84+ capabilities, in eight groups.
This is the unabridged list — the agentic work, the document intelligence, the money, and the operational spine of a personal-injury firm, in one platform at one price. Every item below ships in the PI edition today. What isn’t live is named plainly at the foot of the page. For the AI-forward tour, start on the PI overview; for the whole platform across editions, see the capability showcase.
Documents & drafting
Author once, review, and send — without leaving the matter.
- Multi-pass demand generation — watch Justine draft: dossier → analysis → strategy → narrative → compose → QA flags → exhibit index
- Synchronous drafting for other letter templates
- Draft persistence and full generation history
- Document library — upload, view, organize
- Attorney-review workflow — submit → return for revisions with tasks
- Editing, finalize, and threaded comments
- Firm letter templates
- Send via Outlook (Microsoft 365)
- Cloud import / export — Microsoft 365 + Dropbox
Medical & records
Read the record, build the chronology, price the damages.
- Medical-record analysis — summary, chronology, expenses, gaps
- CPT / ICD codes view, with overrides and a codes-reviewed attestation
- Structured re-extraction on demand
- Quick Analyze — drop a record and auto-build a case
- Medical Damages Matrix — provider × payment source
Case strategy & intelligence
The reasoning surfaced, with its evidence and its confidence.
- Matter Intelligence panel — jurisdiction, venue, coverage, liens, firm-carrier history, opportunity + risk flags
- Value Simulator — drag the drivers, watch the range move
- Net-to-Client Waterfall
- Case strategy and stage brief
- Explainable-AI surfaces — deliberation trace, counterfactuals, factor cards, confidence, anomaly, field provenance
- AI Activity feed
- Streams that refuse to lie — if a reasoning run is interrupted, the surface says it was interrupted and keeps the partial work, rather than rendering a half-written answer as a finished one
- Six of the nine named agents run for a personal-injury firm, and three of those act — the docket sentinel, inbound mail and correspondence. The other three propose and log in shadow. We publish the split rather than the headline number
- OCR that returns “not configured” instead of faking a read. Where a record cannot be read, the product says so rather than producing text nobody wrote
- Outcome statistics — settlement-value distribution and case-outcome statistics from firm + venue history (heuristic, not a trained ML model)
- Ask over your documents — semantic RAG search across the matter’s records and filings, answers with citations
- Similar-case leads — AI-suggested starting points for your own research (unverified; not precedent)
Case management
The whole matter, composed and auditable.
- Full matter CRUD with a status pipeline
- Composed case timeline — activities, chronology, billing, and docs, with treatment-gap flags and citations
- Tamper-evident case audit trail — an HMAC-SHA256 hash chain, each record carrying its sequence number and the hash of the one before it, written through a per-tenant serialised writer and verified for integrity nightly. A record cannot be altered or removed without breaking the chain
- Per-case access control with a real grant screen — who can see this matter, granted and revoked in the workspace rather than by a support ticket
- A matter-visibility switch that shows live assignment coverage BEFORE an administrator flips it — how many matters would become invisible to whom, computed against the firm’s own data, so the decision is made with the consequence on screen
- Case assignment — matters assigned to the people who work them, which is what the visibility switch reads
- Typed case notes — 13 types, pin and search
- Contacts
- Medical treatment ledger
- Expenses & costs
- Liens & subrogation
- Litigation-funding loans
- Discovery register
- Insurance policies
- Settlement-negotiations log
- Appointments
- Pleadings
- Case vitals
- Phases
- Checklists
Deadlines & compliance
The clock, watched — and the packet, prepared.
- Firm-wide and per-case deadlines
- Deadline-reminder automation — 7-day, 24-hour, 1-hour, and overdue emails
- Rules-based California docketing — 19 triggers and 42 rules, statute-cited and advisory, and it states its own limits on screen: calendar days, no court-holiday calendar, and §1013 service extensions described rather than applied
- Deterministic statute-of-limitations tracking across 51 jurisdictions — the 50 states and DC. Read that as coverage, not as verification: four entries are attorney-verified and all four are California, and every entry carries its own flag saying which it is
- Tasks with stats
- E-filing packet prep — checklist and tracker (preparation, not direct submission)
Intake & clients
From first contact to open matter, structured and conflict-checked.
