The whole matter, handled in modules.

Paste the matter facts (client name, scope of work, fee basis) and receive a full engagement letter skeleton built on UK templates: scope, fees, complaints procedure, data-handling notice, and termination clause, ready for partner review and sign-off before it goes to the client.

On day one the firm receives a ready-to-review compliance pack: UK GDPR Art.13/14 client privacy notice, data-handling policy, DSAR response workflow, DPIA, transfer-impact assessment and a draft annual AI-use report, mapped to the SRA Code of Conduct and ICO guidance. Every template opens inside the app with your firm's details already filled in; edit the placeholders and download each as Word or PDF, drafts for partner sign-off, not documents written from scratch.

Drag a whole folder onto the matter (sub-folders and all) or the zip a client emailed, and every file inside is read where it can be. Thirty-one formats are recognised: PDFs including photocopies and phone scans, Word, Outlook .msg exports, .eml and mbox archives, Excel and OpenDocument, RTF, XML, web pages, iCalendar and vCard, photographs, and WhatsApp chat exports rendered as a dated conversation with each message attributed to who sent it. The nine formats a browser genuinely cannot read.doc, HEIC iPhone photos, scanner TIFFs, Outlook.pst files, audio and video among them, say so by name and tell you exactly what to do instead. No file is ever silently ignored, and an email whose substance sits in its attachment says the attachment was not read. From what is read you receive a structured matter card, parties, key dates including any live deadline, claim type and the facts themselves, with each fact carrying the passage it came from, and that passage is checked against the document: one genuinely in the file is quoted, one that is not is labelled as worked out rather than copied, so an invented quotation cannot pass for a real one. The matter's jurisdiction is not taken from the model, it is detected from the text's own signals and chosen by you, within the markets your firm is licensed for. Every fact is a proposal a fee-earner accepts, corrects or rejects before it is written to the matter file. After the matter is open it keeps working: a document added to it is read for facts the moment it lands, and the proposals wait in the matter's Proposals tab. A document identified as written by an AI assistant is never allowed to propose a fact about your case.

Client files remain in the firm's own encrypted vault; the privilege gate classifies Legal Professional Privilege status before any content is sent to the AI, shows exactly what would be sent to the AI provider, and blocks egress unless the solicitor explicitly approves it for that document, that prompt, that moment, satisfying the SRA Code of Conduct duty of confidentiality by architecture rather than policy.

Click "prepare case" on any matter and receive a typed case-graph, parties and their roles, claims and counter-claims, evidence items linked to the documents on file, authorities cited or proposed, and the date spine from first instruction to next deadline, as a draft skeleton ready for the fee-earner to review and edit rather than rebuild from scratch.

Name the counterparty on a matter and the card builds by itself. On a UK matter it queries Companies House for the company profile and the officer list, carrying each officer's appointment and resignation dates exactly as the register records them. That needs your firm's own Companies House credential. Give it the company number and it reads that company; give it only a name and it returns the closest matches, labelled as a name search rather than as a confirmed identification. Where the register reports more entries than came back, the card marks that section incomplete rather than showing a short list as the whole record. The Gazette, and the SRA Register, where you have named the solicitor on the other side, publish no machine interface, so the card deep-links them and marks them "check by hand" instead of counting them as checked. Every line names the register it came from, links to it, and records when that register was read, so you can verify any of it at source. Where a register could not be reached, an outage, or a credential the firm has not connected, the card states that as a gap and counts how many there are. It never hands back a quiet empty card that could be read as nothing adverse found.

Before any text is sent to the AI, the pseudonymisation pass replaces recognised names, National Insurance numbers, NHS numbers, passport numbers, UK postcodes, phone numbers, bank account details and other supported identifiers with stable placeholders, so the AI works on the matter with those identifiers removed. It is pattern-based pseudonymisation with documented gaps, not anonymisation, supporting UK GDPR Art.5(1)(c) data-minimisation duties without any extra step from the fee-earner.

Click the compliance check on a matter and receive a posture report against the SRA Code of Conduct, which Principles touch the work done, where the firm's published AI-use position is consistent with actual matter activity, where a COLP would want a note, plus a draft Annual AI-Use Report built from real matter-activity logs rather than a template filled in from memory. On the attended registry path, the 10 consequential registered skills, 8 drafting, 1 filing and 1 signing, pause for confirm, reject or modify; an unattended request that requires review is refused. Direct and unregistered paths are outside that guarantee.

