Private AI: Why Your Data Stays Within Your Organization
When a company deploys AI that sends data to external servers, it loses control over where that data ends up and who processes it. For banks, private clinics, and public institutions in Montenegro, this is unacceptable — not only due to internal policy, but also because of legal obligations. Private AI running on your own infrastructure (on-prem) or on a private server in the EU is the only model that guarantees data never leaves your perimeter.
In Brief
- On-prem AI means the language model and all data remain on servers you control — with no third parties in the processing chain.
- Banks, clinics, and public institutions have a legal obligation to protect personal data under the Law on Personal Data Protection and GDPR for business with EU clients.
- ProCode develops fine-tuned AI models that are installed directly into the client's infrastructure or on a private server in the EU.
- On-prem installation requires a GPU server, network isolation, and access management — ProCode delivers a complete solution.
- Pricing for custom AI solutions starts from €800, and more complex on-prem solutions from €2,000.
Why Public AI Services Are Not an Option for Regulated Industries
When data is sent to an external AI service, it passes through a third party's infrastructure. For organizations that process clients' financial data, medical records, or confidential government information, this creates regulatory exposure. Financial regulators expect institutions to demonstrate full control over the data processing chain, and sending that data to an external service directly undermines that control.
Health data is a special category under GDPR and requires the highest level of protection. When a clinic uses AI running in someone else's cloud, every query containing patient data becomes a potential regulatory risk. The same principle applies to law firms, accounting firms, and local governments that process citizens' personal data.
What GDPR and Montenegrin Law Require from Companies Using AI
Montenegro, as an EU candidate country, is aligning its Law on Personal Data Protection with GDPR provisions. For companies doing business with EU clients, GDPR applies directly — regardless of where the company is registered. This means a Montenegrin bank, hotel, or clinic that processes data of EU guests or patients must meet the same standards as a company headquartered in Berlin or Paris.
The key requirements relating to AI systems are: transparency of processing, data minimization, the right to erasure, and accountability for processing security. An on-prem AI model meets all of these requirements because the organization always knows where the data is, who accesses it, and how it is processed. With external services, these guarantees depend on the contract and privacy policy of a third party — which is not sufficient for regulators.
This is particularly important for the tourism and hospitality sector: GDPR applies to all organizations whose business is in any way connected to data of EU residents, which in practice covers almost every hotel on the Montenegrin coast.
How On-Prem AI Installation Works
On-prem AI means the language model, embedding models, and all query processing infrastructure operate within the organization's network. Data never leaves your server — neither during training nor during system use. ProCode delivers a complete solution: from fine-tuning the model on the client's data, to installation and configuration on your infrastructure.
1. Analysis and Preparation
Infrastructure assessment, defining use cases, data preparation and cleaning for fine-tuning. GPU capacity and network isolation requirements are established.
2. Fine-Tuning and Testing
ProCode trains the language model on the client's data using PyTorch and FastAPI on its own GPU infrastructure. The model is tested and validated before delivery.
3. Installation and Handover
The model is installed into the client's infrastructure (on-prem or private EU server) using Docker and Kubernetes. Access management is configured via Keycloak and team training is conducted.
On-Prem vs. Cloud AI: What to Choose and When
| Criterion | On-prem / Private Server | Public Cloud AI Service |
|---|---|---|
| Data control | Full — data stays with you | Limited — data goes to third-party servers |
| GDPR compliance | High — no data transfer | Requires DPA agreement and provider verification |
| Model customization | Fine-tuning on your data | Generic model, limited customization |
| Latency | Low — inference is local | Depends on internet connection and server load |
| Initial cost | Higher (infrastructure + development) | Lower (subscription) |
| Long-term cost | Predictable, no per-query fee | Grows with the number of queries |
| Suitable for | Banks, clinics, public administration, law | Internal tools without sensitive data |
A Practical Example: A Private Clinic in Podgorica
Imagine a private clinic in Podgorica that wants to introduce an AI assistant for searching medical records — test results, discharge letters, and protocols. Doctors could ask questions in natural language and immediately receive answers with references to the patient's documentation.
