The governed drafting harness for small firms
AI drafting your regulator would recognise.
Built for small litigation and personal-injury firms in Ireland and the UK - up to about twenty fee-earners.
Attendance notes, chronologies and disclosure lists, with the software and the record on your own server. Each skill is instructed to mark what it cannot confirm, every completed run is recorded, and no draft is released as a document until a named fee-earner has reviewed it.
What the regulators have said
“Appropriate safeguards can include ... setting up AI systems to run locally or within a secured environment controlled by the firm.”
“The use of AI does not diminish or transfer your professional responsibilities. You remain accountable for all work and advice delivered to clients, whether or not AI has been used in its preparation.”
Solicitors Regulation Authority, Misuse of AI - Warning notice
“Solicitors should refrain from providing confidential information to an AI that is not directly managed by their firm since this is at risk of being leaked.”
Law Society of Northern Ireland, Guidance on the Responsible Use of AI
“Any information that you input into a public AI chatbot should be seen as being published to all the world.”
Courts and Tribunals Judiciary, AI Guidance for Judicial Office Holders (England and Wales)
In the Law Society of Ireland's 2025 survey, 45% of solicitors said they never use AI. The rest do. Every regulator in Ireland and the UK has now said what that means. The answer isn't to ban it. It's to run it where your files already live, with a gate in front of anything that leaves.
How it works
Step 1
Installed inside your firm
One Windows machine in the office runs AI-Switchboard; everyone else just opens a browser. Nothing installed on desks, nothing to migrate. Works alongside Clio, LEAP, Osprey, Quill, or no practice management system at all. You hold your own Anthropic account; the software never sees your key.
Step 2
Drafts from your firm's material, with a reviewer in control
Attendance notes, case chronologies, engagement letters and disclosure lists. The drafting skills tell the model to mark points the supplied material cannot confirm as TO CONFIRM, and the Word export highlights those markers. The named reviewer checks the whole draft, including facts and citations, before release. A machine check can fail a run when its defined checks do not pass; it does not replace legal review.
Step 3
Releases like a compliance officer
Every run is recorded: what it read, what it wrote, who approved it. Review is on by default. The Word export refuses until a named fee-earner marks the draft reviewed, and that name is stamped into the released document.
Against the alternatives
| ChatGPT or Copilot | PMS AI (Clio Duo, LEAP) | AI-Switchboard | |
|---|---|---|---|
| Client data location | Provider cloud | Vendor cloud | Your hardware - only the text of each drafting request goes to Anthropic, under your firm's own account |
| Works with your current PMS | — | Only theirs, top tier | Any, or none |
| Unverified facts flagged | No | Partially | Marked by instruction, highlighted for review |
| Review before release | No | No | Enforced by default, reviewer named |
| Audit trail | No | Basic | Every completed run, user named |
| Requires an Azure tenancy or IT function | No | Vendor-hosted | No. One installer. |
| Price | £16–25 (UK) | £117–174+ with PMS (UK) | €115 / £99 per fee-earner, support staff free, plus your own Anthropic usage |
In Ireland, the EU AI Act now applies to your firm.
AI-literacy duties have been live since February 2025 and most obligations apply from August 2026. Drafting tools like this are not high-risk, but you need the policy, the training record, and the reasoning on file. The pack ships with all three, plus a pre-filled DPIA and a register entry for your own records. €115 per fee-earner. Compliance is in the box.
Questions a compliance officer asks
- Is this approved by the Law Society?
- No. The Law Society of Ireland does not approve this software. The trust page explains how its main controls relate to the Society's guidance. Your firm still has to assess the tool, its provider and each use of client information, and remain responsible for the work.
- Does it do legal research or find case law?
- No. AI-Switchboard is designed to draft from material your firm supplies, not to search for authorities or verify case law. That boundary reduces one source of error, but an AI draft can still contain a wrong or invented citation. A named fee-earner must check every citation, fact and legal assertion against the source before release.
- Do my staff need their own AI accounts?
- Your firm needs one Anthropic account, which you hold and pay for directly. Staff sign in to AI-Switchboard with named accounts on your own server; nobody needs a personal AI subscription, and no one uses a public chatbot. Expect roughly €18–35 (£15–30) a month in usage for a heavy user.
- Why don't you hold the AI account for us?
- Because then we would sit in your data chain as a processor, with processor terms, a sub-processor list, and liability for another company's failures. Holding your own key keeps the chain short: your firm, and Anthropic under its commercial terms. We ship the controller-side paperwork so it isn't a project for you.
- What if it's wrong?
- The drafting skills instruct the model to mark points it cannot confirm as TO CONFIRM, and the Word export highlights those markers. A missed marker or convincing wording is not proof that a statement is right. A named fee-earner checks the full draft before release; the export refuses until that review is recorded.
- Where's my data stored?
- In your office. The client file, the drafts, the audit database and the transcripts stay on your hardware. The only thing that crosses the boundary is the text of one drafting request and the model's reply.
Three design-partner slots.
See it on your own matters. Book a 30-minute discovery call, or email Paul directly - pilots run for sixty days on live files, with a named reviewer on every draft.