- AI client intake chat — public, empathetic, one question at a time
- AI attorney intake copilot
- Intake → structured-case extraction
- Three specialist document extractors — Coverage Analyst, Intake Specialist, Investigator — field-level, with source-doc + page citations
- Firm-branded intake — configured per firm, submissions structured into the inbox
- Submissions inbox
- Leads CRM
- Conflict check
- Shared inbox
- Client email — real messages with a per-case thread
- Client texting (SMS) — opt-in, STOP-compliant, with an attorney-alert number-privacy relay (Azure Communication Services)
Trust & settlement money
PI runs on trust dollars — handled to the ledger line.
- IOLTA / trust accounts — last four only
- Immutable trust ledger
- IOLTA three-way reconciliation
- Settlement disbursement statement — gross → fee → costs → liens → net, as PDF
- Disbursement and lien math
Firm ops & insight
The command center for the whole book of business.
- Dashboard command center with AI insights
- Morning Rounds — a deterministic server-side briefing with no model involved. It reviews an unordered 40 matters, not the whole book, and we cap the claim there rather than rounding it up to “your caseload”
- Portfolio risk radar across the book. One honest note on it: the carrier settlement history filters to ACTIVE matters before averaging, so it is a view of the live book rather than of the firm’s complete history
- An approvals inbox that records what Justine DECLINED to do — as its own titled section, not as silence — and distinguishes “we could not ask” from “nothing is waiting”. A system that shows you only its successes is not something you can supervise
- Outbound send is opt-in per firm and off by default. Approving a draft sends nothing until the firm has switched sending on; the machine cannot email your client on its own initiative
- Web push to the attorney — the subscribe toggle fetches the server key before it touches the browser, so it cannot flip on for a channel that could never deliver
- Reports with CSV export
- Firm analytics
- Member management and invites
- Communication templates
- ⌘K spotlight
- Secure multi-tenant isolation (row-level security)
- CCPA / CPRA privacy suite — know, export, delete, correct, with a DSAR ledger
- Full audit trail with WORM archival
- Feature flags
- Impersonation (audited)
Named plainly, so you can trust the rest.
We label roadmap and not-built capabilities as exactly that — they are never presented as shipped. The following are not live today:
- Direct court e-filing submission — Justine prepares the packet and tracks it; she does not file
- In-app e-signature, and the bring-your-own DocuSign connection beside it. The signing envelope, its hash-chained audit trail and the Certificate of Completion are built — but no mail provider is wired, so no signer is ever reached, and there is no signing page for one to land on. Nothing in the workspace can start or finish a signature today. It is close, and it is still on the roadmap: we count a finished thing you cannot reach as nothing
- Fax
- Phone intake / telephony — client intake is a live web chat, not a phone line (on the roadmap)
- Filevine integration — on the roadmap; the turnkey approved integrations today are Microsoft 365 and Dropbox
- Client case-portal — on the roadmap. Separately, the public intake portal page is not currently resolving a firm end to end and is being reconnected; the intake chat and the firm-side intake inbox are unaffected
- Clio and other third-party case-management connectors
- Bank auto-feed for trust reconciliation
- Single sign-on for firms (OIDC / SAML) — not available today; the backend has a flow but no sign-in or settings screen in either app
- The four other practice-area editions (MT, MM, IB, EL) — only PI and WC (applicant-side; CA, GA, TX) are calibrated today
JustineAI™ PI is legal-workflow software, is not a law firm, and its outputs are not legal advice.
A true tool-calling agent. Not a chat box.
Justine is an autonomous Digital Employee. She plans, reasons, calls the firm’s tools, and produces cited work product across every matter — powered by the Eve-Fusion compound reasoning model. The attorney reviews and decides; Justine does the work between the decisions.
