Own AI server project: the steps from the first request to staff using the assistant

How an own AI server project runs

A project for your own AI server runs in four stages, from a first conversation about your tasks to staff using the assistant every day: the checks before the quote, the order and delivery of the machine, the setup of the models with your documents, and a pilot followed by the handover. The 14 steps below say for each one what happens and what you, the customer, prepare or decide, so that a managing director without an IT department sees the whole path before the first call.

The page describes the order of the steps and who does what. It gives no durations: how long each step takes depends on your rooms, your documents and the components, and the lead time of the hardware is confirmed in writing with the quote. Where a step has its own detailed guide on our site, we link it. If you are still deciding whether a server suits your company size at all, start with our guide to an own AI server for a company of 10 to 100 people.

The implementation steps at a glance

Steps from our AI servers and Private AI/ML pages and our guides linked below; the column "your part" is our summary.

The steps do not always follow in exactly this order. A firm that buys only the machine and runs the software itself stops after step 9. A firm that wants the platform from us runs the technical assessment before the order, because the choice of model can change the number of cards.

Before the quote: conversation, room and data

Step 1, the first conversation. It is free of charge. You describe what the assistant should do, for example answer questions from the service manuals, draft offers or write meeting minutes, how many people would use it and which documents it should know. Our Private AI/ML page describes the outcome as two or three possible solution scenarios. Name one person on your side who will own the assistant later; it does not have to be an IT specialist, as our article on an AI server without an IT department explains.

Step 2, the room and data check. We check the rack, the power and the airflow before we quote. For a firm without a server room this means a few plain questions: where the machine would stand, which fuse the circuit has and what else hangs on it, whether the room is cooled all year, and whether there is a wired network port. Our checklist for preparing a server room lists everything in detail. On the data side, list where the documents live today (a file server, a document management system, mailboxes, a wiki), in which formats and languages, and who is responsible for each collection. Our guide to preparing company documents for an AI assistant shows how to do that list.

Quote, assessment and order

Step 3, configuration and quote. From the tasks, the number of users and the room, the machine is sized: number and edition of the cards, processors, memory, storage and network. The written quote states the price, the lead time and the warranty terms; we return a configuration and quote within one business day. For a firm of 10 to 100 people this is usually a workstation with two RTX PRO 6000 Max-Q cards or a rack server with two RTX PRO 6000 Server Edition cards.

Step 4, the technical assessment. If we also build the platform, our Private AI/ML service starts with a paid technical assessment whose price is fixed before work begins. It covers the solution architecture, the choice of model and GPU, and a pilot plan with metrics. Your decision here is the pilot process: one task with a result you can count, such as questions answered from the manuals or minutes written. Our guide to what a private AI pilot should measure explains which numbers to collect.

Step 5, the order. You place a purchase order against the quote. The contract is with Eurokommerz in Austria, with EU invoicing, and one Eurokommerz project lead stays your contact for hardware and engineering. Before technical detail is shared, an NDA is signed, and a data processing agreement is available on request. Whether you need one, for example because our engineering partner Vixen.UNO will work on systems that hold personal data, is for your data-protection officer to confirm.

We supply the machine and, through Private AI/ML, the platform on it under one contract; send us the tasks and the number of people, and we reply with a first scenario and a configuration within one business day.

Assembly, delivery and acceptance

Step 6, assembly and test. The machine is assembled and tested before delivery, with a power-on and basic load test. Nothing is needed from you here, but keep the delivery note and the serial numbers that come with the machine: they are what a warranty case asks for later.

Step 7, delivery and placement. We deliver anywhere in the EU, and other destinations on request through our logistics partner. Commissioning on site is available on request. Your part is practical: a clear route to the room, someone to receive the delivery, the socket or rack position checked in step 2 and a network port. A rack server is heavy and deep; the server-room checklist covers rails and rack depth.

Step 8, your acceptance test. Before the machine goes into use, check it against the order: every card recognised, the diagnostics clean, temperatures and power under load within the limits. Our guide to GPU server acceptance testing describes the tests. Acceptance is easier when both sides agree the tests and pass criteria before the order.

