Your Own ChatGPT, on Your Own Server.
Want AI for your team without sending company documents to an outside service? We install an open model such as Llama, Qwen, Mistral or DeepSeek on your server, sized for your users, with a company login and answers from your own files.
From $15 at store prices, no contract.
- Data stays on your server
- GPU servers from $222.99/mo
- Engineer time $28 an hour
Pick the Server, We Install the Model.
Installation is quoted once we know the model and the number of users. Servers and engineer time at store prices, shown in your currency.
Scripting Service
Python and shell scripts that load your documents, refresh the knowledge base and move AI results where they are needed.
Engineer time, 1 hour
Ollama not using the GPU, slow answers, a model update gone wrong: fixed by an engineer.
Server Administrator, 120 hours a month
A server administrator for 120 hours a month: model and security updates, monitoring and backups.
Private LLM installation
The model, Ollama or vLLM, Open WebUI with company logins, a key-protected API and your documents loaded, then handed over.
GPU dedicated server
From a Tesla P4 or RTX 3090 to RTX 5090 and RTX PRO 6000 Blackwell (96 GB) servers, for models from 8B up to 70B and beyond.
CPU VPS for small models
Small 7B and 8B models for light use, such as one person or a test, on a KVM VPS with enough memory.
How Much GPU Do You Need?
Rough starting points for 4-bit models, the usual choice. Every extra user and every longer conversation needs more memory, so we size it with you.
| Model size | Examples | Download | Memory to plan for |
|---|---|---|---|
| 7B to 8B | Llama 3.1 8B, Qwen 7B, DeepSeek R1 8B | About 5 GB | 8 GB GPU memory, or 16 GB RAM on CPU for light use |
| 14B | Qwen 14B, DeepSeek R1 14B | About 9 GB | 12 to 16 GB GPU memory |
| 24B to 32B | Mistral Small 3, Qwen 32B, DeepSeek R1 32B | 14 to 20 GB | 24 to 32 GB: one RTX 3090, RTX PRO 4000 or RTX 5090 |
| 70B | Llama 3.3 70B | About 43 GB | 48 GB or more: one RTX PRO 6000 (96 GB), or 4 x V100 (64 GB) split across the cards |
Download sizes from the Ollama model library. Larger models such as DeepSeek R1 671B need a multi-GPU server; ask us.
Private AI That Your Team Can Use.
Not just a model download: the parts people struggle with after installing one themselves.
The right model and licence
Chosen for your task and checked for commercial use: Qwen and Mistral are Apache 2.0, DeepSeek R1 is MIT, Llama has its own community licence.
Sized for your users
GPU memory planned for how many people use it at once and how long their documents are, so answers stay fast.
Ollama or vLLM
Ollama for simple setups and a few users; vLLM when many people or apps send requests at the same time.
Company login and a protected API
Open WebUI with staff accounts, and the model API behind a key and HTTPS, never open to the internet.
Answers from your documents
Your files indexed with the right settings, so answers use the whole document and not just the first part.
Connected to your tools
Linked to n8n workflows, AI agents and MCP, so the model can look things up and do simple tasks in your systems.
Systems We Work With
- Ollama
- vLLM
- llama.cpp
- Open WebUI
- Llama
- Qwen
- Mistral
- DeepSeek
- Gemma
- NVIDIA CUDA
- Docker
- Qdrant
- PostgreSQL
- n8n
- MCP
- Ubuntu
From Order to Done.
Tell us the use
Who will use it, for what, and on which documents.
Pick the server
Your own, or a Ucartz GPU server or VPS sized for it.
We install and secure it
Model, logins, API key, HTTPS and your documents.
Hand over and stay on call
Notes for your team, and help with updates.
Our engineers, your company name.
Agencies, hosts and IT firms resell or outsource our work as their own service. Your clients deal with you.
No Ucartz branding
We work under your name. Your clients see your company, not ours.
Resell or outsource
Add server administration, development, design or abuse handling to your own catalogue and bill it yourself.
NDA on request
We keep your client relationships and information confidential.
Scale without hiring
Take on more clients without adding staff: book hours, a task or a full month.
Every Ucartz service is 100% white label. Hosting, radio and video streaming run on white-label hostnames, and anything you buy from us, Hire an Expert work included, can be resold or outsourced under your own name.
Real Feedback From the Businesses We Support
See how our engineering support and infrastructure keep real businesses online.
Private LLM Installation plan, with engineers on call
Getting AI and automation into production takes more than hardware. We design and deploy the whole system on your own server, document it and hand it over. After that, you can call us per task or per hour, with no retainer.
Free Basic Managed Support
Every Ucartz server includes Free Basic Managed Support: complimentary, best-effort help with quick tasks of about 5 minutes, such as access checks, restarts, OS reinstalls and DNS pointing.
What is included →Plug-and-play expert help
Model updates, new workflows, scaling or fixes after handover. From $8 per 15 minutes or $28 per hour, with no monthly commitment.
See on-demand support →AI, automation & custom builds
Private LLMs (Ollama, Llama, Mistral), RAG over your documents, n8n and API automations, GPU/CUDA stacks and custom internal tools.
Plan a project →Frequently Asked Questions
Quick answers about the plans, support and billing.
Run AI on Your Own Server
An open model installed, secured and sized for your team, with your data kept private.