- 01 · Plans
She breaks the work down.
Given a matter or a firm-wide question, Justine plans the steps — which records to read, which authorities to check, which figures to compute — before she starts.
- 02 · Reasons
She thinks it through — in view.
Powered by the Eve-Fusion compound reasoning model, Justine reasons over the matter’s records — chronology, medicals, coverage, statute — citing what she drew on, and streams her thinking as she goes.
- 03 · Calls tools
She uses the tools of the firm.
She calls the real tools she needs — document analyzers, citation verification against CourtListener, valuation and settlement math, the trust ledger, the deadline engine — and reads the results back into her reasoning.
- 04 · Cites & acts
She produces cited work product.
She drafts the demand letter, assembles the filing packet, prepares the disbursement statement, and answers with the source behind every claim — for the attorney to review and decide.
Twelve real legal tools.
These aren’t canned answers. Justine invokes the same tools your firm relies on, against your own data, and reads the results back into her reasoning before she answers.
- Statute of limitations
Computes the deterministic SOL clock for the matter.
- Case valuation
Models special, general, and net figures against comparables.
- Medical chronology
Builds an incident-anchored treatment timeline.
- Analyze medicals
Reads each record for causation, gaps, and expenses.
- Verify citation
Confirms case law against CourtListener — fail-closed.
- Conflict check
Screens parties across the firm before engagement.
- Firm-wide deadlines
Surfaces every upcoming deadline across the book.
- Portfolio summary
Rolls up matter status and risk across the firm.
- Financial summary
Reads trust balances and matter economics.
- Find client
Locates a client across every matter they touch.
- Search matters
Queries the whole book of business by any signal.
- Search case documents
Semantic RAG retrieval — ask a question, get the answer from the matter’s own records and filings, with the source.
- Ask Justine, firm-wide — she plans, then answers across every matter with verified citations
- The War Room — gathers a matter’s record and synthesizes chronology, medicals, valuation, and statute
- Auto-drafts verified demand letters end to end
- Analyzes medical records and deposition transcripts and reports what it finds
Ask her anything. Watch her reason to the answer.
Ask Justine works on one matter or across the whole book of business — no case need be open. She plans, calls the firm’s real tools, reasons over what they return, and answers with the source behind every claim. Pick a question and press play.
“What needs me this month across the whole firm — deadlines and any trust shortfall?”
- Plan
Two questions in one. I’ll pull every upcoming deadline across the book, then reconcile each matter’s trust balance and flag anything short.
- Calls tool
firm-wide upcoming deadlines()
Scanned every open matter — 4 deadlines fall inside 30 days; one is a statute-of-limitations date.
- Calls tool
statute of limitations()
Confirmed the Ramirez SOL runs in 19 days — deterministic, statute-cited.
- Calls tool
financial summary()
Read trust balances across matters and ran three-way reconciliation on each.
- Reasons
The Ramirez SOL is the hard one — it gates the filing packet. One trust ledger reconciles clean to the penny; another is missing a recorded lien payoff, so it will not tie out until that entry posts.
- Cites
Four deadlines in 30 days — Ramirez SOL in 19 days is the priority. Trust reconciles across matters except Okafor, where a $4,200 lien disbursement is un-posted; I flagged the ledger entry. Every figure links to its matter and ledger line for your review.
Illustrative of the work Justine surfaces on a real file. Every answer is attorney-reviewed work product; nothing leaves without your sign-off.
See both sides of the case before the other side does.
The War Room is a dual-perspective view of the matter. Justine builds your case theory — every liability element and damages figure traced to a source in the record — and, right beside it, simulates the other side: the defense and adjuster’s likely attacks, the weak points they’ll press, and the posture they’ll take. It reasons only from the record on file: every element carries a citation, authorities are checked against CourtListener, and nothing is invented. You read the counter-argument before you’re in the room.
