Private AI · in your office or a dedicated cloud

Your firm's own AI. On hardware nobody else uses.

SAFFT AI installs open-source models, your company's documents, a task agent and long-term memory on a machine in your office or a GPU server rented only for you. Your team gets a capable assistant, and your client data stays where it is.

Built for accountancy practices, law firms, wealth managers, fintechs and clinics: teams that can't paste client data into public chatbots.

Leaving your network0 B
Questions answered1,204

Illustration. Every question and document stays inside the line.

Your team already pastes contracts, payslips and client emails into public chatbots. Most firms find out after the fact.

SAFFT AI gives them a better tool, inside your own walls, with a record of every question asked.

Two ways to own it.

The same software and the same screens either way. You choose where it runs and how you pay for it.

SAFFT AI Box

A compact AI workstation installed in your office or server room. It can run with the internet switched off.

Runs on
A high-memory AI workstation (128 GB class), sized to your team after testing
Your data
Never leaves your premises
Internet
Optional. Fully offline is possible
You pay
Hardware and setup once, then a monthly care plan
Ready
After delivery and an on-site install

SAFFT AI Cloud

The same system on a GPU server reserved for you alone, in the UK or EU region you pick. Nothing to install in the office.

Runs on
A dedicated GPU server, not shared with anyone
Your data
Stored in your chosen UK or EU region, encrypted
Internet
Private encrypted link from your office
You pay
A monthly subscription, GPU time included
Ready
Within days of connecting your documents

cloud ⇄ box Start in the cloud and move to a box later, or the other way round. Your documents, memory and settings come with you.

What's inside.

Five parts, all open-source, all running on hardware dedicated to you. We choose, install and tune them. Your team just asks questions.

PartWhat it does for youBuilt with

The model

Reads, writes, summarises and reasons. We test several on your real work and keep the one that does it best.

Open-weight LLMs: Qwen, Llama, Gemma, Hermes

Your knowledge

Searches your contracts, policies and files, and shows the page each answer came from.

Retrieval over your document stores (RAG), respecting folder permissions

The agent

Takes on multi-step jobs: drafts the letter, fills in the questionnaire, prepares the board pack for review.

Hermes Agent, with only the tools you approve

Memory

Remembers clients, decisions and preferences, so nobody has to explain the same thing twice.

Per-person and per-team memory you can view and erase

Controls

Decides who sees what, keeps a log of every exchange, and waits for a person before anything leaves the system.

Role-based access, audit log, approval steps

What your team will ask it.

  1. What notice do we owe Halden if we end the services contract early?Answers from the contract itself and cites clause 14.2.
  2. List every auto-renewal clause that expires before June.Reads the whole contracts folder and returns a table you can check.
  3. Brief me on the Morrow account before Thursday's call.Pulls emails, notes and documents into one page.
  4. Draft answers to this due-diligence questionnaire.Reuses your approved past answers and flags the gaps.
  5. Explain the Q3 admin cost variance against budget.Writes first-draft commentary from your management accounts.

Prove it before you buy.

A pilot runs on a dedicated cloud GPU with your real documents and your real people. You decide on a box or the cloud only after you've seen the results.

Week 1

Scope

Pick two or three jobs with your team and agree which documents the AI may read.

Weeks 2–5

Pilot

Your people use it on real work. We measure answer quality and speed on your tasks, and at the end of week 5 you get both options costed side by side.

After you decide

Care

Model updates, monitoring, backups, new documents and new skills as your team asks for them.

Try it in the cloud. Then own it.

The rules it runs by.

SAFFT AI comes from SAF FinTech Advisory, which works in regulated financial services. We set up private AI the way that world expects: least access, full records, people in charge.

No public chatbot services

The models run on your box or your dedicated server. Prompts and files are never sent to an AI provider.

Answers follow permissions

People only get answers from documents they're already allowed to open.

Every exchange is logged

Who asked what, what came back and which sources were used, ready to export for review.

A person approves actions

The agent drafts and prepares. Anything leaving the system waits for someone to sign it off.

You own the data

Knowledge base, memory and logs are yours to export or delete at any time.

You can see inside

Open-source components mean no black box: you can inspect what runs and where.

See it working on your own files.

Tell us a little about your firm. We'll reply within two working days to arrange a short scoping call.

  • 30 minScoping call. No preparation needed.
  • 1 pageA written pilot plan: the jobs, the documents, the limits.
  • 2 quotesBox and cloud costed side by side before you commit.
Where would you like it to run?

We only use these details to reply about your pilot.

Questions we hear first.

Which AI models do you use?

Open-weight models such as Qwen, Llama, Gemma and Hermes. We don't pick one in advance. During the pilot we test several on your work and keep the best balance of quality and speed for your hardware.

How many people can use one box?

That depends on the model, the length of your documents and how many people ask at the same moment. We measure it during the pilot and size the hardware to your team instead of quoting a number up front.

Is it as good as ChatGPT or Claude?

On questions about your own documents it is often more useful, because it has read them and cites them. On the hardest open-ended reasoning the largest public models are still ahead. The pilot shows you where that line falls for your work.

Can it run without the internet?

Yes, a box can run fully offline. Updates are then installed on a schedule you agree, either on site or over a connection you open for that purpose.

Do you see our data?

Only when you give us access for setup or support, and that access is logged. The rest of the time the system runs without us.

What if we want to stop?

You keep the box hardware. With either option you can export your knowledge base, memory and logs, and we delete any copies we held.