Investor overview

Confidentiality is not
a constraint. It is the product.

Zero-transfer litigation AI with a frontier-quality escape hatch — every safety invariant enforced by the build, not by convention. A practice the attorney owns outright, designed from the rules upward.

Confidential overview — figures are plan, not results.

Pre-revenue. The invariants and the working prototype exist; the full agent set is in active development. Figures below are the plan, and the risks are listed with them.

The problem

The sanctions wave meets the privilege problem

Two separate liabilities, not one: fabricated citations (sanctions, Rule 11) and disclosure to a third party (privilege waiver, HIPAA). Most vendors answer only the first.

17–33%
of queries hallucinate on leading legal-research AI tools Stanford RegLab; general-purpose chatbots are worse
$145K+
in AI-hallucination sanctions in Q1 2026 alone a single quarter
1,450+
documented AI-hallucination incidents leading global database, May 2026

Purpose-built tools are no safe harbour

Sanctions have reached users of major commercial legal platforms. The exposure is not hypothetical and it is not shrinking — it is being written into court records, quarter by quarter.

Privilege is the second liability

Personal-injury and employment practices carry HIPAA and privilege duties at once: every cloud call with PHI or work product is a potential disclosure. A federal court has already held a public-chatbot exchange not privileged — the data went to a third party.

The solution

The privilege boundary is code, not policy

  • Seven agents in a pipeline — intake → documents/OCR → timeline → research → drafting → verification → attorney review. Tireless paralegals, each with one job, all supervised.
  • Cloud calls carrying protected content are blocked by default — prohibited unless deliberately, verifiably permitted.
  • Every assertion is grounded — each citation points to a specific source file and page. A witness testifying only from documents in evidence.
  • Claim-by-claim verification before any human sees the draft — anything unsupported is flagged [VERIFY].
  • Nothing is filed automatically — a licensed attorney reviews and signs every output. The system cannot submit or send on its own.
  • Measured, not claimed: on a twelve-claim controlled set from our synthetic matter, the verifier flagged both a planted $3,800 billing overstatement and a fabricated assertion, passed all ten sound claims, and backed every pass with a verbatim quote checked against the source. Local model, no cloud. Twelve claims is a first measurement, not a validated rate — and our drafting stage does not yet emit citations, so the end-to-end run is not yet demonstrable.

Why now

Frontier-class AI finally fits on a desk

Until recently, privilege-grade AI meant a painful trade: weak local models, or frontier cloud models with confidentiality risk. A desktop AI machine with 128 GB of unified memory now runs a 120-billion-parameter model entirely inside the office.

The trade-off structurally narrowed

Drafting and verification now run at credible quality inside the firm’s walls, on hardware the firm owns and can unplug. The cloud becomes optional rather than mandatory — used only where frontier quality genuinely matters, and only through the de-identification gateway.

A hardware-timing story, not a research bet

The architecture was always right; the hardware finally caught up. A $4,699 device now hosts models that needed a server room two years ago. We are not betting on a research breakthrough — and hardware is a low-single-digit share of the raise.

Compliance

Three shields

The architecture was not built first and checked against the rules afterwards — it was derived from them.

Protected health information

HIPAA

  • Fully local mode: PHI is never disclosed to any third party
  • The hybrid cloud path runs only under a signed BAA plus zero data retention
  • De-identification precedes any transfer
Fla. Op. 24-1 · ABA Formal Op. 512

Bar ethics

  • Generative AI is permitted subject to confidentiality, supervision and honest billing
  • In-house AI that discloses nothing to a third party removes the core confidentiality problem
  • The AI is supervised as a non-lawyer assistant (Rule 4-5.3)
Candour to the tribunal

Rule 11 / sanctions

  • Every assertion grounded in a source file and page
  • Claim-level verification against a verbatim-quotation standard
  • Mandatory attorney signature — the record of reasonable inquiry is built in

Engagement letters disclose AI use expressly; audit logs substantiate every claim made about it. Citations should be verified by counsel before use with a bar audience.