Run the adversary pass on any draft, a particulars of claim, a defence, a transaction advice, and receive a structured critique across thirteen angles: procedure (CPR service and form requirements), evidence (burden and admissibility), limitation (Limitation Act 1980 periods), quantum, legal authority, counter-claim, Part 20 claims, disclosure (CPR 31), costs exposure (Precedent H / Part 36 / CPR 44-47), regulatory exposure, ethics, commercial pressure and commercial optics, with each finding anchored to the relevant statute, rule or authority.

Each recognised neutral citation handled by the England and Wales matter route is checked live against The National Archives’ Find Case Law service. A match confirms only that the citation appears in that service; it does not confirm that the case supports the proposition, remains good law or has not been appealed. Coverage varies, so a non-match is reported as “not found in Find Case Law” and requires solicitor review. Contains information licensed under the Open Justice - Licence v2.0. A case the register does not hold blocks the export as a possible fabrication, and a check that could not run blocks it too, the gate fails closed rather than waving anything through. Statutes are handled the other way, by deep-link to legislation.gov.uk for you to confirm, because a case-law register cannot answer for an Act. An authority the AI left unflagged, or one you have not yet confirmed, holds the document at an amber-warn gate where you choose Cancel or Export Anyway, and the decision is logged to the matter ledger as evidence of the firm's citation due-diligence duty post-Ayinde. Find Case Law holds most courts from 2001 and is not every judgment ever handed down. A miss can mean outside coverage, a misstated citation, or a fabrication, the product says "not found", never "does not exist". Contains information licensed under the Open Justice - Licence v2.0 The check is enforced at export; the confirmation against the live source is the solicitor's, which is exactly the human verification the court in Ayinde required.

Every matter carries a persisted risk register. Deadlines are kept in four separate groups, limitation, procedural, contractual and regulatory, because they are different kinds of date, and one undifferentiated list leaves you to pick the cut-offs out of the hearings and the invoices. The grouping says which family a recorded date belongs to; it does not say what follows if one is missed. Findings sit on three axes: litigation exposure, regulatory exposure and evidential exposure. Every line names the field on the matter it was read from, the statutory basis the record carries, or says plainly that the record carries none, rather than guessing one, and the jurisdiction whose rules produced it. It recomputes as the matter changes. And when an input is missing or a calculation fails, it reports no risk level at all and tells you what is missing: it will not show a clean score for a matter nobody has read.

AI decisions on a matter are recorded to an append-only ledger, and so are the human decisions taken at the product's wired controls, accepted, rejected, edited, signed off, overridden, and the approval or refusal of an external action such as sending a document for signature. Recording is best-effort so it can never block a solicitor's work, and every export names the writes that failed rather than implying none did. The input is fingerprinted (a one-way hash over the whole input, never stored in the clear, so privileged text cannot leak into the export), the output kept as an excerpt, each row HMAC-signed and daily hash-chained, then exported in one click as a signed pack, in your own market's language, for SRA, ICO or in-house COLP review. HMAC authentication and chain links make recorded events tamper-evident when checked with the firm-held key. The export alone is not independently verifiable proof, and the ledger does not establish that an unrecorded event did not occur or that the firm met an obligation.

Four views of the same matter without additional data entry: a four-layer calendar (court dates, statutory deadlines, client meetings and internal actions, drag to reschedule with a confirm step), a firm-caseload Kanban (drag a matter card to flip its status, persisted), a litigation timeline (every dated event on one horizontal axis, colour-coded by source, with Gantt range bars for events spanning a date range), and a case-as-graph view (parties, witnesses, counsel, documents and evidence drawn as nodes with edges, click any node to highlight its connections across the matter).

