What it is
Mary Technology is a fact-management system for litigation. You load a matter’s documents — discovery productions, emails, medical notes, witness statements, expert reports, handwritten records — and Mary reads every page and turns it into a factual record: each fact extracted as its own entry, linked to the page it came from, tagged with the people and dates involved, and marked as supported, contradicted, or unconfirmed. From that record it builds chronologies and evidence tables by issue, person, or period, and it surfaces the things a paralegal chronology usually misses: conflicting accounts, unexplained gaps in time, assertions no document supports, and documents that are referenced but were never produced. A coverage view shows which pages were processed, which were reviewed, and which were never used.
The architectural choice matters. Mary does not answer questions by retrieving chunks from a vector index, which is how the general-purpose legal assistants work; it extracts discrete fact objects first and reasons over those. That is why its output is a reviewable record rather than a chat answer.
Mary was founded in Sydney in 2023 by CEO Daniel Lord-Doyle, barrister Rowan McNamee, Harry Raworth, and Luke Abagi. In March 2026 it closed an A$7M seed led by OIF Ventures, with Sydney Angels and Empress Capital, to open a San Francisco office and launch self-serve sign-up for smaller firms (LawNext, SmartCompany). Named Australian customers include Shine Lawyers, Maurice Blackburn, Hall & Wilcox, and Zaparas Lawyers. The vendor reported 2,000+ lawyers at the raise and now claims 4,000+ lawyers across 150+ firms — vendor figures, not audited.
Why it shows up in legal-ops stacks
- It replaces the paralegal week, not the lawyer. The chronology on a 10,000-page personal-injury or employment file is usually built by hand in a spreadsheet. Mary’s Smokeball marketplace listing claims a chronology in under 13 minutes and 75–85% time saved on drafting and review; LawNext reports customer-stated savings of 50–90%. Treat the range as the honest number: the saving depends on how messy the file is.
- The output is checkable. Every row links to a source page, so the supervising lawyer audits the chronology by clicking, not by re-reading the file.
- Pricing is published. In a segment where EvenUp, Supio, Everlaw, and Opus 2 are all demo-gated, Mary prints its rates.
Pricing reality
Mary meters by page, not by seat. Checked 2026-09-30 on the US pricing page, in USD:
- Pay-as-you-go: $0.30 per page, no commitment.
- Start: $500/month for 2,500 page credits ($0.20/page), unused credits roll over up to 7,500.
- Growth: $1,000/month for 6,000 credits (about $0.17/page), rollover up to 18,000.
- Business: $2,000/month for 15,000 credits (about $0.13/page), rollover up to 45,000.
- Enterprise: custom volume pricing.
New self-serve accounts get 100 free page credits to test on real documents (LawNext). The math: a 10,000-page file costs $3,000 on pay-as-you-go and roughly $1,670 at the Growth rate. Start beats pay-as-you-go once you process more than about 1,670 pages a month. Seats are not the cost driver, so the whole litigation team can work in the record without a per-user bill — the budget question is page volume per month, which you can pull from last quarter’s matter intake.
Best for
A plaintiff or defense litigation team of 5 to 100 lawyers — personal injury, employment, family law, commercial disputes, in-house investigations — where the first two weeks of every matter go to reading the file and building the chronology. It fits best when page volume is steady enough to sit on a monthly plan and the firm already runs Smokeball, whose Prosper+, Grow, and Boost plans in Australia and the US carry a two-way integration that pulls matter documents into Mary and returns the chronology to the matter.
Skip it if your problem is document review for production — that is e-discovery, and Everlaw or Relativity own it. Skip it if you need the tool to draft the demand letter or brief, not just the fact record.
Versus the alternatives
The incumbents for litigation fact work are Everlaw and Opus 2. Pick Everlaw when the matter already lives in an e-discovery review database — its Storybuilder chronologies sit on top of the documents you are reviewing, so there is no second platform. Pick Opus 2 for trial-heavy commercial litigation and arbitration where the fact record has to flow into bundles, transcripts, and hearing presentation. Pick Mary when there is no review database, the file arrives as a pile of PDFs, and the deliverable is a verified chronology with contradictions flagged — at a published per-page price instead of a quote.
In plaintiff personal injury, EvenUp and Supio are the larger players. Pick them when the output you need is the demand package and medical-specials workup, not just the chronology. The fastest-growing entrants are Harvey and Legora; pick them when one platform has to cover research, drafting, and document Q&A across practice groups, and accept that their retrieval-based answers are not a structured fact record. Clearbrief is the complement, not the substitute: it builds hyperlinked timelines in Word and checks the brief against the record once you are drafting.
If none fit — a solo practitioner with a handful of matters a year — the 100 free credits plus pay-as-you-go at $0.30/page is the lowest-commitment way in; below that, a Word table with pinpoint cites is still the fallback.
Watch-outs
- Certifications are not on the current site. The current site describes matter isolation, encryption in transit and at rest, and no training on customer data, but lists no SOC 2 or ISO 27001 report; an earlier version of the Australian site said Mary was hosted on Microsoft Azure with data stored in Australia and was “working towards” SOC 2 and ISO 27001. Guard: before loading client data, request the current attestation letter and written confirmation of the hosting region for US matters, and put both in the engagement file.
- The record is only as complete as the upload. A document that never reached Mary cannot contradict anything, and “no conflict found” reads the same as “no evidence loaded.” Guard: reconcile the coverage view against the production index before relying on the chronology, and assign every “referenced but missing” document to a named person.
- No public API and no MCP server. Outside the Smokeball integration, documents go in and chronologies come out by hand. Guard: if your practice management system is Clio, Filevine, or Actionstep, price the manual export-import time per matter into the pilot, and ask the vendor for its integration roadmap in writing rather than on a call.
- Credits expire past the rollover cap. Rollover is capped at three months of allotment, so a quiet quarter after a busy one burns paid credits. Guard: size the plan to your median month, not your peak, and cover spikes with pay-as-you-go at $0.30/page.
- Vendor metrics are self-reported. The “8 of 9 litigators preferred Mary’s output” head-to-head at an unnamed global firm and the NPS figures come from Mary. Guard: run the 100 free credits on a closed matter where you already have a lawyer-built chronology, and count missed and wrong facts yourself.