Justine builds the case theory from the record.
- Each liability element, traced to a source fact in the file
- Damages by category, with the figures behind them
- Case-theory strength, element by element
- Supporting authorities, verified against CourtListener
She argues the defense and the adjuster against you, so you hear it here first.
- The attack the defense will make on each element
- The weak points they will press
- The adjuster’s likely posture and opening move
- Where your record is thin — before they find it
Plus a settlement band — low, expected, and high, with the likely first offer and the reasoning behind each number — and every authority in the build is independently re-verified against CourtListener before it stands. Reasoned from the record on file; it is analysis to inform your judgment, not a prediction or a guarantee of outcome.
A score you can see her reason to.
Justine scores a matter’s strength from 0 to 100 and models a settlement valuation — and she shows the factors behind both. It is analysis to inform your judgment, reasoned only from the record on file. Not a prediction, not a guarantee.
Viability score for Ramirez v. Delgado — surfaced in the 06:00 briefing with the full reasoning trail.
- Liability is cleanRear-end collision; the defendant admitted fault in his Day-1 recorded statement.
- Causation is documentedThree treating records tie the L4–L5 herniation back to the collision.
- Damages are supportedSpecial damages of $87K with venue comparables running $180K–$320K.
- One gap to explainA 19-day treatment gap the carrier’s IME will press — rebutted by two PT notes.
- The clock is shortStatute of limitations runs in 21 days; the filing packet is gated on it.
The settlement band.
Low, expected, and high — each number reasoned from the record, with the likely first offer and the logic behind it laid out for your review.
- Low
$180K
if liability is contested and the gap sticks
- Expected
$215K
bottom-line authority on the current record
- High
$320K
top of the venue comparable range
Illustrative of a single matter. Settlement analysis is reasoned from the record on file to inform your judgment — it is not a prediction or a guarantee of outcome.
The reason you can put an agent on a matter.
Agentic AI only belongs in a legal file if it refuses to make things up. Justine is built to ground every claim, drop what she can’t verify, and tell you when she doesn’t know — so the record stays sound and the judgment stays yours.
- 01 · Fail-closed citations
She won’t call a citation good law unless it’s confirmed.
Every authority Justine relies on is checked against CourtListener before she stands behind it. If a citation can’t be confirmed, she drops it rather than assert it. This is the anti-hallucination guarantee — the discipline that lets you put AI in a legal file.
- Citation verification runs fail-closed — unconfirmed means unused
- No invented case law, no invented reporter cites
- The authority behind every argument is traceable and checkable
- 02 · Document intelligence — “Justine’s Read”
She reads every record and refuses to guess.
Justine reads each medical and liability document and builds an incident-anchored treatment chronology. When a record is undated, she drops it rather than guess where it belongs — so the timeline you rely on is the timeline the record actually supports.
- Incident-anchored treatment chronology across every provider
- Undated records are set aside, never placed on a guess
- Causation links, treatment gaps, and expenses surfaced for review
- 03 · Jurisdiction-aware reasoning
California by the book. Elsewhere, she flags — she doesn’t guess.
Justine reasons directly under California’s verified law. For other jurisdictions she flags the state-specific rules that need your confirmation and refuses to guess at law she can’t verify — so a cross-border matter never gets silently mis-reasoned.
- California rules reasoned directly, statute-cited
- Out-of-state rules flagged for attorney confirmation
- No guessing at unverified jurisdictional law
- 04 · Document-grounded answers
Answers over your firm’s own documents — with an honest “not found.”
Ask about the record and Justine answers over your firm’s own documents, with inline citations to the source. When the documents don’t support an answer, she says so rather than filling the gap. Every query runs under hard tenant isolation — your files, only your files.
- Inline citations to the source document on every grounded answer
- Honest “not found” when the record doesn’t support it
- Hard multi-tenant isolation — one firm’s data never touches another’s
- 05 · Autonomous, overnight
A nightly case auditor and a morning briefing that’s already done.
