LEXVECTOR assigns a dedicated AI agent to each litigation matter — and houses it inside a privileged workspace built for California's new agentic AI rules. Evidence organization, discovery drafts, deadline tracking, client communication. Entirely through your existing inbox. No dashboard. No software to install. No consumer-grade AI exposure.
Big firms have eDiscovery teams and six-figure Relativity contracts. You have a inbox full of client text messages, scanned PDFs, and bank statements — and one paralegal, if you're lucky.
Photos of receipts. Forwarded text threads. Mislabeled PDFs. WhatsApp screenshots. Every case begins with hours of human labor just classifying and filing what the client sent in. The bill runs $5K–15K before you've drafted a single sentence of the complaint.
"Any update?" "When will you file?" "Did you get my email?" An estimated 30–40% of your day goes to client status updates. You can't ignore them — they'll go elsewhere. You can't bill for them either.
Relativity, Everlaw, and Reveal start at $50K a year. They solve a problem you don't have — document review at scale — and skip the one you do: turning a pile of client materials into an element-by-element evidence map.
The risk runs both ways. In U.S. v. Heppner (S.D.N.Y., Feb. 2026), Judge Rakoff held that documents a defendant prepared with consumer-grade Claude — then shared with counsel — were not privileged. That's your client using ChatGPT. Now flip it: the California Bar's May 14, 2026 Practical Guidance expressly extends Rule 1.6 confidentiality to AI inputs. Typing client data into ChatGPT or consumer Claude — even when no human ever sees the prompt — can itself count as a "reveal" of confidential information if there is a material risk the system accesses, retains, or trains on it. That's you using ChatGPT. Two different rules. Same consumer tool. Same disciplinary exposure.
In four months, the question for every California litigator changed twice. U.S. v. Heppner (S.D.N.Y. Feb. 2026) made client AI conversations a privilege problem. The California Bar's updated Practical Guidance (May 14, 2026) made agentic AI a supervision problem. PAWS — Privileged AI Workspace Structure — was built as the answer to both.
Client AI use happens at the direction of counsel, under terms in the engagement letter — not as independent consumer-grade AI use. The contractual basis matches the doctrinal path Judge Rakoff acknowledged in Heppner: direction, not coincidence, distinguishes privileged from unprivileged.
LEXVECTOR operates as an extension of the firm — full attorney access, every agent action confidence-scored and routed to attorney review when ambiguous. Every client communication confirms back through the attorney before sending. This is the "meaningful lawyer supervision and review" the May 14 Practical Guidance demands, applied as the substrate of the product.
Client materials and AI conversations never train shared models. Case data is encrypted at rest, isolated per matter, and never used for any purpose other than that representation. Built to the Rule 1.6 confidentiality standard the updated Guidance now reads onto AI inputs.
(S.D.N.Y., Judge Jed S. Rakoff) — 31 documents a defendant prepared with consumer-grade Claude, then shared with his defense team, held not privileged — the AI was not at counsel's direction, and the platform was consumer-grade. Your clients are doing the same thing on ChatGPT right now. You cannot stop them. You can give them a privileged channel to use instead.
Every AI interaction is logged into an immutable, privilege-tagged audit log. When opposing counsel files a motion to compel or challenges privilege — or when the Bar asks how supervision was performed — the firm can produce a court-ready privilege log + supervision record meeting FRCP 26(b)(5) and the May 14 Practical Guidance standards. Included in every case, no per-event fee.
You're not paying for a platform. You're paying for deliverables — the same work product a paralegal would hand you, ready for your review.
LEXVECTOR ingests everything the client uploads and organizes it against each cause of action — element by element. You get a matrix showing what evidence supports each claim, plus a missing-evidence checklist to send back to the client.
Based on the pleadings, the agent drafts your RFP, interrogatory, and RFA topic lists, suggests subpoena targets, and writes a discovery strategy memo identifying your strongest evidence and most urgent gaps.
Forward opposing counsel's 30-page RFP to the case email. LEXVECTOR parses each request, searches your case file for responsive documents, drafts a request-by-request response matrix, and surfaces privilege concerns for your review.
