EU-BASED TEAM / AI MACHINES

AI on your
own machine.

Local AI for science, healthcare and government.
Your files stay on your machine.

Machines from €15,000See how it works
SLM / PROS01
AI RUNS HERE● ● ●
02 / THE MACHINES

Choose your machine.

2-MONTH WAITING LIST

All orders will be fulfilled from 2027 onward.

Your delivery date will be confirmed in your quote.

Core

Start with one task.

For research teams, healthcare administration and public-sector departments. Search procedures, summarise documents and draft reports with a small local model.

PROPOSED CONFIGURATION
GPU memory
32 GB
System RAM
128 GB
NVMe storage
2 TB
STARTING AT€15,000
Technical details & uses

Example: one RTX 5090-class GPU. Start evaluation with compact models such as Phi-4-mini (3.8B parameters). Local document search can be added as a software service.

  • Research: protocol and paper lookup
  • Healthcare: administrative documents
  • Government: policy and procedure search

Advanced

More power for more work.

For labs, healthcare organisations and government teams working with larger document collections. More GPU memory for larger models and document search.

PROPOSED CONFIGURATION
GPU memory
96 GB ECC
System RAM
256 GB
NVMe storage
4 TB
STARTING AT€25,000
Technical details & uses

Example: one RTX PRO 6000 Blackwell-class GPU with 96 GB ECC memory. Evaluate 7B–32B models, local embeddings and a document index. Model size, context length and active users determine capacity.

  • Science: larger research libraries
  • Healthcare: internal knowledge search
  • Government: cross-document summaries

Custom

Built around your project.

For research institutes, hospital groups and public agencies with several workflows or strict network requirements. Hardware and integration designed for the project.

PROPOSED CONFIGURATION
GPU memory
192 GB+ total
System RAM
512 GB+
NVMe storage
8 TB+
STARTING AT€55,000
Technical details & uses

Example: two 96 GB professional GPUs, subject to workload and budget. GPU memory is distributed, not automatically one shared pool. Multi-GPU software, storage and network isolation are scoped together.

  • Research institutes and hospital groups
  • Public agencies with multiple teams
  • Projects needing separate AI workflows

Proposed sizing, not fixed product specifications. Hardware starts at the prices shown; software and support cost extra. Final parts, availability, tax and delivery are confirmed in your quote. Sketches are illustrative.

How we size a local AI machine

Memory before model size.

At 4-bit precision, 8B model weights alone take about 4 GB; 32B takes about 16 GB. Actual memory use is higher: allow for the runtime, conversation cache and document tools. B means billion parameters.

Test your actual workload.

We propose measuring answer quality, source accuracy, response time and simultaneous requests using your sample documents. No token-speed or user-count promise before testing.

Local software, chosen for the task.

Options include llama.cpp for CPU/GPU model serving, Phi-4-mini for compact text tasks, and a PrivateGPT-style document-search pipeline. Model licensing and offline operation are checked for each project.

Technical referencesRTX 5090: 32 GBRTX PRO: 96 GB ECCllama.cppPhi-4-miniPrivateGPT

Built around your work.

Science. Healthcare. Government.

01 / SCIENCE

Find the research.

Search papers, lab procedures and internal documentation. Link answers back to the source documents.

Public project reference

The ATLAS collaboration’s chATLAS research describes document search across over 30,000 documents on CERN infrastructure. It is a RAG reference, not proof of an offline SLM deployment.

Read the chATLAS paper
02 / HEALTHCARE

Keep knowledge close.

Search internal procedures and administrative documents. Explore specialist models in a separate, validated research workflow.

Public model reference

Google’s MedGemma offers 4B and 27B model sizes for healthcare development. Its outputs require validation and must not directly guide diagnosis or treatment.

Read about MedGemma
03 / GOVERNMENT

Work through the documents.

Find policies, compare documents and draft summaries for staff review. Access rules and network controls are agreed with your team.

Public project reference

The UK government’s Redbox documents an AI assistant for civil servants to search and summarise documents. It uses LLMs; it is not evidence of an offline SLM system.

Read the Redbox record

These are independent public references, not SLM Pros customers or partners. Suggested uses require project-specific validation.

HOW IT WORKS / THREE LAYERS01 / 03

Start with
your files.

Approved documents become a searchable local library. Your files stay in your environment.

  1. 01 Your files
  2. 02 Your AI model
  3. 03 Your answer

No internet connection.

FILES
MODEL + ANSWERS
SOFTWARE & SUPPORT

Buy the machine.
Add the help you need.

One-off setup or ongoing support.
Quoted for your project and its complexity.

01

Setup

Get your AI ready to use.

  • Select the model, runtime and compression level
  • Add local document search (RAG), embeddings and source links
  • Test answer quality, access rules and outbound connections
02

Maintenance

Keep your machine running.

  • Review model and software updates before installation
  • Monitor GPU memory, storage and response times
  • Agree backup, recovery and support procedures
03

SLM as a Service

Ongoing help with your AI.

  • Ongoing model, document-index and software maintenance
  • Re-test answers when models or documents change
  • Local API connections scoped for your systems
Support scope and response times agreed in your quote.Ask about services
LET’S TALK

Tell us
what you need.

Five people. Based in the EU.

Current waiting list: 2 months. Order fulfilment starts in 2027.

Please leave out confidential information.

Privacy, simply.

We use your name, work email and project details to respond to your inquiry. Requests are stored privately. This website does not use analytics or tracking cookies. The hosting provider may process technical connection logs.

Use the inquiry form for privacy questions. Do not include confidential information.