The daily AI case auditor sweeps every open matter overnight — treatment gaps, missed deadlines, stale providers. By 06:00 the AI morning briefing has a prioritized start-of-day read waiting: what changed, what needs you, and what’s due.
- Nightly autonomous sweep of every open matter
- Prioritized morning briefing / rounds at the start of the day
- Deadlines, gaps, and risks surfaced before they become problems
Powered by Eve-Fusion — a compound reasoning model.
Justine reasons on Eve-Legal F5/reasoner, JustineAI’s configuration of the Eve-Fusion™ compound reasoning model — grounded, jurisdiction-aware, and cited back to the matter’s own records. It is a composition of cooperating models, not a single one — and we never name a third-party foundation model on a customer surface.
Reasoning tiers.
Firms are provisioned on a reasoning tier that scales deliberation depth to the work — from Economy to Maximum. It’s an enterprise configuration set with your team, not a dial in the sidebar.
- 01 · Economy
High-volume, routine work — fast and cost-efficient.
- 02 · Balanced
The everyday default for most matters.
- 03 · Performance
Deeper reasoning for contested, high-stakes files.
- 04 · Maximum
The fullest deliberation, for the matters that warrant it.
Three things your firm already pays for. Replaced by one.
Most PI firms pay for three things — a case manager, an AI layer bolted on top, and a demand vendor that bills $275–$800 per letter. JustineAI™ PI replaces all three, with AI bundled — not metered — and demands unlimited.
A case manager
Headcount that tracks matters, deadlines, records, and trust — one seat at a time.
An AI layer, bolted on
A separate tool stitched on top, usually metered by usage or by seat.
A demand vendor
Outsourced demand writing that bills $275–$800 per letter — per matter, every time.
JustineAI™ PI is one Digital Employee across the whole firm — case management, medical-record and deposition analysis, verified demand letters, IOLTA trust accounting, and California deadline automation. AI is bundled into the seat price. Demand letters are unlimited. One platform, one price.
34 headline capabilities. One seat.
What a personal-injury firm runs on — the agentic work, the reasoning, the demands, the trust dollars, the deadlines, and the operational spine — in one platform, at one price. Every item below is shipped in the PI edition today. This is the headline set, not the whole catalogue: the full PI catalogue runs longer, and neither number is tier-gated — every seat gets all of it.
Agentic AI
A Digital Employee that plans, reasons, and acts across the firm.
- Firm-wide agentic Ask Justine
- The War Room — matter gather + synthesis
- Autonomous multi-step tool use
- Attorney-in-the-loop review on every action
Advanced reasoning
Powered by the Eve-Fusion compound reasoning model.
- Case strategy & assessment
- Case strength / viability scoring
- Settlement valuation reasoning
- Jurisdiction-aware, California-grounded reasoning
Demands & analysis
The document work, bundled — not billed per letter.
- Unlimited demand-letter generation
- Medical-record analysis
- Deposition analysis
- Ask over your documents (semantic RAG, cited)
- Verified legal citations (CourtListener)
Documents & drafting
Author, review, and send — in one place.
- Multi-pass demand drafting + generation history
- Attorney-review workflow (submit → revise)
- Demand Package as a real ZIP — cover, exhibit index, PDF + DOCX
- Send via Outlook · import/export (M365, Dropbox)
Intake & clients
From first call to open matter, structured.
- AI client intake chat (public, empathetic)
- AI attorney intake copilot
- Firm-branded public intake portal + inbox
- Leads CRM, conflict check, client email
- Client SMS with attorney-alert privacy relay
PI money
Trust dollars, handled to the ledger line.
- IOLTA trust accounting
- 3-way reconciliation
- Settlement disbursement statements (gross → net, PDF)
Deadlines & filing packets
The clock and the paperwork, automated.
- California rules-based docketing — 19 triggers, 42 rules
- Statute-cited deadline automation
- A statute-of-limitations table covering 51 jurisdictions
- E-filing packet preparation
Run the practice
The operational spine of the firm.
