Self-hosted AI for developers and teams
For companies whose code and data can't go to cloud AI because of client contracts, regulation or their own security policy. We set up a coding agent for your developers and an AI chat for everyone, on open models, on a GPU server in your name in a data center you choose. Everything is configured as code, and we keep it running for a fixed monthly price.
02 / 08Developers and staff
Developers already use AI, even where it isn't allowed, and client code ends up in outside services. Here they give the task to a coding agent in the terminal or the IDE, and everyone else uses the company chat.
03 / 08Gateway
The Gateway checks sign-in, each person's access key and usage limits, and keeps an audit log. Only then does the request reach the model.
04 / 08Your server
It sits on your premises, or it's rented in your name in a data center you choose. Kubernetes runs the models, and the disk is encrypted with a key you hold.
05 / 08Your code
The agent reads your repository, edits the code and runs your tests. Code, chats and uploaded files go only to the model on your server. No cloud AI ever sees them.
06 / 08Appnestiq, from outside
Every setting lives as code in your Git repository. We roll out updates and new models over an outbound-only connection. We receive only technical operating data (load, errors), never content.
Management connected. We receive only load and error data.
If you leave, the system stays with you: it keeps running, and you get all the data and settings.
07 / 08The process
In the pilot, your developers try it on your own code, and we agree up front on what it has to achieve.
08 / 08The whole picture
Developers give the agent a task, and staff ask the chat. The Gateway checks who is asking and keeps an audit log. Open models do the work on a server in your name. We manage it from outside, over a connection you can cut at any time.
The service
+ Add-onPrefer OpenAI's Codex CLI? We're testing it with open models. Ask us.
Model running on your server. Nothing sent to a cloud AI.
Industries
For any organization whose code and data can't go to cloud AI: because of client contracts and NDAs, the Czech Cybersecurity Act (NIS2), financial regulation, professional secrecy or its own security policy.
You write code for clients whose contracts forbid sending it to outside AI services. Your developers get a coding agent anyway, on a server in your name.
Write unit tests for the discount calculation in billing/discounts.py.
billing/discounts.py
Added 6 tests for the edge cases: zero quantity, an expired code and the stacking limit. All 6 pass.
Firmware, embedded code and technical documentation that must stay inside. The agent and the chat run on your own hardware or a server in your name.
Developers on internal systems, and staff working with policies and client data. Both stay on a server in your name.
Control software, test scripts and internal tools that hold your know-how. Engineers ask in their own words, and nothing goes to a cloud AI.
Contracts and case files under professional secrecy. Lawyers use the chat on their own documents, on a server in your name.
How it fits together
A data center rents you a GPU server and looks after the hardware. That's the foundation, and we build on it too. Which models run on it, how well they work and whether your people use them is our job.
| Topic | The data center provides | We add on top |
|---|---|---|
| The tools | A server ready for software to be installed | A coding agent, an IDE plugin and a company chat, with company sign-in |
| Models | Not provided | Open models chosen for your hardware, set up, tuned and replaced only after testing on your tasks |
| Infrastructure | Hardware, power, network | Kubernetes, model serving and backups, all as code in a Git repository you own |
| Quality | Not provided | A quarterly review of the models on your own test tasks before anything changes |
| Security review | Depends on the data center | Documents for the AI system: architecture, access, models and their origin, incidents, exit plan |
| Management | Server administration if you order it | Updates, security patches and monitoring, over an outbound-only connection you can cut |
| If you leave us | Your rental contract, in your name | The system keeps running, and you keep the repository, data and access credentials |
| Price | Monthly server rent, paid by you directly | One-off setup, then a fixed monthly fee. No per-seat or per-token charges. |
Want to see it on your own code first? Pilot: 4 weeks, €6,000, half deducted from the setup if you continue. See pricing
Pricing
Each option shows the full cost: our one-off setup, our monthly service and the GPU server you rent in your own name from a data center you choose. Not sure which one fits? The pilot tells you.