If the clinic were to use a public AI service, every query would contain patient health data that would be sent to external servers — directly violating GDPR and Montenegro's Law on Personal Data Protection. The only acceptable solution is AI document search installed on-prem: the model is trained on the clinic's internal medical documentation, runs on a server within the clinical network, and does not communicate with external services. Doctors get fast and accurate answers, while patient data never leaves the building.
The same principle applies to a law firm in Kotor that wants AI contract search, or a local government in Nikšić that processes citizen requests. In all these cases, on-prem installation is not a luxury — it is the only legally compliant choice.
Technical Prerequisites for On-Prem AI
A successful on-prem installation requires several technical elements. ProCode evaluates all of these factors during the initial analysis and proposes the optimal solution for the organization's size and budget.
- GPU server: Inference of language models requires a graphics card with sufficient VRAM capacity. The model size determines the minimum requirements.
- Network isolation: The AI system must be separated from the public internet. ProCode configures Kubernetes namespaces and network policies that prevent uncontrolled data egress.
- Access management: Keycloak is used for authentication and authorization, with support for LDAP/Active Directory integration with the organization's existing systems.
- Database and vector search: PostgreSQL with the pgvector extension enables hybrid search — a combination of vector and traditional search — without external services.
- Audit trail: Every query and response is logged locally, providing regulators with full visibility into the data processing flow.
For organizations that cannot or do not wish to invest in their own GPU server, ProCode offers an alternative: installation on a private server in an EU data center, where data remains within EU legal jurisdiction and does not pass through the infrastructure of third-party AI companies.
Which ProCode Services Support the On-Prem Model
All ProCode services can be delivered as an on-prem solution. The most common combinations for regulated industries are:
- AI document search — answers from contracts, policies, and internal documentation with source references, from €2,000
- AI chatbot for businesses — internal assistant for employees or client-facing service, installed on-prem, from €800
- Custom business platforms — ERP, CRM, and DMS systems with built-in AI functionalities, from €3,000
- AI agents and process automation — automation of internal workflows without sending data outside the network, from €800
Frequently Asked Questions
Does on-prem AI mean the company must have its own data center?
No. On-prem can be installed on a single physical server within your network, but also on a private server in an EU data center managed by you or your provider. The key is that data does not pass through the infrastructure of external AI services. ProCode recommends the optimal solution for your technical and budgetary capacity during the analysis phase.
How much does an on-prem AI installation cost?
The price depends on the complexity of the use case, the volume of documentation for fine-tuning, infrastructure requirements, and the number of users. Simpler AI assistants and chatbots start from €800, while more complex solutions such as AI document search with on-prem installation start from €2,000. A detailed overview is available on the pricing page.
How does on-prem AI help with GDPR compliance?
When the model runs exclusively on your infrastructure, there is no transfer of data to third parties that would require a Data Processing Agreement (DPA) with an AI provider. The organization always knows where the data is, who accesses it, and how it is processed — which is the foundation of GDPR compliance. On-prem models eliminate an entire category of risks related to data transfer.
Can on-prem AI be as accurate as public services?
Yes, and it is often more accurate for a specific industry. Fine-tuning on the client's data — internal documentation, procedures, contracts — gives the model a deep understanding of context that generic models lack. A medical clinic gets a model that understands its terminology and documentation; a bank gets a model that knows its internal procedures.
How long does the implementation of an on-prem AI solution take?
A typical implementation takes four to twelve weeks, depending on the complexity of the solution and the availability of data for fine-tuning. The phases are: data analysis and preparation, model training and testing, infrastructure installation and configuration, user training, and handover. ProCode provides support after the solution is handed over as well.
What if the company doesn't have an IT team to manage the server?
ProCode offers the option of installation on a private EU server managed by ProCode on behalf of the client, with guaranteed data isolation. The client gets all the benefits of the on-prem model without needing their own IT capacity. This option is particularly suitable for smaller clinics, law firms, and local governments.
Private AI running on your own infrastructure is not just a technical decision — it is a strategic decision about who controls your most valuable data. For banks, clinics, and public institutions in Montenegro, that decision is increasingly a legal obligation as well. The on-prem model provides full control, regulatory security, and AI tailored to your business — not the other way around.
Let's Talk About Your Project
If you are considering introducing AI that stays within your infrastructure, ProCode can assess your technical prerequisites and propose a concrete solution. No obligations and no technical jargon — just a straightforward conversation about your needs.
