Eve

Legal AI has spent three years being sold to large defence firms and corporate legal departments, where Harvey AI set the template. Eve went the other way, and the choice explains almost everything good about the product.

Plaintiff firms have a different economic structure. They are paid contingently, so time spent is capital at risk rather than billable. Their bottlenecks are specific and well known: reading medical records, drafting demands, chasing providers for documents, answering the phone when a case walks in at nine at night, and noticing the file that has quietly gone stale.

EveOS is organised around those bottlenecks rather than around generic legal research. This review covers what each module does, why the medical chronology work is the most defensible part, what the numbers on the site can and cannot support, and what a managing partner should verify before signing.

What EveOS Actually Runs

• Plaintiff side only, across personal injury, labour and employment, medical malpractice, workers' compensation, and Social Security and disability. That narrowness is the product's main advantage.
• Medical chronologies are the standout capability — automated review of records into a timeline, which is the most tedious and most error-prone task in a personal injury file.
• An auditor runs nightly over open matters, surfacing buried injuries and treatment gaps rather than waiting for someone to re-read the file.
• Security answers the right questions: SOC 2 Type 2, HIPAA compliance, encryption in transit and at rest, US-only data residency, and a statement that client data does not train shared models.
• Backed by a large Series B led by well-known venture firms, which matters when you are handing a vendor the operational core of your practice.

Plaintiff-Side Only, And That Is A Design Decision

Most legal AI products describe themselves as practice-agnostic, which usually means they were built for document review and then pointed at everything else.

Eve names its practice areas and its side of the case. That restriction is what lets the software be opinionated: it can assume there is a client with injuries, a treatment history, an insurer, a demand, and a contingency fee that makes speed worth money. A defence-side or transactional tool — Spellbook works inside Word on contracts, for instance — cannot assume any of those things, so it ends up shipping a search box.

The visible result is a set of modules that map to how a plaintiff firm actually runs — intake, a data layer that structures records and bills, an auditor over open matters, drafting agents for demands and discovery, and an analyst that answers questions about the firm's own performance.

The cost of the same decision is obvious. If your practice is defence-side, transactional, family or immigration, this is not a product you can adapt. Specialisation cuts both ways and here it cuts sharply.

A tool that knows there is an insurer, a treatment history and a contingency fee can make decisions a general-purpose legal assistant is not permitted to make.

The Analyst Is Quietly The Most Interesting Module

Alongside the drafting tools sits a module that answers plain-English questions about the firm: live revenue, margins, settlement pacing.

This is unglamorous and probably the highest-value component for firm leadership. Plaintiff practices routinely run on partner intuition about which case types pay and how long matters take, because assembling that picture means pulling data from a case management system, an accounting system and a spreadsheet nobody trusts.

It also carries an underappreciated risk: an answer given in confident prose is harder to sanity-check than a report with a visible query behind it. Before relying on a figure for a business decision, reconcile it once against your accounting system. If it matches, you have found something genuinely useful. If it does not, you have learned that early rather than in a partners' meeting.

Medical Chronologies Are The Unglamorous Win

If you asked a personal injury paralegal which task they would hand to a machine first, most would name medical record review without pausing.

A serious injury file can run to thousands of pages from multiple providers, in inconsistent formats, much of it scanned, with the clinically significant material scattered through routine notes. Turning that into a chronology is slow, requires real attention, and is where fatigue produces expensive misses.

Eve automates the review into a timeline, and pairs it with a nightly auditor that re-reads open matters to surface injuries that were buried in the records and gaps in treatment that will be used against the client later.

The nightly cadence is the part worth noticing. Records arrive continuously over a matter's life, and the traditional failure mode is that nobody re-reads the file after the initial review. A system that looks again every night addresses a real structural weakness rather than an imagined one.

The necessary caution is that extraction from scanned medical records is genuinely hard, and a missed entry in a chronology is a live risk to the case. Treat the output as a first pass produced by a tireless reader, verified by a person before it drives strategy — which is how the firm's professional obligations require it to be treated in any event.

• Best-fit tasks: chronology construction, treatment-gap detection, provider and bill reconciliation, first-draft demands.
• Supervise closely: anything where a missing record changes the value of the claim.
• Verify independently: dates, provider names and figures that will appear in a filed document.
• Do not delegate: the judgement about what the chronology means for strategy and settlement posture.

The Settlement Numbers Deserve A Second Look

The site carries strong claims: firms handling substantially more cases, settlements 15 to 30 per cent higher, demand letters produced far faster, discovery response times cut dramatically, and a named firm reporting a large jump in annual case volume.

Separate these into two kinds, because they are not equally checkable. Process metrics — how long a demand takes to draft, how quickly discovery responses go out — are plausible and measurable inside a firm. A tool that turns a multi-hour drafting task into a review task genuinely does that.

