AI chat, knowledge intelligence, RAG, agents, multimodal generation and automation on a local AI infrastructure controlled by your organisation.
I3K Local AI is not a workstation with an LLM on top. It is a private AI platform installed inside the company: EuLLM as the inference runtime, RAG Enterprise as the knowledge and document intelligence platform, and a module-extensible architecture on which to progressively consolidate the AI workloads that are today scattered across several external SaaS services. Data, prompts, documents and generated content stay under the organisation's control, and local inference carries no per-token cost.
The same appliance does not serve a single function. It is the organisation's common AI infrastructure: you start with private chat and RAG, and over time you add agents, voice, document intelligence, generation and automation without changing the architecture.
Business applications
Chat · RAG · Agents · Voice · Documents · Coding · Vision · Generative AI · Automation
I3K AI Services
Orchestration, modules, internal APIs, access control
RAG Enterprise
Knowledge and document intelligence
EuLLM
Inference runtime, model management, OpenAI-compatible API
Local hardware
AMD · NVIDIA · I3K-validated accelerators
We show the full direction of the platform, not only what is ready today. Every capability carries its real status: Available is part of the core platform, Optional module is licensed separately, Planned is in development, Roadmap is confirmed direction but not yet in progress, Hardware dependent means it needs the right configuration.
What the appliance does from day one.
Private company chat on local LLM models, with the option to use different models depending on the task and OpenAI-compatible APIs to connect the applications already in use.
Powered by RAG Enterprise. Natural-language querying over company document repositories, with answers anchored to their sources.
Structured extraction and understanding of documents, beyond plain search: this is where RAG Enterprise extends into concrete business processes.
A core EuLLM capability: managing which models run, how they are quantized and on which hardware, without reinstalling the platform.
Licensed separately on the same appliance.
LLMs do not just answer: with authorised tools they can carry out tasks inside company systems. Availability and integrations depend on the modules installed.
The same infrastructure as a private alternative to cloud AI coding services, with code that never leaves the company network.
Unified search extending the RAG Enterprise document base to other company sources.
Where running locally matters most, because recordings are among the most sensitive data you hold.
STT, LLM and TTS can run locally when the hardware configuration allows: customer and staff voices never leave the infrastructure.
Meetings become a searchable knowledge base, without going through external transcription services.
The architecture is designed to progressively host local multimodal generative models. Availability and output quality depend on model and hardware configuration.
Models able to understand text, images, documents, audio and video together.
Local image generation and manipulation for internal creative workflows.
Generative video workloads. We do not state benchmarks or immediate availability: both depend on model and hardware.
Future module for audio generation and processing.
Where local AI stops being a tool and becomes part of the workflow.
Evolution toward a local AI automation platform. The goal is to reduce dependency on external AI automation SaaS, not to fully replace tools like Make or Zapier.
It supports agents rather than replacing them, and integrates with the existing help desk rather than supplanting it.
Commercial and document workloads on the same infrastructure.
Adapting models to the company's domain and language, connected to EuLLM Forge.
Looking ahead, authorised agents may interact with graphical applications, browsers and internal software to automate repetitive digital workflows. Subject to customer-defined permissions and security policies.
The core of the offering is entirely local: the platform default is on-premise and inference runs on site. When the customer requires it, however, individual workflows can integrate external AI services as explicitly authorised extensions. The difference is that this becomes a deliberate, bounded choice rather than the starting condition.
Most companies today pay for five or six separate AI services, each with its own contract, its own per-token price and its own data transfer. I3K Local AI can replace or reduce dependency on several of them, progressively.
We do not promise to replace every external AI SaaS. We promise an infrastructure to consolidate on, one workload at a time.
Inference runs on the hardware on site. No per-token cost, no billing surprise when usage grows. The cost is the hardware plus the software subscription.
The appliance is purchased by the customer, not leased. It stays a company asset even if you change software vendor later on.
We work with AMD, NVIDIA and other accelerators we have validated on our stack. If you have an explicit CUDA requirement, there is a dedicated configuration.
Start with a single appliance and grow toward multi-GPU or a cluster when concurrency demands it, without redesigning the architecture.
Two configurations at the same price can serve opposite purposes. The difference is not what it costs but what you do with it.
Large amount of available memory. Lets you load bigger models locally and work on extended contexts, favouring what fits in memory over responsiveness.
Dedicated GPU memory and higher bandwidth. Built for faster inference, interactivity and more concurrent workloads on the same machine.
Large memory and performance together, with ECC memory. For multi-user, multi-service installations where compute reliability is not negotiable.
Prices exclude VAT, are indicative and subject to confirmation. Every configuration includes initial installation, EuLLM, RAG Enterprise and 12 months of I3K AI Suite.
Ryzen AI Max+ 395 · 128 GB unified memory
from €5,900
Maximum memory capacity in the entry tier, with good energy efficiency. For large models, RAG over extensive corpora and document intelligence where keeping everything in memory is what matters.
32 GB dedicated GPU memory · AMD Radeon AI PRO R9700 or validated equivalent hardware
from €5,900
Same price bracket, opposite philosophy: less memory, more speed. For interactive AI, where response time matters more than model size.
64 GB aggregate accelerator memory
from €8,900
The centre of the range. It can pool the accelerators to serve larger models, or dedicate them to separate loads — LLM on one, vision, embeddings or agents on the other. Departmental AI infrastructure.
NVIDIA configuration · 32 GB
from €9,900
For workloads or applications that require CUDA, whether from proprietary libraries, existing pipelines or internal policy. Same software stack, NVIDIA ecosystem.
96 GB ECC GPU memory
from €22,900
For large-model inference, high session concurrency and multi-service enterprise workloads.
Up to 192 GB unified memory
On request
Planned configuration. For workloads that need even more memory capacity in a single node.
We do not publish tokens-per-second figures until we have I3K and EuLLM benchmarks measured directly on the final hardware. We would rather give real numbers later than convenient numbers now.
Hardware, initial installation, EuLLM, RAG Enterprise and 12 months of I3K AI Suite at the agreed tier.
No per-token fee for local inference.
From the second year the subscription is chosen based on the number of licensed users. Local inference carries no per-token cost at any tier.
€129/month
Up to 25 users
€399/month
Up to 100 users
€699/month
Up to 250 users
€1,190/month
Up to 500 users
€1,990/month
Up to 1,000 users
Above 1,000 users: Custom Enterprise Agreement.
The user count is about licensing, not concurrent sessions: actual concurrency is sized on the hardware and, for large organisations, across multiple appliances or a cluster.
Software maintenance, updates and L2/L3 technical support for the customer's IT team, according to the selected service plan.
The purchased appliance keeps working without renewal. With an active subscription you get:
New modules or new product lines may be licensed separately: the subscription does not automatically include every future I3K product.
Tell us your number of users, the workloads you want to bring in-house and your data requirements. I3K will recommend the appropriate Local AI configuration.
A few details are enough to start. We reply within two working days.
