37 employees for the price of one. Your office. Your server. Your data — and 37 roles, that do the work for you. Data never leaves your company. You pay once 499 000 CZK — including the hardware, installation and training. The same work would cost you 2 160 000 CZK every year.
No commitment, no charge. We'll get back to you within two business days. Your own on-premise server starts at 189 000 CZK — all four configurations.
Our data. Our rules. Our future.
Either you pay for operation on our servers in the Czech Republic, or you purchase the hardware and set it up on your own premises. The same MADLA, the same Czech LLM — the only difference is where the hardware sits and how you pay.
You launch MADLA straight away. We handle the GPU, backups, audit, encryption and updates. The instance is yours, the hardware is ours. Your data never leaves to a foreign cloud.
We deliver a complete solution with a server to your company or public office. The LLM runs behind your firewall, internal data never leaves the building. You pay once.
229tools the model calls on its own
838 479court decisions
17 679legal regulations from e-Sbírka
3 482 845companies on a live map
Each role has one specific job, its own rules and its own tools. You don’t ask them questions — you assign a task and they call the tools themselves. Seven groups, and the last one is the one that makes the difference.
The analysis will produce a ranking — what to deploy first and what isn’t worth deploying at all.
Free analysis →It will invent a clause. Miscalculate a number. Claim it verified something — and verify nothing. No one notices until someone is riding on it.
That’s why every team has a control group. The Opponent breaks down the conclusion. The Number Watchdog recalculates what the model claims. The Evidence Collector traces a source for every statement. The Reality Check verifies that it holds up beyond the screen.
For every one who does something, there is one who takes it apart. There are deliberately four more controllers than performers — oversight is not an add-on, it is half the team.
In our benchmark Skutek 2026 seven systems tackled 971 Czech work tasks across 14 disciplines — company due diligence, sanctions, law, property prices, bulk processing. Every task had a correct answer recorded in advance from a primary source. The CzechAI stack on which MADLA runs finished first.
MADLA is not just a chat. It is your own AI environment that will learn exactly what you need — your work, your documents, your services, your processes and your rules. Not another browser window, but AI you truly have at hand.
It learns your work and your documents. You have it on your computer and on your phone, it remembers what you have done together, and it speaks Czech.
It knows your processes, products, customers and rules. Every employee has their own space, and the company has one shared brain.
This is what MADLA looks like — light and dark theme side by side. Browse with the arrows, click the screen to enlarge it.
Registry, regulations, contracts, budget. Files and personal data stay with you — the server can be located directly in your building.
The model runs behind your firewall, official data never leaves the office. Dedicated offer for the public sector →
You can offer MADLA further — to your customers, members or branches. Get in touch and we will work out how to do it.
And every MADLA can be different — trained for a specific environment. It can run online, remotely on our server, directly on your server or on a ready-made AI server, which we will deliver to you ready to use.
All are one-time payments with no monthly fees. All include delivery, on-site setup and staff training. The difference lies in performance — and in how many people can ask questions simultaneously without anyone waiting.
The most affordable path to AI that keeps your data on-premises. Quiet, fits under a desk.
Who it's for: office, clinic, attorney, tradesperson
When to go higher: The Entry tier is built for documents, answering questions, and search. The Coder is included here too. For heavy programming work, however, 32 GB on a single card isn't enough memory — such tasks can, if you allow it, be sent to our server in the Czech Republic. Anyone who wants to do full-scale programming on their own hardware moves up a tier — data and settings stay with you.
Two cards, 64 GB of memory. Chat and coding assistant run simultaneously.
Who it's for: company with in-house development or IT
When to go higher: Two cards provide memory, not double the speed. For fast responses to many users, images, or advanced programming, go with Max.
Complete AI office: a team that does specific work and checks it itself.
Who it's for: for those who want the entire team ready from tomorrow morning
128 GB of GPU memory. Everything runs at once, nothing switches.
Who it's for: an entire department, or development alongside chat
All configurations include delivery, on-site setup and staff training. Click on the image to open the full configuration sheet.
Don’t want to buy a machine? Use MADLA in our cloud — all you need is a browser or mobile. You pay monthly per user and every month you receive a credit for working with AI.
Everything MADLA can do, except coding.
10 000 credits included every month
MADLA plus your own dedicated developer.
15 000 credits included every month
The full MADLA experience for your team, with administration and a company knowledge base.
20 000 credits included every month
For businesses that also develop their own software.
30 000 credits included every month
With an active plan, you can purchase additional credits with a single click in the app (Credits menu) and pay via PayU. They will be added to the credits included in your plan.
Applications in all plans: web browser (desktop & mobile) and the MADLA mobile app for Android.
Companies can order up to 20 users at once; for more than 20 users we will prepare a tailored offer.
Top up credits: 10 000 credits for 490 CZK. With annual billing you get 2 months free.
The cloud version also works with AI models outside the Czech Republic —
if you need your data kept exclusively on your own infrastructure, choose one of the on-premise configurations above.
Real app screens: chat and agents, images, Finder and voice MADLA. Click on the panel and then on the phone — the screen will enlarge.
MADLA is fully integrable with most systems. We can connect with any existing system or API by agreement — you don't need to change or discard anything.