Private AI questions offices ask before they call

Book a free 20-minute call, or email info@modelside.xyz, or call 516-423-7714. We reply within one business day. On the call, tell us what is slowing your office down and what you cannot put into a public AI tool. We say whether a local setup fits before anything is written. If it does, the next thing you receive is a written fixed-price scope. It says what we build, what it runs on and when it is delivered. If it does not fit, we say that too, and you have lost twenty minutes.

A free 20-minute call, then a written fixed-price scope for the build, then optional ongoing support. No hourly billing. We do not publish prices because the scope changes the price. Three things move it. Whether you need a workstation or already have a capable PC. How many document sources we connect. How many automations we build. Unlike a per-seat subscription, you own the machine and the software on it. There is no monthly fee for each person who uses it. The scope also includes the delivery date, so you know the timeline before you agree to anything.

Yes, and most offices should. Common first steps: ModelSide Chat on a PC you already own, one folder of documents connected, or one repetitive task automated. When the first thing earns its keep, we add the next. If the machine you have is not enough for the models you need, we tell you before any money changes hands. A ready-to-use workstation is the next step up.

Law firms, medical practices, dental offices, and real estate and construction firms are the four we write about on this site. Their rules about client and patient data are strict, and the document work is heavy. Lead-generation, finance and other teams with data they cannot upload to public AI tools also hire us. Paid work so far includes medical practices, a jewelry business in New York's diamond district, funding and finance, construction, and lead-generation businesses. If your office does not fit one of those labels, the question is the same. Is there work you cannot put in a public tool?

It depends on the plan and on your obligations, and the details change. Consumer and small-business ChatGPT plans send what you paste to a third party, and whether a given plan is covered by a Business Associate Agreement depends on that plan's current terms, which can change under you. Plans with stronger terms still put your files with a third party under a contract you have to keep re-reading. A local setup removes the question. The model runs on your own machine, so there is nothing to send and no terms to track.

It depends on the plan, and the terms change. Before any patient information goes into a cloud AI tool, check whether the plan you are on is covered by a Business Associate Agreement. Even where one is offered, the practice remains responsible for how patient data is used. The data still leaves your office. That is the practical reason we build on hardware you own.

Yes to both. For medical and dental practices, the Business Associate Agreement is signed before we see any patient data. For law firms and other offices we sign an NDA. Built to support your HIPAA obligations: patient data is processed on hardware you own, nothing is sent to ModelSide or any cloud, and we sign a Business Associate Agreement. A BAA governs ModelSide's side as your vendor; your own policies, staff training and access rules remain yours.

Your files are processed on hardware you own, and nothing is sent to ModelSide or any cloud. During setup we work on your machine, not on copies somewhere else. If a tool needs a test file, you pick one and it never leaves that machine. A remote support session starts only when you open it, and you can watch everything we do. On-site visits are the same work at your desk. The full description is on our data-handling page, and an NDA is available before any of it begins.

Not for daily use. The models, ModelSide Chat and your document tools run on the workstation itself. It answers questions and drafts documents with the network cable unplugged. An internet connection is used for three things. The initial model download, Windows and driver updates, and remote support when you ask for it. If you prefer a machine that never connects, we can set it up that way. You give up automatic updates and remote help, and we handle those on-site instead.

You do. The machine is yours, the Windows administrator account is yours, and the passwords are handed to you at delivery. Whatever access ModelSide has during setup and support is access you grant. You can remove it at any time without breaking anything. There is no hidden account and no license check that phones home. If you want us to keep a support login so fixes are faster, that is your decision. It is written into the support agreement so everyone knows it exists.

The typical build is a Windows workstation: NVIDIA RTX 5090, 128 GB of RAM, a top-tier CPU and fast NVMe storage. It runs any model that fits on a 5090, which covers the current Llama, Qwen, Mistral and Gemma families and others. It is delivered with the models, ModelSide Chat and your document tools installed and tested, so it works the day it arrives. It sits on your office network, so several people can use it from their own desks. We size the build for your headcount. Pre-built workstations ship nationwide; two have been built and delivered to clients so far.

A private, ChatGPT-style assistant that runs entirely on your machine. There is no account to create, no subscription, and no data sent anywhere: you type, and the model on the workstation answers. It can answer from the documents you connect to it. A question about a matter, a chart or a lease comes back with the passage it was drawn from. It is included on every ModelSide workstation and can be installed on hardware you already own. It is the one tool an office policy can point to and say: use this one.

Open-source models that run on the workstation: the Llama, Qwen, Mistral and Gemma families, and others as they are released. We pick per task, because the model that summarizes a chart well is not always the one that drafts a letter well. When a better model appears, we test it against your work and swap it in under support. Nothing locks you to one vendor.

Works with the files and tools you already use (PDF, Word, scans, email, spreadsheets); direct integrations are scoped per project. In practice that means most first projects run on exports and folders, which almost every system can produce. Where a live connection is worth building, it goes into the written scope. We build it through the software's own interface or through browser automation. We do not list any vendor as integrated on this site. Each connection is scoped and tested for the office that asked for it.

Someone always checks the output, the same way a draft from a new hire gets read before it goes out. Local models are strong at summarizing, searching and drafting from documents you give them. They get things wrong in the same places every AI does: citations, numbers and facts outside the documents. In medical and dental offices the tools are for administrative and documentation help only, not clinical decisions. Bar guidance requires a lawyer to verify AI output no matter where it runs. We build the tools to show their sources so checking is fast.

A workstation is one machine in your office, delivered ready, for offices without an IT department. We set it up, you use it, we support it. Private deployment is for organizations with an IT team. Open-source models are hosted on your own servers or in your virtual private cloud, with access control and audit logging. IT holds the keys. In both cases your data stays in your environment, and nothing is sent to ModelSide or any cloud. The call tells us which one fits.

Both. Our base is New York, New York. We visit offices across New York, New Jersey and neighboring states, and take larger projects further afield. Remote work reaches anywhere in the country, and so do pre-built workstations. A typical project mixes the two. The scoping call is remote and the build happens on our side. Delivery, setup and the staff walkthrough happen at your office when you are close enough for that to make sense. Ongoing support is remote first, on-site when a visit fixes it faster.

We fix it. We diagnose it, remotely or on-site, under support. Software problems are usually a reinstall of the models or a restore of the settings. Your documents live where they always did, so a problem with the AI layer never puts your files at risk. We reply within one business day, and the support agreement says in writing what is covered.

No. The workstation is set up to be used like any other office PC. Under support we handle Windows and driver updates, model changes and fixes. Updates are scheduled with you, so nothing restarts in the middle of a workday. Open-source models keep improving; when a newer one does your tasks better, we test it and install it. Offices that do have an IT team can take a different route. Our private-deployment service hands the keys to your IT staff, with access control and audit logging.