When a client emails to ask for an update, LEXVECTOR drafts the reply and sends it to your inbox. You click [Send] and it goes out from your email address within 30 seconds. Every client-facing message requires your confirmation — no exceptions.
As your case moves toward MSJ and trial, LEXVECTOR builds an exhibit-by-exhibit evidence dossier: chain of custody for each document, foundation analysis for admissibility, hearsay and authentication issues flagged with proposed cures, and a witness-by-witness matrix tying each piece of evidence to the testimony needed to authenticate it. Every exhibit comes with a one-page foundation memo ready for your review.
At 8 AM, one email lands in your inbox. It tells you which cases need a decision today, which clients uploaded materials overnight, what deadlines are coming this week, and how many hours the agents saved you. Ten minutes to plan the day.
Four moments from the same case file. Same client. Same agent. Different stages of the litigation lifecycle. This is what the work product actually looks like.
Robert (the client) gets a Magic Link by email. Twelve-step intake on his phone. When he uploads a document labeled "bank statement," the agent recognizes it's actually the disputed invoice — and asks him to confirm before filing it under the wrong category.
This looks like an invoice, not a bank statement:
Is this the disputed invoice?
A WhatsApp screenshot comes in. The agent's classification confidence is 0.71 — below the threshold. Instead of guessing, it sends a 30-second clarify email. You reply "B" by voice from your car. The task resumes automatically.
Hi Jing,
Robert just uploaded a WhatsApp screenshot dated June 12, 2024 — a conversation between him and someone listed as "David J. — Jones."
I'm not sure how to classify it. Two best guesses (confidence 0.71):
A. Chat record (informal communication → Communications bucket)
B. Evidence photo (substantive content about the unpaid invoice → Damages / Notice evidence bucket)
Looking at it: Jones mentions "I'll wire it Friday for sure" referring to the $200K invoice. That feels substantive to me — but I'd rather you decide.
Just reply "A" or "B" and I'll continue. (Or "hold" if you want to look first.)
— Amy
Once enough materials are in, the agent generates an element-by-element Evidence Map. Six rows. Each fact carries a confidence score and a source link. Click "Invoice_0324.pdf, p.2" — the PDF opens to page 2, highlighted on the "Net 30" clause that establishes breach.
Forward the opposing counsel's discovery request to the case email with subject [OUR-DISCOVERY]. Twenty-four minutes later, a Plan email lands in your inbox: draft RFPs, ROGs, and RFAs — bounded by California's statutory limits — with a strategy memo and source-linked findings. Approve, edit, or clarify in one tap.
I drafted our first round of discovery. Ready for your review.
Targeted at liability and damages, per your forward. All requests within California statutory limits:
→ Strategy memo: strongest evidence on Breach and Performance. Damages calculation needs Jones's Q2 banking records (RFP #14). Mitigation gap remains — flagged for client follow-up.
Same agent, same case file, same inbox — from intake to discovery in four months. No dashboard switching, no manual data re-entry, no consumer AI exposure. Just the work product, attorney-reviewable, source-linked, ready when you are.
We made one counterintuitive but deliberate decision: LEXVECTOR has no main dashboard. Every interaction happens where you already work — Outlook, Gmail, your phone's mail app.
Each morning at 8 AM, you receive a single email. It's a status report from the team of agents working your cases:
A litigator's tolerance for AI error is zero — one bad citation and you're explaining yourself to the bar. Everything below was built with that in mind.
Each fact, claim, and coverage item in LEXVECTOR's output is linked back to the source document, page, and paragraph. One click to verify. The agent is structurally incapable of generating a fact without a citation.
Every output carries a confidence score. High-confidence work is delivered. Low-confidence work is flagged for your review. Privilege and sensitive content never auto-process — they always require a human decision.
Forward a discovery request to the case email and the response pipeline kicks off. Reply to the agent and it understands. No plugin to install, no IT approval needed, no new tool for your paralegal to learn.
Each client's case file lives in its own encrypted workspace, isolated from every other matter on the platform. No cross-firm data leakage. No background training on your case materials. Every document, communication, and AI interaction is treated as confidential by default — only you and your firm have access.
Every output is an attorney-reviewable draft. The agent doesn't write your pleadings, predict outcomes, estimate settlement value, or file anything with the court. Nothing leaves your firm without your name on it.
