For AI agents: a documentation index is available at https://docs.coverbase.com/llms.txt. This page is also available in markdown by appending .md to the URL.
Contract Guardian extracts clauses, detects missing clauses, analyzes liability and indemnity, surfaces AI-usage provisions, and flags deviations from an organization-defined playbook. Rerun it against amendments and addenda so the contractual control surface stays aligned with the live risk profile maintained in Supplier Radar.
The core idea: you write down your standard once, as tiers of acceptable language. Coverbase then finds the vendor’s version of each clause, decides which tier it lands in, and gives your reviewer a short list of deviations with the source text quoted and highlighted, instead of an agreement to read end to end.
Guardian decides which clauses deviate and how far, then marks the asks up for you. It does not negotiate. A person reviews the marked-up file and decides what to send.
What it does
Clause extraction
Parse MSAs, DPAs, SOWs, BAAs, and order forms to extract clauses, terms, and obligations into structured records.
Missing-clause detection
Compare extracted contracts against your playbook and flag clauses that should be present but aren’t (data breach notification, audit rights, subprocessor flow-down).
Liability and indemnity analysis
Surface liability caps, indemnification scope, and limitation-of-liability language so legal teams can focus review on what deviates from standard.
AI-usage provisions
Detect AI-specific clauses, model training restrictions, output ownership, customer data usage in model improvement, and flag where they’re missing.
Clause sets and risk tiers
A clause set is your playbook. Each reference clause in it is one standard you hold vendors to, carrying a severity, guidance that tells the matcher how to find the clause, the contract components it applies to, and a set of variants: versions of the clause language at different levels of acceptability, tiered 1 to 5. Coverbase matches the vendor’s actual language to the closest variant, and the tier drives the verdict:
Sets copied from the packaged library are scored against a target tier instead, so a vendor meeting or beating the ask conforms. Two guardrails sit on top: a low-confidence match is escalated to Needs review rather than auto-passing, and language found outside the components a clause was scoped to is escalated the same way. Where no language is found, the result is Missing when the relevant component is present in the contract, and Not applicable when it isn’t.
Three ways to build one
Clause set library
Copy one of 12 packaged templates, from Enterprise SaaS to HIPAA to Payments: 311 reference clauses between them, each with four tiers of language already written. Your copy is fully editable.
From your own paper
Upload your template MSA or DPA as a baseline contract, or your negotiation playbook with its preferred and fallback positions. Guardian builds the clause set from the exact language in your document. A playbook kept in a spreadsheet imports directly.
By hand
Author reference clauses directly for narrow, high-stakes standards. Clauses are versioned, so edits never rewrite the history of reviews already run.
Review and triage
Component-aware analysis
Guardian first detects the logical components inside the linked documents (MSA, DPA, order form, SLA, NDA, SCCs, AI addendum, amendments), then evaluates each reference clause only where it belongs.
Evidence-backed verdicts
Each result carries the extracted vendor language, the matched variant, a confidence score, the source document and page, and highlighted evidence in the PDF.
Assessment insights
Where the vendor has a completed assessment, flagged clauses are cross-referenced against that evidence and given a suggested disposition (no action, revise language, add clause, or escalate) with citations back into the assessment.
Triage as the unit of work
Reviewers mark each flag as no action, accepted risk, or false positive, or escalate it to a tracked finding. The system verdict and the human decision are both preserved on the record.
Follow-up drafts
Select the deviations worth pushing back on and generate an outbound follow-up draft built from those specific clauses. Nothing sends automatically: copy it, open it in the email composer addressed to the vendor with the redline attached, or escalate to a finding.
Redlined Word export
Where the contract arrived as a Word file, download it back with the vendor’s wording struck and your playbook’s inserted as tracked changes, plus a comment per clause carrying the ask. It opens in the Review ribbon the vendor already uses.
Draft versioning and redline history
A contract is a family of documents, not a file, and Guardian is built on that model. Drafts, redlines, executed originals, amendments, restatements and exhibits are all held together and ordered, so “what did we agree, and how did we get here” is answerable without opening six PDFs.Typed lineage between documents
Coverbase draws a typed relationship between any two documents in a family: amends, restates, attaches to, redline of, and duplicate of, each recorded with the evidence that produced it. A version chain is data, not a naming convention.
Redlines are their own component
A marked-up draft is detected from its tracked changes and its content, and routed to the redline and negotiation history component. It is kept for the record, and it is never in force: an unsigned Net 45 draft can never compete with the signed Net 60 amendment for the payment-terms slot.
Supersession is explicit
Amendments resolve as an ordered series where each supersedes the last, with the effective date deciding. A superseded instance stays visible in the history rather than being deleted, and anything contributing nothing in force, a redline or a duplicate, renders struck through.
Every answer cites its source
Each resolved term names the document, the page and the effective date it came from: “Net 60, from Amendment 3 effective 1 Jan 2024, supersedes Net 30 in the master agreement.” You can open the source span from the answer.
- Conformed copies are handled deliberately. A master re-issued with its amendments already incorporated is recognized as such, and you choose whether the absorbed amendments stay as separate terms or are marked incorporated. Counting the same amendment twice is the failure mode this exists to prevent.
- Clause sets are versioned. Editing a reference clause never rewrites the history of reviews already run against the previous wording, so a review from last quarter still shows the standard it was actually judged against.
The review’s own history
Alongside the document lineage, each Guardian run keeps its own record: the system verdict per clause, the matched variant and tier, the confidence, the evidence with page citations, and the human triage decision layered on top. Re-run Guardian after an amendment and you get the new verdict beside the old one, so what changed in the controlled obligation surface is visible rather than inferred. Where the contract arrived as a Word file, Guardian can return it with the counterparty’s wording struck and your playbook’s inserted as tracked changes, plus a comment per clause carrying the ask. It opens in the Review ribbon your counterparty already uses, which keeps the negotiation in the format legal teams actually redline in.How to integrate
Document submission
Submit contract documents (PDF, DOCX) for analysis. Returns structured clause extraction and playbook deviation report.
Re-evaluation on amendment
Add amendments and addenda to the contract and rerun the review against the existing paper, surfacing what changed in the controlled obligation surface.
Common workflows
Pre-execution redline review
Pre-execution redline review
Submit drafts to Guardian before legal review. It surfaces deviations from your playbook, missing clauses, and unusual liability terms so reviewers focus on the deltas rather than reading every contract end to end.
Contract repository ingestion
Contract repository ingestion
Bulk-process your existing contract portfolio. Guardian extracts obligations, surfaces gaps against your current playbook, and produces a prioritized list of contracts that need renegotiation.
Living obligation tracking
Living obligation tracking
Guardian-extracted obligations feed the Coverbase obligation surface, where they can be assigned, tracked, and linked to controls. When a vendor’s risk profile changes in Supplier Radar, contractual obligations are already in scope.
Field-level contract inventory
Field-level contract inventory
Pair clause review with Document Insights to populate the contract record itself: value, term dates, notice periods, liability caps, and governing law extracted from every document with page citations.
Setting this up? The Contract Guardian guide walks through building a clause set, writing risk-tier variants, and running your first review. Contract Guardian results integrate with the Coverbase obligation and finding surfaces, accessible through the Export API and MCP server.