- Conflict checks
- Client intake portal
- Time tracking
- Reporting & analytics
- Secure multi-tenant isolation
See the full capability catalog →
On the roadmap, not yet live — in-app e-signature and the DocuSign connection beside it (the envelope, its hash-chained audit trail and the Certificate of Completion are built, but no mail provider is wired and there is no signing page, so no signer can complete a signature — we count a finished thing you cannot reach as nothing), direct court e-filing submission (Justine prepares the packet today), fax, phone-intake / telephony, Filevine integration, a client case-portal (the firm-branded public intake portal is live and distinct from it), single sign-on, and bank auto-feed for trust reconciliation. The turnkey approved integrations today are Microsoft 365 and Dropbox. We label roadmap items as roadmap; they are never presented as shipped.
The client hears from the firm — in their language.
Justine keeps the client informed by text and email, with a number-privacy relay so attorney alerts never expose a personal cell number — every message attorney-reviewed before it goes out.
- Client texting with a privacy relay
Opt-in, STOP-compliant SMS keeps the client updated on their case, and attorney alerts route through a number-privacy relay so the client never sees a personal cell number. Transmitted over Azure Communication Services on a toll-free line.
- Per-case client email
Real client messages threaded to the matter, sent through Microsoft 365 / Outlook — the correspondence lives with the file, not in a separate inbox.
Not legal advice · not a law firm · ABA-conduct aligned · the attorney decides. Client communications are attorney-reviewed work product.
Questions, answered honestly.
Is JustineAI™ PI a chatbot?
No. Justine is a Digital Employee — an autonomous agent that plans, reasons over the case file, calls the firm’s tools, and produces work product across every matter. A chat box answers a question; Justine does the work, under the attorney’s supervision.
Will the AI make up case citations?
No. Every authority Justine relies on is verified against CourtListener, and she will not assert case law she cannot confirm. Verification is fail-closed — if a citation can’t be confirmed, she drops it rather than assert it. The reasoning is grounded in the record and in verifiable law — no hallucinated citations. As with any work product, the attorney reviews before it goes out.
Can I ask Justine about the whole firm, or just one file?
Both. Ask Justine works on a single matter and firm-wide — with no case open — across every matter, client, deadline, and trust dollar. She’s a true tool-calling agent: she plans, calls real tools like firm-wide deadlines, portfolio and financial summaries, and find-a-client-across-matters, then answers with the source behind each result.
Will Justine answer from our own documents?
Yes. Ask about the record and Justine answers over your firm’s own documents with inline citations to the source. When the documents don’t support an answer, she says “not found” rather than filling the gap. Every query runs under hard multi-tenant isolation — your files, only your files.
What about matters outside California?
Justine reasons directly under California’s verified law today. For other jurisdictions she flags the state-specific rules that need your confirmation and refuses to guess at law she can’t verify — so a cross-border matter is never silently mis-reasoned. Additional jurisdictions are on the roadmap.
Is the e-signature DocuSign?
Neither is live today. Native in-app e-signature (ESIGN / UETA, hash-chained audit trail, Certificate of Completion) is built, but no mail provider is wired and there is no signing page, so no signer can complete a signature yet. The bring-your-own DocuSign connection has no screen in the workspace either — the served bundle does not contain the word DocuSign. Both are on the roadmap; we count a finished thing you cannot reach as nothing. Engagement letters, releases and settlement agreements are drafted and exported on your letterhead for signature out of band.
Can we control how deeply Justine reasons?
Yes — at the firm level. Firms are provisioned on a reasoning tier, from Economy to Maximum, that scales deliberation depth to the work. It’s an enterprise configuration set with your team during onboarding, not a dial in the sidebar.
Are demand letters really unlimited?
Yes. Demand-letter generation is bundled into the seat price — not billed per letter. Generate as many as your matters need. A multi-pass pipeline drafts and flags QA issues for your review; the attorney finalizes. Medical-record analysis and deposition analysis are bundled the same way.