Structure

The lawyer owns the practice

Two entities by design: a 100% attorney-owned PLLC that practises law, and a technology MSO paid in fixed, fair-market fees.

PLLC — practises law

100% owned by a licensed attorney
  • Owns every client relationship
  • All legal judgement, case selection, fee setting and signatures live here
  • Technology is staff, not a partner

MSO — sells infrastructure

Owns platform, hardware and IP
  • Co-owned by the operators; the attorney holds equity too
  • Paid in fixed, fair-market fees only
  • Never a percentage of legal fees; no participation in case decisions

The bright line — no fee splitting, no non-lawyer control. Follows Texas Ethics Op. 706 and Fla. Rules 4-5.4 / 4-5.7, and deliberately inverts every fact of the TIKD case: only the lawyer selects matters, sets fees and directs representation. Stated honestly: Florida has no MSO-specific opinion yet — an advisory opinion request is planned before scaling.

Business model

Two engines, one flywheel

Engine 1 — contingency fees

The owned practice

  • Personal-injury and employment matters — the pipeline compresses a 15–25 hour demand package into a few hours of attorney review
  • Revenue: standard contingency fees
  • Every matter proves the product on real work
Engine 2 — fixed fees per seat / matter

The MSO platform

  • The same kit, white-labelled for other firms as a preconfigured appliance plus service
  • Never a share of their recoveries
  • Recurring revenue smooths the contingency cycle
Jurisdiction · judge · adjuster patterns

The flywheel

  • Every matter enriches a de-identified outcome corpus no national vendor has
  • Each engine strengthens the other
  • The moat is the licensed-firm wrapper plus the data plus the position of trust — not a model, which commoditises

The plan

Twelve months to revenue-ready

Headline figures from the seed plan. The full P&L, hiring schedule and use-of-funds reconcile line by line — available under NDA.

$1.3M
seed ask ~80% to the team; equipment + formation ≈3.6%
3 → 10
headcount over year one every hire tied to a pipeline or caseload trigger
Day 90
fully-local vs hybrid quality verdict judged on real matters — the bet stays reversible
~$910K
year-one burn, optimistic case we say it straight rather than pretend fees arrive early
Roadmap milestones. Two external dependencies are marked honestly: workstation shipping dates (until then, existing PCs) and the timing of the bar advisory opinion — which gates aggressive MSO scaling, not the owned practice.
WhenMilestone
Month 1Entities formed and papered; hardware received; insurance bound; pipeline migrated; first owned matters opened
Day 90Fully-local quality verdict on real matters → product positioning decision (local-only versus hybrid)
Months 3–6De-identification layer hardened; 2–3 MSO partner firms signed; caseload carried with paralegal support
Months 6–9Associate attorney onboarded; workstations evaluated as they ship; MSO appliance packaged
Month 12Production pipeline on live matters + first white-label MSO deployment running; first settlements arrive

Risks

Stated plainly

Mitigations are implemented or budgeted, not promised.

RiskMitigation
RegulatoryFlorida has no MSO-specific opinion; an adverse reading is possible — met with strict fair-market papering, attorney control, and a pre-emptive bar advisory opinion request
Model qualityLocal models trail the best cloud models on the hardest persuasive drafting — which is exactly why hybrid mode exists; the 90-day verdict decides on data rather than hope
Revenue timingContingency revenue is back-loaded and year one is capital-intensive even optimistically — recurring MSO revenue is the smoothing engine
Hardware timingWorkstations are announced but not shipping and device prices are volatile — a budgeted reserve line covers it; attorneys work on existing PCs until devices arrive
Verification limitsNo automated checker is ever 100% — attorney oversight is structural, not optional: nothing is filed without a human signature
Build maturityThe reasoning agents are not yet written; the invariants are. The order is deliberate — the boundary and the audit chain cannot be retrofitted, the agents can be built on top of them

Next step

A structure session and a live demonstration

The strongest close available is the one claim that can be tested in the room: a live run of the verification gateway on a matter whose answer is already known.

Request the full deck

Full P&L, hiring plan and use-of-funds are shared under NDA. Write to hello@lexvault.associates, or see the product overview for law firms.