AI services
Private AI for business.
On hardware you control.
Open-weight language models running on your own server, workstation or cloud tenant, so sensitive data never goes to a public AI service. Scoped up front, built on hardware you buy, handed over to your team. If a business tier is enough, we will say so.
Private AI for business means running an open-weight language model on hardware you own or control, so sensitive data never goes to a public AI service. Pro IT NW, a Seattle-based, labor-only consultancy, scopes, builds and hands over private AI on your on-premises server or your own cloud tenant. It suits firms whose data cannot leave their control. We resell no hardware or software, and we will say when a business tier is enough.
Do you need private AI, or is a business tier enough?
Start with the AI your staff already use, because that is where the exposure is today.
People are already drafting, summarizing and analyzing with ChatGPT, Claude or Gemini, often on personal accounts with no company policy and no admin control. For many firms the right first step is not a private model at all. It is moving that use onto company-owned business or enterprise accounts with single sign-on, admin controls, written data terms and an acceptable-use policy. That is the work on our secure AI for business page, and it is often enough.
Private AI makes sense when
- A contract, regulator or client says certain data must stay on systems you control
- A site has to work with limited or no internet connection
- You want one model version held steady while a process depends on it
- The work is mostly questions over your own documents, not open-ended writing
A business tier is usually better when
- Staff mainly need general drafting, summarizing and research help
- The vendor's written data terms already meet your obligations
- No one inside the business can own patching and model updates
- Usage is light, so dedicated hardware would sit mostly idle
Scope
What does a private AI engagement include?
A private AI deployment is a small system your team will own, so we scope every part of it, not just the model.
- Use-case definition. The specific questions or tasks the model is for, who uses it, and which data it may touch.
- Model selection. An open-weight model chosen for the task, with its license terms flagged for your own review.
- Hardware sizing. What the workload needs in general terms, on premises or in your own cloud tenant. You buy it from the vendor you choose.
- Retrieval over your documents. An index of your own material that respects existing permissions, so people only get answers from files they can already open.
- Access control and identity. Sign-in through your existing directory, such as Entra ID, with access limited to the people who need it.
- Logging. A record of who asked what and which documents were used, kept where your team can review it.
- Patching and ownership. A documented process for operating system, software and model updates, with a named owner on your side.
- Handover. Documentation and hands-on training for your internal IT staff.
How does a private AI project run?
Every project runs in the same five steps, and the first one can end with a recommendation not to build.
- Discovery. What AI is already in use, what data is involved, and what problem a private model would solve. If a business tier fits better, we say so here.
- Design and scope. Use case, model, hardware requirements, access model and logging, written down with a fixed scope before work starts.
- Build. On the hardware or cloud tenant you procured, connected to your directory and your documents.
- Test with your people. Real questions from the staff who will use it, including checks that permissions hold.
- Hand over. Documentation, training, and the update process passed to your team.
What do you have at the end?
- A working private AI system on hardware you own or a tenant you control
- Sign-in through your existing identity system, with access limited by role
- Logs your team can read
- Written documentation of the design, the model, and how to update both
- Internal IT staff trained to run it without us
What private AI is not
- Not a like-for-like replacement for the largest hosted AI models on every task
- Not hands-off: someone on your side has to own updates
- Not a hardware sale. You buy from the vendor you choose; we implement
- Not legal advice on data obligations or model licenses
Planning new server hardware anyway? Private AI is often worth scoping alongside VMware exit and server modernization. Not sure what AI your staff already use? Start with the AI usage review, or read about secure AI for business.
FAQ
Common questions about private AI.
What is private AI for business?
Private AI for business means running an open-weight language model on hardware the business owns or controls, either an on-premises server or workstation or the business's own cloud tenant, so prompts and documents are processed there instead of being sent to a public AI service. Pro IT NW, a Seattle-based, labor-only consultancy, scopes, builds and hands over private AI deployments.
Do we need private AI, or is a business tier of ChatGPT, Claude or Copilot enough?
Often a business tier is enough. Business and enterprise tiers of the major AI tools give you company-owned accounts, single sign-on, admin controls and written data-handling terms you can read before you sign. Private AI earns its cost when a contract, a regulator or a client says certain data must stay on systems you control, when a site has to work without an internet connection, or when you need to hold one model version steady. If none of those apply, Pro IT NW will say so and point you to setting up a business tier properly instead.
What is an on-premise LLM?
An on-premise LLM is a large language model that runs on servers or workstations in your own building or data center rather than on an AI vendor's cloud. Open-weight models make this possible because their files can be downloaded and run locally. Running one also means owning its access control, logging and updates.
Can a private AI search our own documents?
Yes. The usual pattern is retrieval: the system searches an index of your documents and gives the relevant passages to the model, so answers draw on your material rather than on the model's general knowledge. The important design rule is that the index respects existing permissions, so a user can only get answers from documents that user is already allowed to open.
Do you sell the hardware or the AI software?
No. Pro IT NW resells nothing: no hardware, software or license margins and no partner badges. We describe what the workload needs in general terms, you buy the server, workstation or cloud capacity from the vendor of your choice, and we implement on it. That keeps the hardware recommendation free of any sales incentive.
Who maintains a private AI system after it is built?
Your team does, and the engagement is built around that. Pro IT NW hands over documentation covering the model, the hardware, access control, logging and the update process, and trains your internal IT staff to run it. If no one inside the business can own patching and model updates, that is a strong sign a business tier of an existing AI tool is the better choice.
Free checks
Not ready to talk yet? Start with one of these.
10 questions
Could you take back the keys?
Ten yes / no questions: could you remove your IT provider's access without their help and keep control? Admin accounts, domain, licenses, backups, contract dates.
Open →
Runs in your browser
Who controls your domain?
See who runs your domain's DNS, where your email goes, whether you publish DMARC, and whether on-prem Exchange is still visible. Nothing stored.
Open →
Sample report
See a Copilot readiness report
A sanitized sample assessment: findings scorecard, readiness ratings and remediation roadmap, so you can see what you would get.
Open →
Find out whether private AI fits, before you buy hardware.
Tell us what data is involved, who would use it, and what is driving the question. We will come back with scope, or a plain answer that a business tier will do.