A brand of iTech Group

We deploy local language models in your company

An AI assistant on your servers, with no data sent to the cloud.

I work with a team that has 30 years of IT experience. We analyse your needs, select the hardware and models, install the environment, connect your company documents and train the team that will use it.

Get in touch How a deployment works

A team with 30 years of experience

Deployments on your servers

Your data stays in-house

Team training after go-live

AI can help with everyday work. Your data does not have to leave the company.

Public cloud models are convenient, but they mean sending questions, procedures and documents outside the company. You also pay per user and depend on an internet connection. That is why we deploy the assistant on your servers.

  • Your data stays in-house

    Questions, documents and the model run on your servers. That matters for GDPR and for protecting trade secrets.

  • Predictable costs

    You set the environment up once. You do not pay a subscription for every user, and you are not waiting for the next price rise from a cloud provider.

  • Works without the internet

    Once installed, the environment runs locally — including when there is no connection to the outside world.

  • An experienced team

    I work with a team that has 30 years of IT experience. We deploy what we teach: the environment has to run on your servers, not just in a slide deck.

Deploying a local AI assistant

We run language models on your servers. Questions, documents and answers stay in the company. The scope is always tailored to your situation — no one-size-fits-all package.

  • Analysis of needs and your current environment
  • Hardware and model selection
  • Installation of Ollama and Open WebUI on your servers
  • A knowledge base from your company documents
  • Security: accounts, HTTPS, backups
  • Integration with your existing environment
  • Training for your team after go-live
  • Ongoing maintenance

Also available: reviews and workshops

Consulting

Readiness assessment, policies and workshops

We help you assess whether the company is ready to work with a local model, draft rules for using AI, and run workshops for IT teams. This is technical and organisational support — it is not legal advice.

How a deployment works

We work step by step, on your servers. You are left with a working environment, documentation and a team that knows how to use it.

  1. Analysis

    We discuss your tasks, data and servers. We determine whether a local assistant makes sense and where to start.

  2. Hardware and model selection

    We recommend a server, a graphics card — if one is needed — and models suited to your use cases. There is no single setup that fits everyone.

  3. Installation and go-live

    We install Ollama and Open WebUI on your servers, set up accounts, HTTPS and backups, add a knowledge base from your company documents and, if needed, connect the environment to your existing tools.

  4. Training and maintenance

    We train the team that will use the assistant. We leave documentation and can look after the environment after go-live.

Who we are

iTech AI is a brand of iTech Group. Implementations and training are led by Jakub Wojtanowski. He works with a team that has 30 years of IT experience.

We work with networks, Linux systems, containers, security and local language models. We deploy what we teach: an AI assistant should run on your servers, not just in a presentation.

Jakub is the lead trainer and is responsible for how deployments are carried out.

  • CCNA (Cisco)
  • CCNP (Cisco)
  • CCNA CyberOps (Cisco)
  • CCAI — Cisco Certified Academy Instructor
  • SSCP (ISC2)
  • MCT — Microsoft Certified Trainer
  • MCSA (Microsoft)
  • MCITP (Microsoft)
  • LPIC-1 (Linux Professional Institute)

Training

Alongside deployments, we run practical courses: from local language models, through containers and Kubernetes, to Linux hardening. Training is delivered on-site or remotely, on virtual machines. Dates and fees are agreed individually.

Course · 2 days

“AI under control”

Local language models in the company, with no data sent to the cloud. Two days of hands-on labs: from starting a model, through the interface and API, to working with company documents, automation and security.

Course agenda

Day 1 — models, Ollama, Open WebUI and APIs

  • Introduction and Lab 0 — environment check
  • Module 1. What a language model is
  • Module 2. Cloud or on-premises
  • Module 3. Choosing hardware and a model
  • Lab 1. Ollama: installation, models and throughput measurement
  • Lab 2. Open WebUI in Docker
  • Lab 3. The Ollama API and the OpenAI-compatible API
  • Module 4. Prompt engineering in a company

Day 2 — RAG, automation, security and operations

  • Module 5. RAG — answers grounded in documents
  • Lab 4. A knowledge base from PDF documents and quality tests
  • Lab 5. Automating an administrator’s work
  • Module 6. Fine-tuning models (LoRA) — an overview
  • Module 7. Security
  • Lab 6. nginx with TLS, access control and backups
  • Module 8. Operations
  • Lab 7. An IT-department assistant from scratch, following a checklist

The discussion of GDPR and the AI Act is technical and introductory. It does not replace legal advice.