Can Justine e-file directly with the court?
Not yet. Today Justine prepares the complete e-filing packet, ready to submit. Direct court e-filing submission is on the roadmap. In-app e-signature (ESIGN / UETA, with an audit certificate) is built but not live today: no mail provider is wired and there is no signing page yet, so a signer cannot complete a signature — and we count that as nothing until you can reach it.
What jurisdiction does it cover?
California today. The rules-based deadline automation and demand workflow are California-grounded and statute-cited. Additional jurisdictions are on the roadmap.
How is our firm’s data secured?
JustineAI™ PI is multi-tenant with row-level isolation, so each firm’s data is scoped to that firm. It runs keyless on managed identity — no shared secrets — and it does not log prompts or client PII.
Are you SOC 2 certified?
SOC 2 is in progress. We do not claim a certification we do not yet hold. The platform runs on Microsoft Azure, which carries its own attestations at the infrastructure layer; our own report is being pursued and we will say so plainly when it is complete.
The refusals are the product.
The question that decides a legal-AI purchase is not what the system can do. It is what it does when the record will not support an answer. Every item below is a place Justine declines — named, in the product’s own words, with the evidence it would not fabricate. A machine that tells you which of its own inputs it could not stand behind is worth more to a practice than one that always has a number.
An interrupted answer reports itself interrupted.
If a reasoning run is cut short — a deploy, a dropped connection, a stop — the partial work is kept and labelled as partial. It does not render as a finished answer. This is the quietest failure mode in agentic software and the most dangerous one in legal work, because a truncated analysis looks exactly like a complete analysis that found less.
- Statutes of limitations
Fifty-one jurisdictions, and it tells you which four it can stand behind.
The statute-of-limitations table covers fifty states and the District of Columbia, and every entry carries its own verification flag. Entries that have not been attorney-verified are printed as AI-compiled references to confirm against the current statute — not as deadlines. Four entries are attorney-verified, and all four are Californian. Recognising a jurisdiction is not the same as vouching for it, and the table says which is which.
- Reading scanned records
Where it cannot read, it says so.
Optical character recognition runs on Azure Document Intelligence under managed identity. Where that route is not available to a tenant, the read returns an explicit not-configured result and the document is left unread. Nothing is inferred from a page that was never successfully read — which is the only version of this feature that is safe to put a medical record through.
Restraint, not absence.
Tenancy is fail-closed and actually armed in production, not staged behind a flag: Postgres FORCE row-level security with a tenant_isolation policy on every table carrying tenant_id, an application role that is NOBYPASSRLS, and CASE_ACCESS_ENFORCED and ENTITLEMENTS_ENFORCED both on in production. A missing policy on this shape of database does not raise an error — it silently returns nothing, which is why the policy shape is asserted by a build-breaking check rather than trusted.
- Hash-chained audit trail
An audit trail that would show tampering.
Every row carries a sequence number and the HMAC-SHA256 hash of the row before it, written through one serialised writer per tenant so the chain cannot fork, and verified end to end nightly. Editing history means re-deriving every hash after the edit — which is the point of a chain rather than a log.
- Per-case access
Per-case access, granted rather than assumed.
Case access is a grant a person makes, not a role that happens to have read rights. Before an administrator turns matter visibility on, the screen shows how much of the book is actually covered by assignments — so nobody flips it and finds out afterwards.
What shipped in PI, and when — 29 entries covering new capability and corrections an attorney would want to know about. Security, access-control and confidential-data handling changes are deliberately not itemised publicly, and the page says so.
Put one Digital Employee to work across your whole firm.
JustineAI™ PI is live for personal-injury practices in California. Request a 14-day trial — no credit card. AI is bundled, demands are unlimited, and Justine is working matters on day one.
14-day free trial · No credit card required · Self-serve signup
JustineAI™ PI is legal-workflow software, is not a law firm, and its outputs are not legal advice.