Outcome metrics are a different matter. Attributing a higher settlement to software requires holding constant the venue, the adjuster, the injury severity, the policy limits, the firm's willingness to try the case and the general movement of the insurance market. Those are the variables that actually determine settlement value, and no software vendor can control for them.

That does not make the figure false. A better-documented demand supported by a complete chronology very plausibly settles higher, and every experienced plaintiff lawyer believes something like this. It means the number is customer-reported experience rather than a measured effect, and should be weighed as such.

The stronger evidence sits elsewhere: a large Series B from investors who conducted diligence, a long list of named firms willing to be associated publicly, and a stated count of firms on the platform. Named references are worth more than percentages, and they are the thing to ask for.

Privilege Is The Whole Security Question

For a law firm, vendor security is not an IT matter. Client confidentiality is an ethical obligation, and in these practice areas the files also contain medical records, which brings a second regulatory regime with it.

Eve's published answers are the right ones. Its security page states SOC 2 Type 2, encryption at rest for all databases and TLS in transit, hosting on AWS with all data located in the United States, least-privilege access with quarterly reviews, background checks and security training for staff, annual third-party penetration testing, and a formal incident response process. Elsewhere the site states HIPAA compliance and that client data is not used to train shared models, with isolation at organisation, user and workflow level.

That set covers the questions a firm's professional responsibility counsel will ask. Two things are worth pursuing directly rather than accepting from a web page: the subprocessor list, including which model providers process matter content, and the retention and deletion schedule when an engagement ends.

Neither absence is unusual — most vendors keep both behind an NDA — but both belong in your file before client material moves. If you carry cyber liability cover, your insurer will want the same answers, and the review is the same shape an accounting firm would run on Black Ore.

• Ask for the SOC 2 Type 2 report itself, with its observation period and any exceptions.
• Request the subprocessor list and confirm which model providers see matter content.
• Get retention and deletion terms in writing, including what happens at termination.
• Confirm the no-training commitment appears in the contract, not only on the website.
• Check the terms against your jurisdiction's rules on confidentiality and supervision of non-lawyer assistance.

How To Pressure-Test Eve In A Pilot

There is no trial and no published price, so evaluation happens through a sales process. A written plan protects you from judging the product by its demo.

The bar to clear is step two. If the chronology on a file you know well misses something material, that is the finding — not a reason to look at the next feature.

1. Choose three closed matters you know intimately, including one that was document-heavy and one that settled badly.
2. Run the medical chronology on all three and check it against what you know is in the file. Count the misses and the inventions separately; they are different failure modes with different consequences.
3. Have it draft a demand for a matter whose demand you already wrote, then compare. Judge it against your own work product, not against a blank page.
4. Put the analyst module against a number you can independently verify in your accounting system, and reconcile.
5. Run intake in parallel with your existing process for two weeks rather than switching, so a missed lead is caught by the old path.
6. Only then discuss commercial terms, with your own measurements in hand instead of the vendor's percentages.

How Eve Is Bought

There is no free tier, no trial and no published pricing. Access begins with a demo and proceeds through a sales conversation, which is standard for software sold into firms rather than to individuals.

In this category, contact-sales pricing usually means the figure scales with firm size, seat count, matter volume and contract term, and that implementation and onboarding are part of the engagement rather than a self-serve signup.

The site does offer a way to think about value, presenting the return in terms of additional settlement value earned annually. Treat that framing as a sales instrument. The honest version of the same calculation is one you run yourself after a pilot, using your own measured before-and-after on demand turnaround and case capacity.

Budget for the evaluation itself. A serious pilot consumes partner and paralegal time, and that time is the real cost of getting this decision right.

Eve: The Firm-Level Facts

Which Firms Eve Fits

Right for your firm if

• You run plaintiff matters in personal injury, employment, medical malpractice, workers' compensation or disability.
• Medical record review is a genuine bottleneck and paralegal capacity limits how many cases you can carry.
• Your intake leaks — calls missed after hours become cases that went to the firm advertising next to you.
• You want firm performance data without assembling it by hand from three systems.
• You can staff a real pilot on closed matters rather than judging the product by demonstration.

Wrong for your firm if

• You practise on the defence side, or in transactional, family or immigration law — the specialisation that makes it good makes it inapplicable.
• You are a solo or very small practice without the volume to justify enterprise software and an implementation.
• You need to try software before buying, since there is no trial and no published pricing.
• Your firm cannot supervise AI-produced work properly, because the professional obligation to verify does not transfer to a vendor.
• You practise outside the United States, where the data residency, the forms and the procedural assumptions do not apply.