This isn't a software subscription. LEXVECTOR is priced like court reporters and e-discovery vendors — case-level disbursements that pass through to your client under standard California fee agreement provisions. Your firm pays a $500 one-time activation fee. Everything else moves through the case.
UPL risk is real. Bar discipline is real. LEXVECTOR is designed so neither falls on you. Here is exactly where the line sits.
LEXVECTOR is priced like the litigation support services already on your invoice — court reporters, e-discovery vendors, expert witnesses — not like a per-seat SaaS tool. Pass through to the client at a 30% discount (default), or pay it from the firm at full price (used for contingency, pro bono, or relationship matters). PAWS architecture, privilege log export, and the full client experience module are included in base pricing — no add-ons, no per-event fees.
Onboarding, security setup, billing integration, matter templates, and engagement-letter language support. Covers firm-level setup; never passed through to any client.
Tiered by total active caseload across your firm:
Standard "costs and expenses" provisions in California fee agreements already authorize pass-through of matter-specific technology costs to clients — the same mechanism that lets you bill clients for court reporters, e-discovery vendors, and expert witnesses. The 30% volume discount is a vendor-level discount tied to billing path, not a kickback or markup — structurally identical to per-matter pricing Westlaw, Lexis, and Relativity have offered for decades without raising CRPC concerns.
| Component | Pass-Through | Firm-Pays |
|---|---|---|
| Monthly Technology Fee · 12 mo | $1,908 | $2,724 |
| Discovery Support Package | $2,500 | $3,570 |
| MSJ Support Package (50% probability) | $1,600 | $2,285 |
| Typical case total | $4,000 – $4,900 | $5,700 – $7,000 |
Every new matter defaults to pass-through. Firms can opt into firm-pays on a matter-by-matter basis. No firm-wide commitment required — each case is decided independently.
In the post-Heppner world, your client's AI use is either privileged or it isn't. PAWS architecture, the audit log, and court-ready privilege log export inside LEXVECTOR represent roughly $700 of each year's case cost — and they exist so a single privilege challenge doesn't undo the rest of the work on the file.
On a $200K commercial matter, a privilege waiver can cost the client the core narrative of the case. $700 of disbursement is a rational price for that protection — far less than the deposition transcript on the same matter, and orders of magnitude less than the cost of arguing privilege from a weaker position.
Not for big-firm litigation support teams. Not for in-house counsel. For solo practitioners and small firms who handle their own evidence work and have never had access to enterprise-grade tooling.
Heppner (S.D.N.Y. Feb. 2026) held that documents a defendant prepared with consumer-grade Claude — and later shared with counsel — were not protected by attorney-client privilege or work product doctrine. The court's reasoning turned on the absence of attorney direction and the consumer-grade nature of the tool. LEXVECTOR is structurally the opposite: client AI use happens at the explicit direction of counsel under engagement-letter terms, inside a privileged enterprise workspace with full attorney supervision, with an immutable audit log evidencing every interaction. We call this architecture PAWS — Privileged AI Workspace Structure.
What this means for California. Heppner is a federal decision from the Southern District of New York, but its reasoning applies to any client communication with consumer AI — including communications by California clients with their California attorneys. California is independently moving in the same direction: on May 14, 2026, the California Bar Board of Trustees approved updated Practical Guidance that, for the first time, specifically addresses agentic AI and requires meaningful lawyer supervision over any AI system that makes substantive legal determinations. California litigators should expect that Heppner's privilege analysis will be cited by adversaries in California state and federal proceedings, and that the May 14 Guidance — which is effective now — defines the supervision standard LEXVECTOR was built to.
Probably yes — and the California Bar just said so directly. The May 14, 2026 Practical Guidance, together with proposed Comment 2 to Rule 1.6, expands the definition of "reveal" to include exposing confidential information to AI tools where there is a material risk the system may access, retain, or use the data inconsistently with the duty of confidentiality. This catches consumer-grade ChatGPT, consumer Claude, and similar public platforms — whose terms of service expressly permit retention, training on user inputs, and disclosure to third parties.
