Enable
AI Setup
ChatGPT, Claude, Copilot or Gemini business plans set up properly, with single sign-on, admin controls, a usage policy and staff onboarding.
Enable · For sensitive work
Private AI for law, accounting and advisory firms
Private AI & Knowledge Assistants is Pylon Digital's private AI for law, accounting and advisory firms that can't put client data into public tools: AI set up for sensitive work, with a staff chat assistant plus an "ask our documents" assistant over your precedents, policies and procedures. Its answers cite the documents they came from, so staff can check them.
Last updated:
Is it right for you
Who it's for
Private AI suits 10–100 person law, accounting and advisory firms whose work involves client material that should not go into public AI tools. The usual trigger is a partner ruling out ChatGPT for client files, or staff losing time hunting for the current precedent or procedure.
- Law firms with precedents, practice notes and procedures spread across SharePoint, network drives and email
- Accounting firms with engagement procedures, workpaper checklists and technical notes that juniors struggle to find
- Advisory firms with methodology libraries and engagement templates built up over years
- Firms whose clients, insurers or engagement terms restrict where client information can go
How does an assistant answer from our own documents?
It searches your documents first, then writes an answer using only what it found. This approach is called retrieval-augmented generation, or RAG. When a lawyer asks about the firm’s standard limitation-of-liability clause, the assistant finds the matching passages in your precedent bank, gives them to the AI model with the question, and returns a short answer with a link to each source.
Because documents are looked up at question time rather than built into the model, answers follow your collection. Update a precedent and answers change at the next refresh; remove a superseded procedure and it stops appearing. If nothing relevant is found, the assistant is designed to say so rather than guess.
Why not use ChatGPT for client work?
For some work you can, but not for all of it. As a matter of best practice, the OAIC recommends that organisations do not enter personal information, and particularly sensitive information, into publicly available generative AI tools (OAIC guidance). For law firms holding privileged material, and accounting firms holding tax file numbers and financial records, that guidance touches much of the daily work. Our comparison of private AI and ChatGPT for law firms sets out where each fits.
This is general information, not legal advice.
Deliverables
What you get
You get a private assistant staff can use every day, the document collections behind it and the controls that decide who sees what.
Staff AI chat
A chat assistant for drafting, summarising and checking, opened in a browser with your firm's existing sign-in, for work you would not put into a public tool.
Ask-our-documents assistant
Answers questions such as 'what is our file-closing procedure?' from your precedents, policies and procedures.
Source citations
Every document answer links to the passages it drew on, so staff can open the original and check it before relying on it.
Permission-aware answers
Designed so each person only gets answers from documents they can already open in your existing systems.
Curated document collections
We work with you to choose which libraries go in, set aside superseded versions and agree how new documents are added.
Monthly run and support
Hosting, monitoring, document refreshes and model updates, delivered through Managed AI.
Process
How it works
The path is a discovery call, a scoped pilot on one document collection, then a wider rollout and a monthly service that keeps it running.
- Step 0145 minutes, free
Discovery call
We look at the work you want kept out of public tools, where your documents live and who needs access.
- Step 02Agreed in writing
Scope and document review
We agree the first document collection, the permission model and the hosting approach, and write all three down.
- Step 03Timeline in your proposal
Build and pilot
We set up the assistant, load the first collection and test it with a small group against questions they already know the answers to.
- Step 04Timeline in your proposal
Rollout and training
We open access to the wider firm, add further collections and train staff to check sources.
- Step 05Monthly, ongoing
Run
We host, monitor and update the assistant as your documents, staff and AI models change.
Integrations
Systems we work with
The assistant reads documents where your firm already keeps them and uses your existing sign-in. Documents in a practice-management system can be included where the system allows; we confirm this in discovery.
- Microsoft 365
- SharePoint
- OneDrive
- Google Workspace
- Actionstep
- LEAP
- Clio
Security & data residency
Where your data goes, and who can see it
Private AI exists for work that can't go into public tools, so the controls are the product. Client data is stored in fully GDPR-compliant data centres, and answers are designed to draw only on documents each person can already open.
- Where data is stored
- The assistant and your document collections are stored in fully GDPR-compliant data centres. Our security and data residency page explains how we protect them.
- Which AI models answer
- Answers are generated by an AI model named in your proposal, and we show you that provider's terms on model training before the build.
