English edition · Czech edition is available at /cenik/
PRICING / TRANSPARENT ENTRY
We first define the specific work problem, data, accountability and value metric. Only then do we choose the model, infrastructure and scope. Below are realistic entry budgets for typical engagements.
All prices are indicative starting amounts from and are normally shown excluding VAT. Third-party licences, travel, extensive data or system integrations and non-standard SLAs are priced separately.
01 / CORE PACKAGES
The packages are designed so a company does not buy a platform before validating the use case. Each next phase follows only from evidence produced in the previous one.
Expert definition of the problem, data, risks, baseline and economics. The output is a prioritised, actionable 90-day plan.
fromCZK 49,000excl. VATRECOMMENDED START
One specific workflow validated on real data, with an evaluation set, human oversight and measured value.
fromCZK 229,000excl. VATProduction-grade enterprise AI environment with integrations, identity governance, model gateway, monitoring and operating documentation.
fromCZK 590,000excl. VAT02 / EXPERT CAPACITY
Hourly work is suitable for clearly bounded expertise. Implementation is normally priced as a package or sprint so the budget is tied to an outcome.
Architecture, models, local inference, RAG, agents, evaluation or an independent technical review of a supplier.
Up to 4 hours including preparation, facilitation and a concise output with priorities and next steps.
Up to 7 hours including preparation, workshop and structured recommendations for management or the technical team.
Reserved senior and implementation capacity up to 32 hours per month, priority scheduling and ongoing coordination.
03 / OPERATIONS & INFRASTRUCTURE
The price covers operational responsibility: gateway, monitoring, backups, updates, evaluation, incident process and dedicated support. Compute hardware is only one component.
Monitoring, updates, regression evaluation, monthly operations review and basic incident capacity.
Dedicated 96GB GPU capacity with governed access, monitoring and an agreed operating limit. Minimum commitment: 6 months.
Urgent work, changes outside the agreed scope, non-standard migrations or incident response beyond normal support.
04 / OPERATING MODEL COMPARISON
For a fair comparison, we show typical monthly cost models. This is not a universal price list for every provider; it is an indicative scenario for a team working with documents, analysis and repeated queries throughout the month.
One standardized A4 page = 1,800 characters including spaces. The default five-person scenario assumes approximately 7,500 A4 pages read a 750 A4 pages generated. This is a modelled workload, reviewed 7 August 2026; it is not a provider quote. One page can be processed multiple times during retrieval, checking and summarisation. Public chat products are modelled as per-user licences. Own hardware includes GPU, server, electricity, monitoring and a share of an operator’s time.
Typically priced per user. Fast to start, but with less control over data, workflow and auditability.
Cost grows with every document load, summary, check, agent step and longer context. We therefore express the workload in standardized A4 pages rather than provider-specific technical units.
The GPU alone looks inexpensive on paper. Real operations also include the server, electricity, backups, monitoring, security and admin/DevOps time.
Managed operations for an existing deployment start at CZK 39,000 per month. If dedicated 96GB capacity is also required, a typical fully managed model starts at CZK 78,000 per month excluding VAT.
1× NVIDIA RTX PRO 6000 Blackwell 96 GB, using the supplied acquisition price: CZK 22,978 incl. VAT, approximatelyCZK 18,990 excl. VAT a CZK 528 / month when spread over 36 months.
Own hardware makes sense when it is surrounded by process, security and operational accountability. Otherwise the company has bought a component, not an AI operating system.
05 / INDICATIVE ECONOMICS
The calculator converts a conservative time saving into indicative economic value. It excludes quality improvements, faster response, lower risk and new revenue.
Calculation: users × hours saved × 4.33 weeks × hourly cost × realised benefit. This is a scenario, not a guaranteed saving.
06 / COMMERCIAL RULES
The goal is to remove ambiguity rather than shift the risk of an undefined brief onto the customer or supplier.
AI Compass: 100% before start. AI Pilot: typically 50 / 30 / 20%. Private AI: typically 40 / 40 / 20%. Retainers and operations are paid monthly in advance.
A new system, data source, language, role or SLA is described and priced first. The original budget does not change without approval.
We deliver the agreed outputs, test evidence and transparent limitations. Accuracy, savings or ROI cannot be guaranteed without specific data and scenarios.
07 / FAQ
Most value and cost come from process analysis, data integration, evaluation, security, operational responsibility and support. Hardware matters, but it does not create a reliable enterprise system by itself.
Yes. For a clearly bounded expert question, a minimum two-hour senior consultation is available. For an implementation decision, an AI Compass with a concrete output is usually more useful.
For a typical single workflow, yes, after a short qualification. Scope increases with multiple systems, strict data boundaries, non-standard integrations, multiple languages or high-availability requirements.
No. Compute and usage have an agreed limit based on the model and workload. Overages are handled by increasing capacity or using a transparent usage model.
At the supplied RTX PRO 6000 Blackwell purchase price, GPU amortisation alone is only about CZK 528 per month excluding VAT. Real operations add the server, electricity, backups, monitoring, gateway, security and at least part of an operator’s time for updates and incidents. That is why a realistic monthly model for a single-GPU platform is closer to CZK 24,000–35,000 than to the hardware amortisation alone.
Briefly describe the process, number of people, data sources and sensitivity. We will recommend the smallest scope that can produce credible evidence.