Course · 2 days

Docker: containers, Harbor, Trivy and Portainer

A workshop for administrators: installing the engine, building a secure image, data, networking and Docker Compose, then a private Harbor registry with the Trivy scanner and host management in Portainer. 10 labs on Debian, Ubuntu and Rocky Linux.

Course agenda

Day 1 — engine, images, data and networking

  • Introduction, topology and architecture
  • Module 01. Fundamentals and installation
  • Module 02. Images and the container lifecycle
  • Module 03. Your own image and BuildKit
  • Module 04. Data and restoring a backup
  • Module 05. Networking and publishing ports
  • Recap and lab-environment check

Day 2 — Compose, registry, security and management

  • Module 06. Compose — a web application and a database
  • Module 07. Harbor with Trivy
  • Module 08. Trivy and a security gate
  • Module 09. Portainer CE and Agent
  • Module 10. Hardening, rootless and a short look at Swarm
  • Final exercise and close

Course · 2 days

Kubernetes: from k3s to a kubeadm cluster

A practical workshop for administrators. On day one you work with the k3s distribution; on day two you build a kubeadm cluster with Calico from scratch. You deploy, upgrade, deliberately break and repair the same application across 16 labs.

Course agenda

Day 1 — the k3s mini-distribution on a single node

  • Module 1. From a process to a cluster
  • Module 2. Mini-distributions and installing k3s
  • Module 3. Core objects and persistent data
  • Module 4. Deployments, Services and troubleshooting
  • Module 5. Ingress and Helm
  • Module 6. Alternatives and choosing a distribution

Day 2 — a full kubeadm cluster on three nodes

  • Module 7. From k3s to kubeadm
  • Module 8. Bootstrapping a kubeadm cluster
  • Module 9. Images and deploying the application on kubeadm
  • Module 10. Application operations
  • Module 11. Security and node maintenance
  • Module 12. Scaling, observability and upkeep

Course · 2 days

Linux security and hardening

A workshop on securing Linux servers: from a threat model and a Lynis audit, through accounts, sudo, PAM, SSH with keys and MFA, a firewall and SELinux, to auditd, AIDE and monitoring with Wazuh. 11 labs on Rocky Linux, Debian and Ubuntu.

Course agenda

Day 1 — the system, accounts, access and access control

  • Opening and the lab environment
  • Module 1. Threat model and a baseline
  • Module 2. Updates and reducing the service surface
  • Module 3. Accounts, sudo, PAM and password policy
  • Module 4. SSH: keys and MFA
  • Module 5. Firewall and protection against dictionary attacks
  • Module 6. SELinux and AppArmor
  • End-of-day recap

Day 2 — audit, integrity, monitoring and compliance

  • Module 7. Logs and audit: journald, rsyslog, auditd
  • Module 8. File integrity: AIDE
  • Module 9. Wazuh: agents, SCA, FIM and alerts
  • Module 10. Service security and systemd sandboxing
  • Module 11. Compliance and a repeat audit
  • Summary and an implementation checklist

Common questions

Who carries out the deployment?

Jakub Wojtanowski leads deployments together with a team that has 30 years of IT experience. Hardware, models and scope are agreed individually.

Do we need a graphics card?

Not always. Smaller models can run on a capable server without a dedicated GPU. Larger models and a more comfortable working pace usually need a graphics card with enough memory. We discuss hardware individually — there is no single setup that fits everyone.

Does the model know current information and our procedures?

The model itself only knows what it learned during training, and it does not reach out to the internet or your files on its own. To answer from your company documents, we add a knowledge base (known as RAG): it first searches your materials, then prepares an answer based on them.

Does any data leave the company?

The point of a local installation is that questions, documents and the model stay on your servers. During the deployment we set up the network, HTTPS and backups so that this is the case. This does not relieve the company of its own obligations under GDPR.

How long does a deployment take?

It depends on the scope: from simply starting the assistant to adding a knowledge base, user accounts and usage rules. The timetable and stages are agreed individually.

What about GDPR and the Artificial Intelligence Act?

A local installation helps keep data in the company, which matters for GDPR and the protection of trade secrets. The AI Act adds further duties that depend on how the model is used. We do not provide legal advice — we can help on the technical side and point out where you should consult a lawyer.

Will this replace public cloud models?

For work with company data and procedures, a local assistant lets you keep questions in-house. It is not a copy of a particular cloud service, but your own environment, under your control. Some uses can still be left in the cloud — that is the company’s decision.

Let’s talk about your environment

Tell us what you need. We usually start with a local AI assistant deployment or with team training. We will get back to you and agree the next steps.

You can also write directly to kontakt@itechai.pl.