AI on-premise · sovereign by design

Your data.Your model.Your infrastructure.

ONPREMA.ai deploys generative AI in your data center or private cloud. Zero prompt leakage to third-party APIs. Zero per-token lock-in. Full compliance with GDPR, DORA and the EU AI Act.

Book a technical call See the architecture

data leaves your network

100%

control over model & weights

4-8 wks

typical pilot timeline

Why on-premise

Cloud APIs are not an option for everyone

Public sector, finance, healthcare, defense and industry face hard data constraints. We build AI that respects them.

Data sovereignty

Personal data, legal documents, source code and trade secrets never leave your infrastructure — ever.

Predictable cost

CAPEX instead of ever-growing per-token OPEX. You know deployment and maintenance cost upfront, with no surprises as usage scales.

Regulatory compliance

GDPR, NIS2, DORA, EU AI Act, sector regulators. Models run inside your trust zone and inherit your security policies.

Vendor independence

Open weights (Llama, Qwen, Mistral, Bielik). Swap models without rewriting your app — no ecosystem lock-in.

Latency & availability

The model runs next to your data. No dependency on public internet, no third-party API rate limits.

Fine-tuning on your data

We tune models to your domain, jargon and processes. The result is your IP — it never becomes training data for someone else's model.

Solutions

From ready integrations to custom deployments

We ship working connectors for the systems you already use. We start there and grow into your process.

Integration

AI Copilot for Mint Service Desk

Automatic ticket classification and prioritization, suggested replies from your knowledge base, long-thread summarization, KB article generation from resolved tickets. Ships as a native Mint SD plugin (mintsd.com).

30-50% faster ticket handling
Auto-routing to the right support group
Real-time SLA-at-risk detection

Integration

Semantic search on Apache SOLR

Hybrid search (BM25 + vectors) over your SOLR index. Embedding model runs locally, the index stays in your infrastructure. RAG without shipping documents to external APIs.

Natural-language search in EN and PL
Cross-encoder result reranking
Source citations — full auditability

Custom

Private AI assistant for your org

Chat grounded in your documents (RAG), with SSO, per-document access control and a full audit log. Model, embeddings and vector store all live inside your network.

SSO: Entra ID / Keycloak / LDAP
Row-level security aligned with your IAM
Audit log for every prompt and response

Custom

Fine-tuning and domain models

We tune open models (Bielik, Llama, Qwen) on your data — docs, tickets, code, transcripts. The full training pipeline runs on your hardware.

LoRA / QLoRA on available GPUs
Evaluation on your own test sets
Model and experiment versioning (MLflow)

Architecture

What it looks like under the hood

A reference stack we adapt to your environment — bare metal, VMware, OpenShift, Proxmox or private cloud.

your-network.local

01Application layer

Mint Service DeskInternal portalsCustom applications

02AI layer

Orchestrator (LangGraph / custom)RAG pipeline · re-rankerGuardrails · PII redaction

03Model layer

vLLM / TGI / OllamaLLMs: Bielik, Llama, Qwen, MistralEmbeddings · Reranker · Whisper

04Data layer

Apache SOLR (hybrid search)Qdrant / pgvectorMinIO · PostgreSQL

05Infrastructure layer

GPU: NVIDIA H100 / L40S / RTXKubernetes / OpenShift / bare metalMonitoring: Prometheus · Grafana · Loki

All AI traffic stays inside your network zone. Zero calls to public model APIs.

Delivery process

From first call to production

step 01

Discovery (1-2 wks)

Workshops with your team: use cases, data, constraints, compliance requirements, available hardware.

step 02

PoC / Pilot (4-6 wks)

We run the chosen scenario on your infrastructure. Measurable KPIs agreed upfront.

step 03

Production

Hardening, SSO, backup, monitoring, documentation, training for your IT team.

step 04

Evolution & support

SLA, model updates, additional use cases, Managed Service if you want us to run it.

About ONPREMA.ai

A technology brand by ITSM Software S.A.

ONPREMA.ai is an initiative by ITSM Software S.A. — a Polish software vendor with a track record of building enterprise-grade systems (including Mint Service Desk, used across Poland and abroad). We combine years of on-premise delivery, integration and production support with strong generative AI, RAG and MLOps capabilities.

itsmsoftware.pl

Polish software vendor

Contracts, support and documentation in English or Polish.

Enterprise track record

Years of delivery in public sector and finance.

Owned products

Mint Service Desk and connectors built in-house.

Let's talk

Book a technical call

30 minutes with our AI architect. No sales deck — we ask about your use case, data and infrastructure, and tell you what's realistic.

sales@itsmsoftware.pl