For UK firms the client-account view states the SRA Accounts Rules Rule 8.3 five-week interval calculated from the reconciliations saved on this screen, never as the firm's complete record or a breach finding, compares the three balances with a live discrepancy classifier and exports a sign-off-details PDF for the COFA or a manager to sign. The invoice-age view groups the chaseable invoices it has read successfully by days since issue date (up to 30, 31-60, 61-90, over 90), per currency, each at its full recorded total: part-payments are not deducted, so no figure is a remaining balance or a finding that an invoice is overdue. A provider that fails, a matter left unread or a list cut off at its page limit is named on the card, and the affected totals are shown as a floor. Invoicing creates a UK invoice PDF and a structured JSON sidecar in the browser, with no remote invoicing API, for the firm's accountant to review or transform for its accounting workflow; it is not certified, is not sent to HMRC and is not checked for full compliance with VAT invoicing rules, and no direct import into, connection to, or submission through any accounting or MTD service is implemented or proved. Settlement shows a historical base rate over the firm's precedent bank, the pseudonymised outcomes of closed matters of the same type and jurisdiction, only when at least three of them carry a recorded final outcome; it does not count case law and is not a prediction for the matter in hand. On a matter's Billing tab the AI activity figure is labelled a modelled manual-time difference, an assumption and not measured savings. The per-fee-earner dashboard keeps measured figures (time actually recorded, drafts, AI invocations) apart from fixed, unvalidated modelled time and utilisation assumptions, and the modelled figures must not be quoted as savings.

Open a matter, pick a document, and receive a fully-drafted starting point in one step: eleven HMCTS civil forms (N1, N9, N9A, N9B, N180, N181, N215, N225, N228, N244, N265) pre-populated from matter facts, nine ship with the blank PDF bundled; the default-judgment pair N225 and N228 pre-fill the same way once you upload the blank, which is free from the court service, and eight pleading drafters chosen from the matter screen, Particulars of Claim, Defence, Counterclaim, Defence and Counterclaim, Reply, Witness Statement, Statement of Costs N260 and Skeleton Argument. Eighteen workflows are offered on a UK matter, chaining documents into common matter shapes (LBA, Part 36 offer, default judgment N225/N228, N244 application, disclosure N265, mediation brief, freezing order, trial prep). Litigant-in-person mode carries through to the rendered documents as well as the AI-drafted ones: party-in-person signature framing, no SRA identity line, and the CPR 46.5 / PD 46 costs-cap basis on Precedent H instead of solicitor Guideline Hourly Rates. E-bundle preparation is in CPR PD 39A §3.2 shape. The contract library holds twenty-six templates across the five markets, nine of them English-law, NDA, Services Agreement, Employment Contract, Settlement Agreement, Witness Statement, Part 36 offer, Precedent H, Letter Before Action and CMC directions, each with its statutory anchors cited inline, and a filled template saves onto the matter so it goes through the same review, citation gate, signing and Word export as any draft.

The agent fleet surfaces inside Microsoft Word, Outlook and Excel, against the matter you pick. The Word ribbon drafts, stress-tests the wording you have selected, pulls counterparty intel, and files the open document to your document system. The Outlook task-pane, which you open on the message in front of you; it does not watch your mailbox, suggests which matter an email or calendar entry belongs to, files it to correspondence, drafts a matter-anchored reply, and reads dates out of the text for you to confirm before any of them reach the matter timeline. The Excel add-in checks a column of counterparty identifiers against that market's own company register and, if you tick the box, screens each name against the sanctions list your firm has imported, results written beside your selection. Companies House (with your firm's own credential) and the Polish KRS and Biała Lista VAT answer automatically; the German, Dutch and Spanish company registers are not connected, and a row for one of them says so and links you to the register rather than leaving a blank that reads as a clean check. A row that could not be checked is never written as one that was.

When a firm opts in, off by default, with an explicit confirmation by the firm, the journal records the AI actions connected to it: the five case-preparer dispatches and the solicitor's verdict on a draft, with bounded input and output summaries, routing and redaction count. Direct drafting, evidence-analysis and review-grid clients are outside this journal. Text fields are pseudonymised before the write and an entry is dropped if that step fails; this is not anonymisation, and detection has gaps. On an installed Enterprise profile the journal is held in the firm's controlled vault, reached over authenticated HTTPS. Journalling uploads no entries to an AI provider or a central learner. The confirmation is the firm's product choice, not consent from the people in its files. The disclosure pack is an editable template the firm adapts to its actual hosting, access and processing and adopts within its own governance.

The complete matter record is kept in an encrypted vault controlled by the solicitor or firm, not in a vendor database or browser cache. Exact originals, verified facts, deadlines, audit, document memory and firm knowledge use one attributable record; every case view is rebuilt from it. A fresh browser reconnects to the same authority, and recovery is reported only after an independent restore is verified.