deterministic code, so the same contract produces the same answer twice and every step is inspectable.
03 / The guarantees
Properties of the system, enforced in code and in CI.
These are not projections or averages. They are properties the pipeline enforces on every contract, locked in by the evaluation gate so they cannot quietly regress.
The principle
Keep the judgment with your people. Give the reading to the system.
— How every RivetData agent is scoped
Try it on your own paper
Send it over and we'll return a cited findings report on your own document — free, and with no obligation.