When practitioners hear "on-premise AI," many assume they are being asked to build something. A server room. A dedicated IT department. An infrastructure project that runs for months before the system is useful.
That is not what this describes.
A private AI deployment for a small professional office is a single dedicated computer, placed in a back office or server closet, connected to the office network. Staff access it through a web browser on any device connected to that network. Nothing leaves the building. No third-party vendor is involved in ongoing operation.
Here is what that looks like from the initial assessment through steady-state use.
The assessment
An engagement begins with a review of what the practice has and what it needs. This covers the documents the practice wants the system to work with, the hardware already in use, the network configuration, the number of people who will use the system, and any compliance or confidentiality requirements specific to the practice.
The assessment produces a written document: what the system would do, what it would not do, which hardware tier is appropriate for the practice's volume and user count, and what deployment involves. The practice has a complete picture of costs, timeline, and scope before any purchase decision is made.
The hardware
At the scale appropriate for a solo or small practice, the hardware is a Mac Mini M-series or Mac Studio M-series, depending on the number of users and the size of the document corpus. A Mac Mini M-series Pro with 64GB of unified memory handles one to three users and up to 3,000 documents. A Mac Studio M-series Max handles three to eight users and up to 15,000 documents.
The computer sits on a desk, a shelf, or in a rack. It connects to the office router via Ethernet. There is no server room requirement. No rack is required. A compact dedicated machine in a closet is adequate for a practice at this scale.
The hardware runs continuously. It does not need to be managed day to day. Updates to the software and models happen through a defined procedure, not through automatic updates that could introduce regressions during business hours.
The document corpus
Before the system can answer questions, it needs documents. The initial ingestion involves collecting the files the practice wants the system to know about, processing them through the pipeline, and loading them into the retrieval index. The practice controls what goes in. The system draws only from those documents. It has no access to the internet and no knowledge beyond what the practice provides.
Adding documents after deployment is a routine operation. The index updates without downtime. Removing documents is equally straightforward: the relevant entries are deleted from the index and the system no longer draws from them.
What staff sees
Staff interact with the system through a web browser. The interface is a query field. A user types a question in plain language: "What are the indemnification terms in the Henderson matter?" or "Summarize the patient history for this chart." or "Which suppliers appear in deviation reports from the past eighteen months?" The system returns an answer with citations to the specific documents it drew from. The user can follow those citations to the source material and read the original passage.
The system does not replace the document management software, billing system, calendar, or case management platform the practice already uses. It adds a search and drafting capability that works from the practice's own documents. Existing tools remain in place.
What an engagement looks like
Three phases, each with a defined scope.
Assessment is a written deliverable: hardware recommendations, a proposed scope for deployment, and an honest account of what the system will and will not do for this practice. Billed as a flat fee.
Deployment covers hardware procurement, configuration, document ingestion, and staff orientation. Timeline depends on document volume and firm size. A solo-practice deployment is measured in days. A small-office deployment is measured in weeks.
Support is an ongoing arrangement covering software and model updates, document additions, configuration questions, and incident response. The practice has a defined point of contact, not a help desk queue.
The judgment about what to do with the output remains with the professional. The system retrieves and organizes. The attorney, physician, accountant, or operations director decides.
What changes and what does not
The practice's documents remain stored where they are stored now. Existing document retention policies apply. Existing practice management tools are unaffected.
What changes is what staff can do with those documents. A question that previously required a manual search through a filing system returns cited passages in seconds. A drafting task that required locating and reviewing several prior matters produces a structured starting point from the practice's own precedents.
The judgment about what to do with that output remains with the professional. The system retrieves and organizes. The attorney, physician, or accountant decides.