The short answer: a law or accounting practice can use AI aggressively — for contract summarization, deadline tracking, matter briefs, and document search — provided one architectural condition holds: inference happens inside the office. A private AI appliance satisfies that condition by design; a cloud chatbot cannot.
The problem with "just use ChatGPT" at a firm
Every professional-responsibility conversation about AI lands in the same place. The tools are impressive; the data path is disqualifying. Pasting a client's agreement into a cloud service means the document leaves your custody, transits systems you cannot audit, and sits under terms that change without your consent. For a profession built on privilege and confidentiality, that isn't a workflow adjustment — it's a category problem.
The fix isn't abstinence. It's moving the model to the documents instead of the documents to the model.
What a private appliance actually does for a small firm
Contract and document summarization. Hand it a 90-page agreement and get the terms, obligations, and unusual clauses in minutes — with the exact passages quoted, so verification takes seconds. Document stays in building unless enabled.
Version comparison. "What changed between the March and May drafts?" answered precisely, including the changes opposing counsel didn't mention in the cover email.
Matter and client briefs. Before a client meeting: the last three interactions, open items in both directions, and what changed since. Assembled from your own files and correspondence.
Deadline vigilance. Every date in every engagement letter and court document, tracked, with reminders that escalate. The malpractice-insurance argument for this one writes itself.
Institutional search. "Find the engagement where we handled a spin-off like this one" — decades of firm knowledge, searchable in seconds by the newest associate.
The architecture, in one paragraph
Digital Twin Pro is a small appliance that sits in your office like the router does. Client folders connect to it locally; the assistant (Hermes) reads and reasons on the device itself. Nothing is uploaded, nothing is subscribed to, and per-user access profiles keep the bookkeeper in billing folders, not case files. Every action lands in a local audit log you can export, which is the answer to the partner who asks "and how would we prove what it accessed?"
What it costs, in firm terms
$1,699 once — less than most firms bill in a day — with no required subscription. Your data stays on the appliance; you can copy backups to your own drive whenever you like, and a one-time service can set that up with you. Compared to per‑seat AI subscriptions for a five‑person office — plus the potential cost of a confidentiality breach — the appliance is far cheaper.
What it will not do
It won't file documents with a court, sign anything, or replace judgment — drafts are drafts until a professional reviews them. It also won't stop you from routing a task to an external model if you explicitly choose to; it simply makes "local" the default and the exception visible. Honest boundaries are part of the confidentiality story.
The full workflow list: Private AI for law, accounting & finance — 30 use cases →
Updated July 15, 2026.
Digital Twin Pro — the private AI appliance. $1,699 once, no required subscription, 15-day evaluation return window. See the appliance →