How It Works: A Plain English Guide

LegalDocs-Assist is a drafting and analysis platform for UK fee-earners working in three areas: RTA credit hire, housing disrepair, and clinical negligence. You feed it the letters and documents that come across your desk; it reads them, analyses the legal arguments inside, and helps you draft the reply.

Think of it as a very well-read assistant that has read 133 UK credit hire and motor claims authorities, knows what CPR Part 32 requires of a witness statement, and never gets tired at 5pm. It will not, however, sign your letters for you. Every output is a draft for a fee-earner to review.

Every AI feature in LegalDocs-Assist runs on the same model: Google's Gemini 3 Flash (preview), accessed through an enterprise gateway. We deliberately chose a single model rather than a mix, so that the safety controls described below apply uniformly to every AI call the system makes, there is no second-tier model running on a quieter code path with weaker controls.

The enterprise gateway terms prohibit Google from using your inputs or our outputs to train their models. That covenant sits above our own commitment not to train on your data.

This is the single biggest risk of using AI in a legal setting. The model is perfectly capable of inventing a plausible-looking case name and citation that does not exist. We treat that as the central engineering problem of the platform, and have built three independent layers of protection that every AI output passes through. The same three layers run on every AI feature in the product, not just selected ones.

For the text‑based tools, your source documents stay on your computer. When you upload a text‑based PDF or Word document into the Insurer Letter Analyser, the Witness Statement Generator, the Liability Assessment, or any of the other text‑based tools, the file is read and converted to text inside your browser. The original file itself does not leave your device; only the extracted text is sent for analysis. Even then, the text goes through our privacy filter first (the next section explains how).

There are two documented exceptions. Scanned or image‑only PDFs are briefly uploaded so we can OCR them on the server, then deleted within the hour. Medical pagination uploads the PDF itself; it is dealt with separately further down so you can see exactly what happens to it.

The AI drafting itself happens on our server, under Zero Data Retention terms with the AI provider, and the generated output is deleted shortly after it is returned to you (see "How long is my data kept" below).

Before any text reaches the AI, it passes through our PII masking gateway. "PII" stands for Personally Identifiable Information, the kind of detail that identifies a real person and that you would expect to be handled carefully.

The gateway looks for personal details and replaces them with neutral placeholders before the text reaches Google's model. So instead of "Mrs Lisa Chen", the AI sees "[NAME_REDACTED]". Instead of "john.smith@example.com", the AI sees "[EMAIL_REDACTED]". The AI does its analysis on the cleaned text, and the placeholders are reinserted in your output so the final draft reads naturally.

No. We do not use your data to train AI models, and the enterprise contract we hold with Google prohibits Google from using it either. The AI provider operates under the Lovable AI Gateway's Zero Data Retention terms, in force since March 2026: your prompts and the model's responses are not retained by the provider and are not used to train any model. Your correspondence and case details are used for one purpose only: to generate the specific output you asked for.

LegalDocs Assist is designed not to retain personal case data. Uploaded source documents and the outputs the tool generates — witness statements, paginated bundles, chronologies, timelines, advice, correspondence — are deleted shortly after they are produced or on session close. A scheduled purge runs continuously as a backstop, so nothing is ever retained beyond 24 hours. Firms are expected to download or save their own copy of any output they need, and to re‑run the tool if they need it again. Medical‑bundle pagination files are deleted within about an hour of completion. Inactive login sessions expire on their own. The only data retained is the minimal account record needed to run the subscription. Activity and audit logs are retained as described in the Privacy Policy.

Uploaded files are never included in any backup. The database has standard disaster‑recovery backups, but because case content is deleted within 24 hours those backups hold essentially no personal case data, and once a record is deleted there is nothing to restore it from.

Special category (health) data. Medical records are core to clinical‑negligence and personal‑injury work, so the Service routinely processes special‑category health data. The safeguards are browser‑side processing of source documents, server‑side PII masking, encryption in transit and at rest, and immediate deletion within 24 hours. As data controller, your firm is responsible for the Article 9 lawful basis.

Deletion is immediate and permanent. When you delete a record or document, it leaves the live database and, if it has files attached, those leave storage too. Because storage is not backed up, there is no copy to restore from.

If your firm leaves the platform, we permanently delete your firm's entire tenant — all case content, all documents and bundles, and the user accounts that go with it. The admin path that does this is hardened and is executed only by CaseFlow Automation Ltd platform administrators: a firm has to be archived first (a reversible step that gives you a thirty (30) day grace period during which the firm can be reinstated), then permanently deleted from the archived list as a second gate. Firms cannot self-delete. Where the trigger is contract termination, the permanent delete happens within thirty (30) days of termination. We record the fact of the deletion in an audit log — who, when, and the counts — but never the content itself. The only thing we keep is the minimal account and billing record the law requires us to keep, with personal data stripped out where we can. This is a contractual commitment we honour operationally, not an automated platform feature.

UK GDPR gives you the usual rights — access, rectification, erasure, restriction, portability and objection.

For data about your firm's clients and claimants, your firm is the data controller and we are the processor. So requests from those individuals go to your firm; we help you fulfil them, including by deleting or exporting the records on your instruction.

For data we hold about you as a user of the platform — your login, your usage logs — you can contact us directly at info@caseflowautomation.co.uk. You can also complain to the Information Commissioner's Office at any time.

No. Each Client organisation on LegalDocs-Assist operates in its own walled-off space. Your firm's correspondence, drafts, witness statements, history, and templates are invisible to every other firm on the platform. This is enforced at the database level by row-level security rules that sit underneath the application, it is not a setting in a configuration screen that could be accidentally turned off.

UK GDPR requires organisations to collect the minimum personal data necessary, to protect it properly, and to be transparent about what they do with it. We are registered with the Information Commissioner's Office under registration number ZC013423 and operate under UK GDPR. Here's how the design of the product lines up against those duties:

No. LegalDocs-Assist is decision support. It helps a fee-earner draft and analyse faster, working from a structured knowledge base; it does not replace the judgement of the qualified solicitor who signs the letter. Every AI output in the product is labelled as a draft requiring review, and outputs are never auto-sent.

We make that point clear throughout the platform: you will see it on first login, on every page that invokes the AI, and alongside every generated response.