AI consulting · on-premise AI · AI agents — vendor-independent, on your hardware or hosted in Europe.
100% vendor-independent No cloud lock-in GDPR-compliant Open-weights models Infrastructure in Europe
We take an AI project through all three stages: we advise vendor-independently on what is worth doing, we build the system with open-weights models, and we keep it running — on your premises or hosted in Europe. You decide where the data lives; we document what that choice means for cost, operation and data protection.
Alongside that, we build AI agents: programs that plan a task, call the right tools and run it through to a result inside your systems. Planning means the agent breaks the work down, orchestration means several agents and services cooperate, and automation means the process runs without someone starting each step.
| AI consulting | Potential analysis, cost-benefit, roadmap — independent recommendations without product ties. |
| On-premise AI | Local AI systems set up on your premises: GPU server hardware, open-weights LLMs running locally, integration into your systems. Data stays entirely in-house. |
| Solutions & training | RAG and knowledge assistants, workshops and enablement, compliance and governance — so AI actually delivers. |
Each can be commissioned on its own, and every recommendation stays free of product ties.
|
nokkela.ai — AI consulting & solutions Vendor-independent AI consulting and on-premise systems with open-weights models — including a potential and cost-benefit analysis before you invest. |
nokkela.io — AI agent systems Agents that plan a task, call the right tools and see it through — built for real workflows on infrastructure in Europe. |
We compare the three ways to run AI so the decision is made on facts rather than on hype.
| Criterion | US cloud AI | nokkela.ai · hosted | On-premise (your building) |
|---|---|---|---|
| Operation / location | US cloud, worldwide | Data centres in Europe | On your premises |
| Data sovereignty | Data leaves Europe | Data stays in Europe | Data stays in-house |
| GDPR | Third-country transfer critical | GDPR-compliant (EU) | No external processing |
| Cost model | Per token / subscription | Predictable (per token or package) | Upfront investment + running costs |
| Vendor lock-in | High | Open (open weights & various third-party models) | Open, full control |
| Offline operation | No | No (hosted) | Air-gapped possible |
On-premise AI is an infrastructure project as much as a model project: the GPU node, the storage holding the documents and the network around them decide whether the system is usable. We build that part too.
|
nokkela.host — Compute & GPU VMs, containers, CEPH block storage and GPU nodes in data centres in Europe — what a hosted model actually runs on. |
nokkela.cloud — Storage for your corpus S3-compatible object storage and Nextcloud in Europe — where the documents behind a retrieval index live. |
Around them: nokkela.network for firewalling and Zero Trust, nokkela.systems for the wider infrastructure, and nokkela.app for the application layer, handed over under the MIT licence.
We choose the model for the task, not for a vendor relationship. Because the weights are open, the model stays runnable on hardware you control.
Mistral Qwen Gemma 4 GPT-OSS Nemotron DeepSeek GLM MiMo MiniMax Kimi-K2 Phi Whisper (STT) Kokoro (TTS)
Use case, the languages involved, the answer quality required and the available GPU memory decide the choice — and we test the shortlist against your own material before committing.
| Knowledge management | Make internal documents searchable (RAG) — answers instead of searching. |
| Document processing | Capture and structure invoices, contracts and emails automatically. |
| Customer communication | Assistance for support, FAQ bots and draft texts. |
| Data analysis | Detect patterns, create reports, support decisions. |
| Content creation | Marketing and technical texts efficiently, with quality control. |
| Automation | Streamline workflows — often together with nokkela.io. |
We assess each use case for effort and value before anything is built, so nothing is automated for its own sake.
What does "open weights" mean, and why not simply call a cloud AI API?
Open-weights models are models whose trained parameters are published for download, so we can run them on hardware you control instead of calling someone else's API. That is what makes the on-premise and European-hosted options possible at all: the model file sits next to your data. Licence terms differ from model to model, so we check each one for your intended use and name it in the concept.
Can an on-premise system run completely offline?
Yes — it can be operated air-gapped, because the open-weights model and the retrieval index sit on your own hardware and no request goes to an external API. Model files, the inference stack and updates come in through your own change process instead of a live download. The hosted variant cannot do this: it runs in data centres in Europe and needs a connection.
Are our prompts and documents used to train models?
Whatever you put into an on-premise system stays inside your network: the model runs on your own hardware, so prompts, documents and the index built from them are not sent to an external provider and are not available for anyone else's training. In the hosted variant, processing takes place on infrastructure in Europe instead of with a US provider. Where a use case calls for a third-party model, we name it and its terms in the concept beforehand.
What do you need from us to build a RAG knowledge assistant?
Access to the document sources — file shares, wikis, ticket systems, contracts — plus your access rules, because the assistant must respect the permissions that already exist. We start with one clearly bounded body of documents rather than everything at once.
What hardware does an on-premise AI system need?
A GPU server, a local inference service and integration into your existing systems; we specify, procure and install the hardware on your premises. Sizing follows the workload: the size of the chosen model and the number of concurrent users determine how much GPU memory is needed, so we fix that in the concept rather than guessing. Network, storage and backup are planned with it.
| 1 · Initial talk | Free and non-binding, up to 30 minutes | We understand your goal. |
| 2 · Analysis | Potential, maturity, use-case assessment | A clear recommendation — including "not worth it". |
| 3 · Concept | Model shortlist, sizing, data flow | Cost and timeline in writing, before implementation starts. |
| 4 · Delivery & operations | Implementation, training, support | Ongoing operation with named contacts. |
- Enquiry form: https://nokkela.ai/en/#contact
- Email: contact@nokkela-it-concept.com
- Phone: +49-151-56092554
- Security contact: security@nokkela-it-concept.com (security.txt)
nokkela.ai and nokkela.io are operated by Nokkela-IT-Concept GmbH — one legal entity, one contracting party for every domain in the group.
Lucas-Cranach-Straße 14, 96317 Kronach, Germany · Amtsgericht Coburg (Coburg Local Court), HRB 7780 · EUID DED4401V.HRB7780 · D-U-N-S 317285761 · VAT ID DE464043147. Managing directors: Daniel Plominski, Manuela Plominski.
The company profile of the group is at github.com/nokkela-it-concept-com (English) and github.com/nokkela-it-concept-gmbh (German). All websites are published in English (default) and German. Last reviewed: 9 August 2026.