Every finding cites its source page

Your discovery pile,
reviewed overnight.

DocuLightning reads, classifies, and reviews legal documents end to end. It hands back a first-pass review your associates can actually check, because every line traces to the page it came from.

14 days, 500 pages, no card. SOC 2 Type II. Your documents never train a model.

MSA_Delaney_v7.pdf page 22 of 118

11.4  Each party shall indemnify and hold harmless the other from any claim arising under this Agreement, without limitation as to the aggregate amount of such liability, and this obligation shall survive termination.

Uncapped indemnity, off playbook

Your standard requires a cap at 12 months of fees. Clause 11.4 has no cap and survives termination. Compare against the 6 prior Delaney agreements, none of which omitted the cap.

DEL-004221  p.22  cl.11.4
In production at Kirkland & Ellis Latham & Watkins Sidley Austin Jones Day Morgan Lewis
Audited against SOC 2 Type II ISO 27001 ISO 27701 ISO 42001 GDPR and UK GDPR CCPA
The problem

Most legal AI gives you an answer. It rarely gives you a way to check it.

Courts have made the position clear: the lawyer signs, so the lawyer verifies. A tool that produces a confident summary with nothing behind it moves the work rather than removing it. DocuLightning is built the other way around, from the citation up.

Reads what the last tool skipped

Vision-model extraction handles handwriting, stamps, redlines, faxes, and thirty-year-old scans. Tables keep their structure and multi-column pages keep their reading order, so nothing quietly drops out of the index.

Agents that finish the task

Give one instruction. The orchestrator plans the pass, extracts clauses against your playbook, cross-checks the earlier versions, flags what deviates, routes it for approval, then reviews its own output before it reaches you.

Defensible from the first run

An immutable audit log records every step, prompt, model version, and human override. Export the full chain for opposing counsel, your GC, or a regulator without reconstructing anything after the fact.

How it works

Four steps. You keep the last one.

The pipeline is deliberately boring and inspectable. Each stage writes its output to the record before the next stage reads it.

01 / INGEST

Point it at the pile

Drop a folder, connect iManage, NetDocuments, SharePoint, or a Relativity workspace. 40+ formats, no page limit.

02 / PARSE

Turn paper into structure

Layout-aware extraction produces clean text with coordinates, so every downstream claim keeps a pointer back to the pixel.

03 / REVIEW

Run the pass

Agents classify, extract, compare against your playbook, and escalate anything ambiguous instead of guessing at it.

04 / SIGN OFF

You decide

Findings arrive ranked with sources attached. Accept, edit, or reject. Every action lands in the audit log under your name.

Customers

Who is running it today.

Six Am Law 100 firms, eleven in-house departments, and three managed review providers run DocuLightning on live matters. Reference calls are available once you are in an active evaluation.

6
Am Law 100 firms, four of them in the top 50
11
In-house departments, Fortune 500 and mid-market
3
Litigation support and managed review providers
2.4M
Pages reviewed since the first production deployment
Kirkland & Ellis

Diligence on a nine-figure carve-out

A 14,000 document data room screened for change of control, assignment, and exclusivity in a single overnight pass. The team opened 62 agreements instead of all of them, and every flag arrived with the clause attached.

1,900 attorneys
Corporate and M&A
Deployed in their own VPC
Jones Day

First-pass privilege screen

A 240,000 document production cut to a defensible review set with a documented reason on every call. Contract reviewer hours dropped by roughly two thirds, and the audit log went into the record without reconstruction.

1,100 attorneys
Litigation and e-discovery
US data residency, model versions pinned
Sidley Austin

Vendor paper against a playbook

Every inbound MSA and DPA marked up against 34 fallback positions before it reaches a lawyer. Median turnaround on a routine vendor agreement went from four days to under two hours.

28 person legal team
Regulated industry, EU residency
Integrated with iManage
Measured on real matters

What the first ninety days look like.

Aggregate figures across our design partners in the first quarter of 2026. Your numbers depend on document quality and how tight your playbook is.

94%
of first-pass review completed without human touch, on contract sets with an established playbook
6.2h
median turnaround on a 10,000 document production, from upload to ranked findings
100%
of findings carry a page-level citation, verified by a second pass before delivery
Where teams start

Four workflows, live today.

Every one of them runs on the same parse and audit layer, so a finding in diligence looks identical to a finding in litigation.

Contract review against a playbook

Load your fallback positions once. Every incoming paper gets marked up against them, with the deviation, the risk tier, and the clause you would normally counter with.

Diligence and data room triage

Change of control, assignment, exclusivity, and MFN provisions surfaced across thousands of agreements, sorted by what actually threatens the close.

Privilege and responsiveness screening

A defensible first cut on productions, with confidence scores and a documented reason on every call, plus mandatory escalation on anything borderline.

Legacy archive recovery

Boxes of scans, faxes, and handwritten amendments turned into a searchable record with the marginalia intact and attributed to its page.

Honest about the limits

What DocuLightning will not do.

It will not give legal advice

It surfaces what a document says and where it deviates from a standard you defined. The judgment call stays with a licensed human.

It will not hide uncertainty

Ambiguous documents get escalated rather than resolved. You see the confidence score and the reason, not a clean number covering a guess.

It will not replace the reviewer

The output is a first pass. Firms that treat it as a final answer are the ones ending up in the hallucination case law, and we would rather not sell to them.

Run it on one matter.

Upload a real set, compare the output against your own first pass, and keep the audit log either way. Most teams know inside a week.

Start free trial Book a demo