On-premise & local AI
Private AI for Sensitive Data
Want the benefits of AI but can't send client data to ChatGPT? We install your own in-house AI on your premises or in a private UK cloud, so your data never leaves your control.
Three ways to run it
On-premise AI server
A dedicated AI machine in your office or comms cabinet. Prompts, documents and answers stay inside your building, and it can run with no internet access at all.
Private cloud in the UK
Your own AI environment in a UK data centre region, isolated to your business. No shared consumer service, and no hardware to house.
Hybrid
Sensitive work goes to the private model; everyday tasks can use an approved business tool like Copilot or ChatGPT. Clear rules decide which is which.
What your team can do with it
Ask your own documents
Search contracts, policies, case files and procedures in plain English, with answers that point back to the source document.
Summarise and draft
Summarise long documents and correspondence, and produce first drafts from your own templates and precedents.
Meeting and call notes
Transcribe and summarise recordings on your own hardware, so confidential conversations aren't uploaded anywhere.
Redaction and data extraction
Pull key details from invoices, forms and reports, or find personal data that needs removing before a document is shared.
Private automations
Connect the private model to your workflows, so documents can be processed automatically without leaving your network.
A familiar chat screen
Staff get a ChatGPT-style interface, signed in with their existing Microsoft 365 or Google account, with access set by role.
Long-term memory
A company memory that grows like a web
Most AI chats forget everything when the conversation ends, and can only read so much at once. Your private AI keeps its own wiki of your business instead, an approach popularised by OpenAI co-founder Andrej Karpathy. Every document, email and meeting note it's given is written up into linked pages, so its knowledge keeps building over months and years.
Linked pages, not a pile of files
One page per client, project, supplier, procedure and decision, cross-linked to each other. Ask about a client and it follows the links to everything related.
Memory far bigger than one chat
The AI reads its index first, then opens only the pages it needs. Your company's knowledge can grow far beyond what fits in a single conversation.
It keeps itself tidy
Regular check-ups flag contradictions, out-of-date information and gaps. The pages are plain text your team can read and correct too, and they never leave your systems.
Built and run by a security team
An AI server is another system holding your most sensitive data, so it needs the same care as the rest of your IT. We secure it, look after it and keep it up to date, as part of your managed IT.
We use well-established open-weight models and check each model's licence allows business use.
- Data stays on hardware or cloud you control
- Can run fully offline for the most sensitive work
- Sign-in with your existing work accounts and MFA
- Access controlled by team or role
- Activity logging for accountability
- Patched, monitored and backed up by Senri
How it works
- Step 1
Discovery
We work out which tasks and data need to stay private, how many people will use it, and what it must connect to.
- Step 2
Design and pilot
We choose the hardware or cloud setup and open-weight AI models to suit, then test on your real tasks with a small group.
- Step 3
Install and secure
We install it, connect your documents, set up sign-in and access, and train your team to use it well.
- Step 4
Manage
We keep it patched, monitored, backed up and up to date with better models as they're released.
Private AI is especially popular with solicitors and accountants. It can also power the custom AI agents we build.
Talk to us about private AI
Tell us what you'd like AI to help with. We'll reply by email, usually within a few hours.
Private AI: common questions
Private AI is an AI assistant that runs on your own hardware or in a private cloud environment dedicated to your business, instead of a shared public service. Your prompts and documents stay under your control and aren't sent to a third-party AI provider.
Yes. Open-weight AI models can run entirely on a local server or workstation with no internet connection. That makes it suitable for highly confidential work, although you'll still want a secure way to install updates.
For many everyday business tasks, such as summarising, drafting and searching your own documents, modern open-weight models do a good job. The largest public models are still stronger at the hardest reasoning tasks, so we test on your real work during the pilot and tell you honestly what to expect.
Not ChatGPT itself, which only runs on OpenAI's systems. But you can run very capable open-weight models on your own server, including models OpenAI has released for this purpose, with a ChatGPT-style chat screen for staff. Your prompts and documents then stay on hardware you control.
Yes. A private AI installed on your premises or in a private UK cloud gives staff a ChatGPT-style assistant, but prompts and documents aren't sent to a public AI provider. It's the option we recommend when client confidentiality rules out public tools.
It can build a long-term memory. Instead of forgetting after each chat, it maintains a wiki of linked pages about your clients, projects and procedures, updated as new documents arrive. It reads the relevant pages when you ask a question, so its knowledge can keep growing without hitting the limits of a single conversation.
Businesses whose data can't go to a public AI service, or whose clients won't allow it: solicitors, accountants, financial advisers, healthcare and care providers, HR teams, and anyone handling commercially sensitive information.
It depends on how many people will use it and which models you need. For a small team it can be a single powerful workstation or server; for more users, a dedicated AI server or a private cloud setup. We size it during discovery so you don't overspend.
It can make compliance simpler, because personal data isn't shared with an external AI provider and you control where it's stored and who can access it. You still need a lawful basis and good data handling, which we help you put in place.
It depends on the setup, number of users and whether it's on-premise or in the cloud. We agree the scope on a short call and confirm costs before anything is ordered.