Get your head out of the cloud.
Your own AI, running on hardware you control. Every file, conversation and memory stays under your roof and inside your jurisdiction.
Where does our data go? Your computer.
Who holds the keys? You do.
Who trains on it? Nobody.
Every prompt is a copy of your work, sent somewhere else
AI is now where the thinking happens: the contracts, the patient notes, the formulas, the strategy. Each time it goes into a public tool, a copy of it lives on someone else's computer.
Your best knowledge leaves the building
The work that makes a firm valuable is exactly the work people paste into AI: client files, research, pricing, unreleased ideas.
Where it is stored depends on the fine print
A licence tier, a residency add-on and a subprocessor list decide which country holds your data. Any of the three can change next quarter.
"Customers outside the EU may have their queries processed in the US, EU, or other regions."Microsoft, on Copilot data residency
Regulators have started asking
Professional bodies in law, health and finance now ask where client data goes when AI is used.
"An AI tool which stores data offshore could lead to unintentional breaches."Ahpra, guidance for health practitioners
Public AI is "akin to putting it in the public domain".Law Society of NSW
The cloud is someone else's computer. This one is yours.
OnSite AI Labs builds private AI that runs on a machine you own. The model, the files, the memory and the logs all sit inside your four walls, so the answer to "where does our data go?" is one sentence, and it stays true.
Sovereign
Your hardware, your jurisdiction, your keys. Your work lives where you decide, whatever a provider's policy says next year.
Private
Every person's conversations, files and memory are kept separate. Colleagues see what their role works on and nothing more.
Governed
Administrators decide who uses what, whether an agent may search the web or run code, and can audit every action it takes.
Everything you use AI for today, kept in-house
The same everyday power people reach for in public tools, on infrastructure you control.
Chat and analysis
Writing, planning, summarising and problem solving, with a quick mode and a deep-thinking mode.
Web research
Searches and reads current sources, compares them and answers with the sources attached.
Reads any file
PDFs, Word, Excel and images. Ask questions, extract details, compare documents, analyse spreadsheets.
Creates documents
Word reports, PowerPoint decks, Excel workbooks and PDFs, straight out of the conversation.
Sees images
Screenshots, photos, charts and scanned forms, read and explained.
Runs code, sealed off
Calculations, data analysis and file processing inside an isolated sandbox.
Remembers how you work
Preferences, working style and standing instructions, held per person.
Learns from corrections
Gets better each time it is put right, and the learning stays scoped to you or your team.
Agents with a job, and only the access that job needs
Each agent has its own instructions, its own tools and its own connections. A finance agent sees the ledger. A research agent sees the web. Each keeps to its own lane.
Research
Reads the web and your library, comes back with sources.
Document analyst
Reviews contracts, reports and records, and flags what matters.
Finance
Works the numbers with the books it is cleared to see.
Operations
Handles the recurring admin: inboxes, follow-ups, scheduling, filing.
Support
Answers from your own policies and knowledge, in your voice.
Your own
Build agents for any team or workflow, with exactly the access you grant.
Connects to the tools you open every day
Mail, files, calendars, chat, projects and the books. Connected one agent and one person at a time, so each connection goes only where you send it.
Gmail
Outlook
Google Drive
OneDrive
SharePoint
Dropbox
Google Calendar
Teams
Slack
Zoom
Notion
Google Sheets
Xero
HubSpot
Trello
Asana
Jira
GitHubBuilt on an open connector standard, so the list grows with the tools you use.
Industries it can work for
Anyone whose work is worth more because nobody else has it.
Personal knowledge work
Your notes, drafts, research and ideas, with an AI that knows them and keeps them to itself.
Proprietary work and R&D
Formulas, designs, source code and unreleased products stay inside the company that owns them.
Health
Clinical notes, referrals and patient records, worked on inside the practice.
Law
Privileged files, contracts and matter notes, reviewed and drafted in-house.
Finance and accounting
Client ledgers, tax files and forecasts, handled under your own confidentiality rules.
Care and disability services
Participant files, care plans and incident reports, kept with the provider responsible for them.
Consulting and advisory
Client strategy, board papers and deal work, separated client by client.
Government and defence supply
Sensitive programmes on dedicated hardware you can account for.
Your own model by default, a frontier model when you choose
Private comes first. Reach for the biggest public models only for the work you decide is safe to send.
Private model, on your machine
Runs locally for everyday work. Client data, files and memory stay inside your walls.
Claude or ChatGPT, on your own key
Switched on per person, for the tasks you choose, billed to your own account.
Questions people ask before the call
Your files, conversations and memory are stored and processed on your own machine. Two things can reach outside, and both are yours to switch on or off: web research, which sends the search query, and a frontier model on your own key, for tasks you choose.
It is configured for you: your documents, your instructions, your agents and your connections. Your work stays out of every training set.
You do. The hardware is yours, the data on it is yours, and you control who can reach it.
Only where you allow it. Each person's conversations, files and memory are separate, and each agent sees only the folders and connections its role needs.
Backups are kept off the machine, and restoring from them is part of every setup. It is the first question we expect, so it is the one we plan around.
We spec the hardware, set it up, connect your tools, and look after it once it is running. Your people just use it.
See what private AI could do for your work
Thirty minutes. Tell us what you would hand to an AI if you knew it would stay private, and we will show you how it could work.
- What you could run privately, first
- Which tools it would connect to
- What the hardware looks like for a team your size