For companies with 50 to 500 employees

Your own AI assistant. On your network, without a usage-based subscription.

A chat and coding assistant for your staff running on your own NVIDIA DGX Spark, Stack of Sparks, Mac Studio or RTX hardware. In a fully local configuration, models, documents and prompts remain inside your network. You work with me directly, not with a sales team.

Book an intro callSee the packages
Entry configurations under €10,000 one-offincluding suitable hardware, setup and training. Exact price and timing depend on configuration, integrations and delivery; local inference has no usage-based token fee.
AWS CERTIFIED SOLUTIONS ARCHITECTAWS CERTIFIED DEVOPS ENGINEERDEVOPS SINCE 20175.0 / 5 FROM 40 REVIEWS
CLOUD AI VERSUS LOCAL AI

The difference is not quality. It is who carries the responsibility.

Data processing
CLOUD PROVIDER
In the provider's infrastructure, in the EU or a third country depending on the contract
LOCAL, AT YOUR SITE
On your infrastructure; outbound connections can be technically restricted
Contracts
CLOUD PROVIDER
Usually a processing agreement; subprocessors and transfers must be reviewed
LOCAL, AT YOUR SITE
Less external processing; agreements are still needed if service providers access personal data
Cost model
CLOUD PROVIDER
Per user, per token or flat-rate, depending on the service
LOCAL, AT YOUR SITE
Investment plus power, maintenance and optional operations; no token fee
Model changes
CLOUD PROVIDER
Depends on the provider's contract and controls
LOCAL, AT YOUR SITE
Approved by you, tested and documented
PACKAGES

Three levels

Every package includes installation, models, an access concept and training. NVIDIA systems or Apple Mac Studio, whichever fits your use cases and server room. You either buy the hardware yourself or source it through me. The price is fixed after the intro call. Local inference has no usage-based token fee; power, maintenance and an optional operations service remain ongoing costs.

PACKAGE 01
Pilot
One use case, one team. Proves internally that it works before the bigger investment.
  • NVIDIA DGX Spark or Mac Studio
  • One use case in production
  • Chat interface for up to 25 users
  • Handover session for your IT
Fixed price after the intro call
MOST CHOSEN
PACKAGE 02
Department
Several use cases, connected to your systems. The environment the company works with daily.
  • Stack of Sparks, Mac Studio or RTX server
  • Search across your own documents
  • Directory and role integration
  • Training for IT and business teams
  • Operations manual and documentation
Fixed price after the intro call
PACKAGE 03
Operations
Setup plus ongoing operation, for companies that do not want to assign their own team.
  • Everything in package 02
  • Maintenance and model updates
  • New use cases added over time
  • One named contact
Monthly flat fee
DGX Spark
DGX Spark
128 GB unified memory, desktop format, quiet enough for an office.
Stack of Sparks
Stack of Sparks
Several compact systems, neatly stackable and scalable when one computer is not enough.
Mac Studio
Mac Studio
Mac Studio with an Ultra chip and up to 512 GB unified memory. Substantial model capacity in a compact, quiet system.
RTX systems
RTX systems
Workstations and servers when existing hardware should be reused.
WHO BUILDS IT

Dimitri Tarasowski

AWS Certified Solutions Architect and AWS Certified DevOps Engineer. I have been building cloud and infrastructure environments since 2017, and for several years I have taught cloud and DevOps in year-long programmes with more than 20 participants per course, in German and English.

A local AI environment is the same work: hardware, Linux, networking, access concepts, automation and documentation. Not a product that only runs while someone stands next to it.

You talk to me from start to finish. I build it, train your people and step back, or stay on for operations if that is what you want.

LinkedInGitHubResources
Dimitri Tarasowski
"He never solved problems only on the surface, he always looked for solutions that hold up in the long run."
CHRISTIAN HEIMKE, PROGRAMME LEAD (TRANSLATED)
HOW IT WORKS

Four steps to a running environment

From the first call to handover to your IT. Every step produces something you can check.

01
Intro call, 45 minutes
Use cases, number of users, constraints. Free of charge.
02
Fixed-price proposal
Hardware, models, integrations and dates in writing.
03
Setup on site
Installation, models, access concept, integration with your systems.
04
Handover and training
Documented, reproducible, operable by your own IT.

Common questions

Are local models good enough?
Open models can be sufficient for document search, summaries, drafting, classification and code assistance. We test whether their quality, speed and context length meet your requirements using representative data.
Do we need dedicated staff?
Whether Linux experience alone is enough depends on your availability, security and integration requirements. I train your IT team and document the environment. Operations can be contracted separately if you prefer.
Who buys the hardware?
Either way works. You buy it and I build on top, or you source the systems through me and get everything from one place.
How long does setup take?
The first use case is usually in production a few weeks after the hardware arrives. Delivery time is normally the longest item.
YOUR RISK

The first step costs you nothing but 45 minutes.

The intro call is free
No deck, no sales pitch. If local hardware does not make sense for you, I will say so.
Fixed price, no change orders
What the proposal says is what you pay. If I misjudge the effort, that is my problem, not yours.
Acceptance, not blind trust
The agreed use case runs at handover, or I keep working until it does.
No lock-in afterwards
The environment is yours, documented and with no licence back to me. Ongoing operation is an offer, not a condition.

Let's spend 45 minutes on which package fits you.

Free, and not a sales call. If local hardware does not make sense for you, I will say so.

Pick a time