Start here
€6,000one-off
Your developers and a few colleagues use the coding agent and the chat on your own code, on a GPU server rented in your name. We compare two models side by side. Server rent is paid separately.
Where should the server run?
A GPU server rented in your name from a Czech or EU data center you choose. We pick the configuration and arrange it. You sign with the data center and pay it directly.
Setup is paid once: we prepare the server, deploy Kubernetes, the models, the agent and the chat, connect company sign‑in, and train your admins and developers.
For One development team
Our service€1,600/ month
Coding agent, IDE plugin and company chat on a server with two GPUs. One model setup, chosen in the pilot.
For Several teams, plus chat for the company
Our service€3,000/ month
Everything in Team, on a server with four GPUs, so the agent and the chat each get their own capacity. Documentation for your supplier security review included.
For Large development organizations
Our servicefrom€5,500/ month
The largest open models on a server with eight GPUs, or several servers. Scoped and priced individually.
One fixed price for the tools, the models and someone who keeps them running.
FAQ
No. Open models are behind the best cloud models, most of all on long, complex tasks. They do well on scoped work: fixes, tests, refactoring, explaining code. The pilot measures the difference on your own code before you commit.
Your code would still go to an outside AI provider. We're for companies whose client contracts, NDAs or security policies don't allow that. If yours do, cloud tools are cheaper and smarter.
Prompts, code and chat history stay on the server. Sign-in goes to your identity provider, updates come from sources we pin and mirror, and we receive only technical health data such as load and errors.
We never receive your code or chats, only technical health data such as load and errors. Our administrators have logged maintenance access to the server, and you can revoke it at any time. The system keeps running without it, but we can't do maintenance until you turn it back on.
Open-weight models chosen and tested for your hardware and tasks. Today the strongest for coding come from Chinese labs, such as Qwen and GLM. They run fully on your server, with no connection to their makers. If your policy rules them out, we use Western models from Mistral, Google or OpenAI instead. You get a list of every model with its origin and license.
We're testing it with open models. Our default is OpenCode, which works reliably with them. If you need Codex, tell us in the demo.
Almost. The agent, the IDE plugin, model serving and Kubernetes are open source. The chat, Open WebUI, is source-available with a branding clause. The models are open-weight: you can download and run them, but their training data isn't published.
Per developer, no. Once you count company-wide chat as well, it can come out similar. You pay to keep AI inside your company, not for a lower price.
With us, you do rent it yourself, in your name. We pick the models, set up the agent, the chat and sign-in, test new models on your own tasks, apply security patches and keep it all running, for a one-off setup and a fixed monthly price.
A demo, yes, in a few days. Running it for your developers every day is the hard part: testing new models before switching, keeping model serving and the agent compatible, applying patches and answering the security review. We do that for a fixed monthly price.
The server is in your name, and the whole configuration is in your Git repository with written procedures. The system keeps running without us, and another team can take it over.
It runs on one server, so it isn't highly available, and support is on Czech business days. If you need more, we'll scope it with you.
Yes. We specify the server, you buy it for your server room, and we install and manage it. We don't sell hardware, so the hardware never ties you to us.
Companies that are allowed to use Copilot, Claude or ChatGPT and are happy with them. For them, cloud services are cheaper and smarter.
Team
There are two of us. You talk directly to the people who build it.
Sales and customer relations
Lukáš is the first person you talk to. He finds out what your developers and staff want to use AI for, sets up the demo, and agrees with you on what the pilot has to achieve.
LinkedIn (opens in a new window)lukashellesch.com (opens in a new window)
Engineering
Jakub designs the infrastructure, from Kubernetes and model serving to the agent setup, and keeps it running. He's your contact for technical questions and your supplier security review.
LinkedIn (opens in a new window)kubadvorak.com (opens in a new window)
Contact
Tell us what your developers and staff would like to use AI for. We'll reply within one business day and set up a 30-minute demo.
Or reach us directly
Request a demo
We reply within 1 business day