- Who has access
- Staff sign in with their existing Microsoft 365 or Google Workspace accounts. The assistant is designed to follow your source systems' permissions, so a document restricted to one practice group stays out of other staff's answers.
- Model training
- Your documents are looked up when a question is asked, not built into a model. Your proposal shows the model provider's terms on training, and we switch on the settings it offers to keep client data out of it.
- Privacy
- Each system's handover documentation records what personal information it holds, where it is kept and who can see it, so your firm can answer privacy questions from clients and regulators.
Pricing
Pricing guidance
Pricing model: Setup + monthly
Setup fee plus a monthly fee, both quoted in writing after the free discovery call. Setup depends on the number of document collections, how much tidying they need and your permission structure. The monthly fee covers hosting, monitoring, refreshes and model updates. Your proposal shows whether AI model usage is included or billed separately.
Illustrative example
Example: a 40-person law firm with precedents spread across SharePoint
Talk through a similar buildIllustrative scenario: junior lawyers ask senior staff which version of a precedent is current, because the precedent bank holds several copies of most documents. With Private AI, a senior associate and the practice manager choose the current versions, and staff ask questions such as "what is our process for closing a conveyancing file?" The assistant answers in a few sentences and links to the procedure and precedent it used.
Commitments
What you can hold us to
These are the working rules behind every engagement, from the first discovery call to the monthly Run service. Ask us on the call how each one works for the system you have in mind.
Scope and price in writing first
Every build is scoped and priced in writing before work starts, so there are no open-ended hours.
A person approves what agents send
Our AI agents draft; a person at your firm approves what they send. The After-hours Receptionist answers live only from wording your firm approves in advance.
Every agent action logged
Every action an agent takes is recorded in an audit log, so you can see exactly what it did.
GDPR-compliant data centres
Client data in the systems we build and run is stored in fully GDPR-compliant data centres.
Accessible, fast websites
Websites are designed and tested against WCAG 2.2 level AA and Google’s Core Web Vitals “good” thresholds.
Built and run under one roof
A build can move straight onto a monthly Run service with us, so it is not handed over and forgotten.
Melbourne and Victoria
Areas we serve
We are based in Melbourne and work with businesses across Melbourne, Geelong and Victoria, on site by arrangement or remotely.
Questions
Frequently asked questions
Is private AI safe for client-confidential work?
It is designed for that work. Private AI keeps prompts, documents and answers in a system set up for your firm and stored in fully GDPR-compliant data centres, with sign-in through your existing accounts and answers designed to draw only on documents each person can already open. Staff still check answers against the cited source before relying on them, as they would check a junior's work.
How is private AI different from ChatGPT or Copilot?
The difference is where your data goes and who sets the rules. ChatGPT and Copilot run on the vendor's own service, under the vendor's terms and storage locations. Private AI is set up for your firm, on hosting chosen for your firm, and answers from collections you curate. A firm can use both: a public tool through AI Setup for general work, and Private AI for client matters.
Where is our data stored?
In fully GDPR-compliant data centres, in a system set up for your firm. Access is limited to the people and systems that need it, and your proposal sets out how your data is handled. Your originals stay in your existing systems; the assistant keeps a searchable index of the collections you choose, refreshed as documents change.
Will staff see documents they shouldn't?
They shouldn't. The assistant is designed to follow the permissions in your source systems: if a staff member cannot open a file in SharePoint, the assistant should not use it in their answers. We test this during the pilot with restricted documents before any wider rollout.
What happens if the assistant gets something wrong?
It can, which is why document answers cite their sources. AI models sometimes produce confident but wrong text, known as hallucination. Answering from your own documents and showing the passages used makes errors easier to spot. Staff are trained to open the source before relying on an answer, and anything going to a client should be reviewed by a person first.
Secure by design. Set up correctly. Fully managed.
Talk to us before you commit to anything
Start with a free 45-minute discovery call. We look at your systems and priorities, then recommend a first step with a fixed scope, or tell you if we are not the right fit.
Keep exploring
Related services and reading
Run
Managed AI
Hosting, monitoring and support for your agents, automations and private AI, plus seat admin for public AI tools, run by the team that built them.
Industry
Law firms
Private AI over your precedents, supervised agents for intake, email and first drafts, and weekly matter-level WIP and lock-up reporting for Australian law firms.
All services
Enable, Build, Run
See the full line-up of 10 services and how an engagement moves from setup to a monthly Run service.