Setting up the software and connecting documents

Step 9, the base software. On request the machine arrives with the operating system, the NVIDIA driver, CUDA (NVIDIA's software layer that lets programs use the GPU) and a container runtime (the tool that runs each piece of AI software in its own sealed package). Your IT service provider, if you have one, gives the machine a network address and an address for its remote management. For IT staff, our guide to a new GPU server setup from BIOS to the first model covers this step technically.

Step 10, models and platform. Next come the model itself, the serving engine that runs it (such as vLLM), the web interface staff type into, and a gateway that issues keys and writes a query log, a record of who asked what and when. The engineers work in your environment under your access controls and do not take copies of production data out of it. Two decisions are yours: which model to use, from the shortlist the assessment produced, and how long the query log is kept. Our guide to choosing an open LLM for a company assistant covers licences and languages.

Step 11, connecting the documents. The document collections from step 2 are connected so that the assistant can search them and name the source of every answer. Each user sees only what their access rights allow. This is where your heads of department decide: which collections go in, which version of a document is the current one, who may see what, and who keeps the collection up to date. The quality of the answers depends more on these decisions than on the hardware. Our article on RAG on company data explains why.

Users, rules and training

Step 12, accounts and rules. Staff sign in with accounts from your directory service if you have one, or with accounts created for the platform. Before access, three things should be settled. A written usage policy tells staff what may go into the assistant and what moves there from public chatbots, as our guide to AI usage policies describes. Under Article 35(1) of the GDPR, a data protection impact assessment is required before processing that "is likely to result in a high risk to the rights and freedoms of natural persons"; your data-protection officer decides whether that applies, and our DPIA guide shows the procedure. Where your company has employee representatives, involve them early; our article on the works council and an internal AI assistant covers this from the EU side.

Short user training belongs in the same step. The AI Act defines a deployer as a person or body "using an AI system under its authority", outside personal non-professional use, so a company using its own assistant is one. Article 4, as amended by the Digital Omnibus on AI and in the consolidated text of 27 July 2026, reads: "Providers and deployers of AI systems shall take measures to support the development of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf". Chapters I and II of the AI Act, which include Article 4, apply from 2 February 2025. This is general information, not legal advice; your legal counsel confirms what it means for your company. A first session usually covers what the assistant is good at, that its answers can be wrong, how to check the named source and what must not be typed in. Our guide to AI literacy training under Article 4 and our article on whether a small company needs an AI officer go further.

Pilot and handover to daily use

Step 13, the pilot. A few users work with the assistant on the process chosen in step 4. They grade the answers against a set of real questions with known correct answers, and the gateway logs how many requests arrive in the busiest hour and how long answers take. At the end you decide with these numbers: extend to all staff, change the model, add documents, or stop. Our Private AI/ML service scales only what has proved its value in the pilot.

Step 14, handover and support. The team that will run the platform is trained to run it and develop it further. After that the choice is yours: your own staff, or our support under an agreed SLA. The manufacturer warranty on every component is handled through us, defective-on-arrival units are replaced, and the assembly is covered by us. For larger rollouts in waves, our guide to rolling out a private AI assistant to 1,000 employees continues from here.

Our Private AI/ML service starts with the free first conversation and a pilot on one process with clear metrics; describe the process you would start with, and we propose a pilot and the machine for it.

Who you deal with

You sign one contract, with Eurokommerz, and work with one Eurokommerz project lead from the first call through support. Eurokommerz supplies the hardware with EU invoicing and the manufacturer warranties. The engineering for the models, documents and platform comes from our partner Vixen.UNO, inside that contract. Hosting, if you prefer it to a machine on your premises, runs on dedicated hardware in a Tier-3 data centre in Lithuania; then steps 2, 7 and 8 change, and the rest stays the same.

What we supply

You can buy this system from us: our team sizes the configuration to your workload and budget and replies within one business day. We build AI workstations and servers with two, four or eight GPUs to order, assembled and tested before delivery, delivered anywhere in the EU on one EU contract and invoice, with manufacturer warranty on every component handled through us; quantities and lead time are confirmed in writing with the quote. The operating system, drivers, CUDA and a container runtime are installed on request, and the models, documents and pilot are our Private AI/ML service.

FAQ