Private AI for business, on infrastructure you control
Your contracts, client files, financial records, product data and the records in your CRM, ERP and other systems are the most useful information you have. They are also the information you can least afford to send to an outside vendor. Private AI runs open-weight models, the free models you can host yourself, on a server in your office or in your own cloud account. Your team gets a private ChatGPT-style assistant that answers questions, drafts documents and works with that data and your systems. Every question, document and answer stays inside your walls. We build it, host it where you decide, and hand you the keys.
- Your cloud account
- Your own server
- Air-gapped
Who private AI is for
Law, accounting and advisory firms that must keep client matters confidential.
Engineering and architecture firms protecting drawings and intellectual property.
Healthcare-adjacent businesses that hold patient information.
Financial firms with regulatory limits on where data can go.
Manufacturers with process knowledge locked in old documents.
Any company whose legal team said no to a vendor’s terms, or that simply does not want a subscription that reads its files.
A private ChatGPT for your team
Your staff are already pasting client emails and contract clauses into a free chatbot. This is the version you can allow. The same chat experience, running on a model you control, answering from your own documents with the source cited, with logins, team permissions and an audit trail. It drafts emails and proposals in your house style, summarizes long documents and meeting transcripts, pulls data from PDFs into tables, and lets each team see only its own files.
How it is built
An open-weight model (Llama, Qwen, Mistral or Gemma, chosen for your hardware and tasks) served with vLLM or Ollama. A chat interface such as Open WebUI, or a custom one. A retrieval layer that indexes your SharePoint, Google Drive, file server, email archive and CRM. Your single sign-on, per-team document permissions and an audit log. The same stack later powers agents in your apps, CRM and other systems, so nothing has to leave for automation either.
Ways to deploy
Your cloud account
A GPU instance in your own AWS, Azure or Google Cloud account. The fastest start, and easy to move on-premises later.
Your own server
A workstation or rack server in your office. Nothing touches the internet except what you allow.
Air-gapped
No internet path at all, with updates delivered by hand. For the most sensitive environments.
Hardware you need
For a company of a few dozen to a few hundred people, one server with a single professional GPU is enough for everyday use. We size it to your team and your document volume, specify it in writing, and you buy it in your name at supplier price. Power and cooling costs are modest. Larger teams or heavy document loads need more, and we quote that case by case.
Where self-hosted models fall short
Open-weight models on one GPU are a step behind the biggest vendor models on the hardest reasoning tasks. For summarizing, drafting, extracting, searching and answering over your own documents, which is most of what a business needs, they are more than good enough. Where you want a frontier model for a specific task, we route only approved, non-sensitive requests to it under your own API account. We tell you which tasks fall where before you buy hardware.
What it connects to
Microsoft 365 and SharePoint, Google Workspace and Drive, file servers, email archives, and HubSpot, Zoho, SuiteCRM or Salesforce. SuiteCRM can sit on the same server as the model.
How a private AI project runs
Scope and sizing
Which documents, which teams, which tasks, and where it will run.
Outcome: a hardware specification and a plan.
Install and secure
Server or cloud instance set up, hardened, with the model tested on your own tasks.
Outcome: a working, locked-down base.
Connect your documents
Ingestion from your sources, permissions mapped to your teams, single sign-on.
Outcome: an assistant that knows your business.
Pilot with a team
A small group uses it on real work while we tune it.
Outcome: evidence it is useful.
Roll out and hand over
Runbook for your IT team, monitoring, and a support plan if you want one.
Outcome: a system you own.
Related pages and reading
Questions people ask about private AI
Something else on your mind? Ask an engineer directly.
Is a self-hosted model good enough?
For document questions, drafting, summarizing, extraction and internal search, yes. We show you on your own documents before you commit.
Can it read our CRM?
Yes. HubSpot, Zoho, SuiteCRM and Salesforce connectors are standard.
Who maintains it?
Your IT team with our runbook, or us under a support plan. Model updates are a routine task.
Can we try it before buying hardware?
Yes. Start in your cloud account, or run a proof of concept, and move on-premises when you are ready.
Is it suitable for regulated data?
A private setup removes the outside vendor from the picture. We scope the rest, access, logging and hosting, with your compliance lead.
See it answer from your own documents
Send us a handful of non-sensitive documents and we will show you what a private assistant does with them.