The practical implication is that the disciplinary risk runs to you, not just your client. When your client uses ChatGPT, the consequence is privilege waiver — bad for the case. When you use ChatGPT to analyze your client's documents, the consequence is a Rule 1.6 violation — bad for your bar card. Most small-firm attorneys still treat ChatGPT as a productivity tool. After May 14, the State Bar treats it as a third-party disclosure.
How LEXVECTOR solves this. Every matter lives in its own encrypted workspace, isolated from every other case on the platform. We do not train shared models on client data. We do not share inputs with third parties. The contractual terms between LEXVECTOR and your firm are written to satisfy the "material risk" test in the new Comment 2 — so that when you analyze a client document inside LEXVECTOR, you are not making a Rule 1.6 disclosure. You are exercising your duty of confidentiality, not breaching it.
The California Bar's updated Practical Guidance (approved by the Board of Trustees on May 14, 2026) sets out three requirements that map directly to LEXVECTOR's design philosophy from V0:
1. Meaningful lawyer supervision over agentic systems. Every agent action in LEXVECTOR carries a confidence score. High-confidence work proceeds; anything ambiguous is routed to an attorney decision. Every client-facing message requires attorney confirmation before sending. The full audit log produces a record of supervision that can be inspected if the Bar ever asks.
2. Confidentiality covers AI inputs under Rule 1.6. Per-matter isolation, zero-retention infrastructure, encryption at rest, and a contractual prohibition on cross-matter or shared-model training. Client materials never leave the privileged container of the case.
3. Verification of all AI-cited authorities under Rule 3.3. Every fact LEXVECTOR generates is structurally linked to a source document, page, and paragraph. The system is incapable of producing a citation that does not trace back to a verified source — eliminating the hallucination class of risk that has produced the Bar's sanctions trend since 2023.
We did not retrofit these features in response to the May 14 update. They were the design philosophy from V0. The Bar's guidance and LEXVECTOR's architecture were converging on the same answer to the same question.
We provide recommended language covering three things: (1) pass-through of case-specific technology costs as a disbursement, with explicit volume-discount disclosure; (2) attorney's direction of client AI use as part of the representation; and (3) an exclusive-channel commitment asking the client not to discuss the matter on consumer AI platforms like ChatGPT. The Exclusive Channel provision is new in V5 and exists to give attorneys an ethical defense if a client violates it.
All data is encrypted at rest with AES-256 and in transit with TLS 1.3. We maintain a full audit log of every access event. We're built BAA-ready and on a path to SOC 2 Type I within twelve months of launch. Your data is never used to train shared models. Each matter is isolated; cross-matter access requires explicit firm-level authorization.
Every client-facing message requires explicit attorney confirmation — no auto-send is possible. Intent classification screens for legal-advice language and routes it to your review. The client-facing email templates are restricted to status reporting; they don't make legal judgments. Your master services agreement names you as the final responsible party.
Every output is an attorney-reviewable draft with source links on every fact. If a fact is wrong, you'll see it before it leaves your firm. The agent flags low-confidence items rather than guessing. We maintain an evaluation framework with 50+ ground-truth cases and run regressions on every model update.
No. LEXVECTOR works through your existing email — Outlook, Gmail, or any IMAP-compatible client. There is no plugin requirement, no IT approval needed. An optional Outlook add-in is available for one-click case routing, but the core workflow runs entirely on email forwarding.
Your firm pays a one-time $500 activation fee. Everything else — the monthly per-case technology fee ($89–$189) and stage-specific support packages ($500–$3,800) — is passed through to the client as a case disbursement, the same way you bill for court reporters and e-discovery vendors. A typical commercial matter runs $4,000–$4,900 in total LEXVECTOR fees across its 12-month lifecycle, or about 2–2.5% of a $200,000 client matter. Of that, roughly $700/year is privilege-architecture (PAWS) protection — included in base, not an add-on. See the full pricing breakdown for details.
V0 supports California commercial litigation — breach of contract and small business disputes. Additional practice areas will be added based on design-partner demand and engineering capacity.
We're onboarding a select group of California firms as design partners. Twelve months free, hands-on support, direct input into the product roadmap, and 50% off list price for the lifetime of the engagement. In exchange, we ask for honest feedback.
Spots remain in the Design Partner program for California litigation firms. One email is all it takes.
Apply for early access →