--- title: "On-premise AI: self-hosted LLM on your servers — G&G Technologies" description: "On-premise AI and self-hosted LLMs: models run on your own servers and documents never leave your infrastructure. Built in Europe, from San Marino." publisher: "G&G Technologies" lang: en canonical: https://ggtechnologies.sm/en/services/on-premise-ai/ translation: https://ggtechnologies.sm/servizi/ai-on-premise/index.md --- # AI that works where your data already is. *On-premise AI* Models that run on your machine or on your server. The documents stay where they already are, and the person deciding stays the person who decided before. - [Talk to us](mailto:info@ggtechnologies.sm?subject=Information%20request) Or write to us: [info@ggtechnologies.sm](mailto:info@ggtechnologies.sm) - **1997** the first project - **18** years in manufacturing - **900** workers in the study *Where we start* ## The data you would need AI on is the data that cannot leave. Many companies stop short of AI for one reason: the data they would need to use it on is exactly the data that cannot leave. Medical records, contracts, designs, customer and supplier data. The fix is less exotic than it sounds: the model runs inside your own infrastructure and the documents never leave it. It is the principle Podz.AI is built on, and the same one we apply in tailored projects. #### Based in Via Marino Moretti 23, 47899 Serravalle, Republic of San Marino. #### Development Software designed and developed entirely in Europe. #### Product Podz.AI, the personal AI workstation, built on the DigiSense® framework. *The proof · Podz.AI* ### This page describes a method. You can download it. Everything written above — the model running inside your own infrastructure, the documents that never leave it, personal data masked when the cloud is needed — is already inside a program you can install today. It is called **Podz.AI**, it is our own product, and the trial runs for thirty days. You download it, give it your documents and watch what it does. If somebody tells you about on-premise AI, trying what they built is the quickest way to find out whether they can build it. The [release history](https://github.com/G-G-Technologies-Srl/digisense-releases/releases) is public too, along with the price list. It shows how long we have been maintaining it, which is something a quotation does not tell you. *How it works in practice* ## Three configurations, depending on where the data must stay and how much power is needed. ### Fully local Models and data on the same machine or server. Processing works with the network unplugged. - The data stays inside your infrastructure - Open, self-hosted models, keys kept in-house - The limit is the GPU you have available ### Hybrid with anonymisation The processing stays on-premise. When cloud power is needed, personal data is masked before it is sent and restored in the answer. - Masking of names, dates and addresses - Masked fields are replaced before the request leaves your machine - What may leave is decided case by case ### Integration with systems in use AI reads documents and business systems where they already are, and there they stay. - Controlled access to archives - The data stays in your own archives - Traceability of who asked for what *How we work* ## The choice between local, hybrid and cloud starts from the data, not from the technology. 01 ### Classify the data Which data is involved, who may see it, and which task has to be done. 02 ### Draw the boundary Where the model runs, what it sees and what it must never see. This is where the architecture is decided. 03 ### Install on-premise Installation inside your own infrastructure, integration and testing with the people who will use it. 04 ### Maintain Maintenance, model updates and evolution over time. *FAQ* ## Frequently asked questions ### What machine does it take to run AI on-premise? It depends on model size and load, and in practice on how much GPU memory you have. For one person, a Mac with Apple Silicon or a PC with a dedicated GPU. For continuous business use, a server with a GPU. Exact sizing we do together, before any quote. It is a trade we practised before AI needed it: in the late nineties our chief executive was assembling workstations with the graphics cards chosen to run CAD, and the constraint was the same one. ### Is local AI less capable than cloud AI? On generic tasks, frontier cloud models remain more powerful. On a defined task — reading one type of document, extracting certain fields — the gap narrows considerably, and the confidentiality advantage weighs more. ### Can I start on-premise and open up to the cloud later? Yes, that is the most common setup. You start fully in-house and open to the cloud only for the steps that need it, with personal data masked before sending. ### Is anonymisation enough for compliance? On its own, no, and the distinction is a technical one: for GDPR purposes the compliance assessment belongs to the data controller — whoever decides why and how that data is used. What you need in order to make it, we document: which fields are masked, what actually leaves, and how you verify it. ### Where does my data get processed, if you are in San Marino? On your own servers. In an on-premise architecture the model runs inside your own infrastructure: the place of processing is yours, and our home country bears on the contract rather than on the path the data takes. That is the difference from a cloud service, where that path runs through the supplier by design. Where a service of ours processes data on your behalf, the applicable law and the place of processing sit in the contract, before anything starts. *Insights* ## Technical analysis on the same subject. ### [The AI Act and client data](https://ggtechnologies.sm/en/insights/ai-act-client-data/) What a firm using AI on client data actually risks, between the AI Act, the GDPR and Italian Law 132/2025, and what to do from Monday morning. ### [Before you hand an AI agent the keys](https://ggtechnologies.sm/en/insights/autonomous-agents-boundaries/) To an AI agent, the content it reads and the instructions it receives are the same text: that is prompt injection. Three questions to ask first. ## Where to go next ### [Artificial Intelligence](https://ggtechnologies.sm/en/services/artificial-intelligence/) Agents that read documents and query your systems, with the measure agreed first. ### [Podz.AI](https://ggtechnologies.sm/en/podz-ai/) The product, ready to install: it reads your files, 30 days to try it. ### [DigiSense®](https://ggtechnologies.sm/en/digisense/) What sits underneath, and why your project does not start from scratch. ## Do you have data that cannot leave and work AI could do? Describe the task and the kind of data. From there we can tell whether you need fully local or a hybrid architecture. - [Email us](mailto:info@ggtechnologies.sm?subject=On-premise%20AI) - [Call us](tel:+3780549900824)