We power.Hyrax AI-Ready Computers
Get a laptop or desktop selected for your needs, configured with powerful AI models, and ready to work. Hyrax handles the hardware, installation, training, and support.
Chosen for the AI work you want to do, not the biggest number on a spec sheet.
Open-weight models set up, tested and ready when you switch on.
Practical apps and workflows configured for your tasks.
We show you how to use it, and stay on hand afterwards.
Six quick questions, no technical knowledge needed. You'll get a recommendation you can send straight to us for a quote.
The finder gives starting recommendations. It doesn’t check stock or prices, and we confirm every specification in your quotation.
Recommended starting specifications. Every configuration is quoted for your needs.
Students, teachers, office workers, entrepreneurs and everyday professionals.
Researchers, developers, creators, consultants and advanced professionals.
Development teams, AI engineers, laboratories and specialist professionals.
Businesses, universities, NGOs, hospitals, government agencies and research institutions.
Which models run well depends on memory, usable GPU memory, operating system, quantisation, context length and the task. We don’t publish fixed prices or stock levels: hardware prices in Nigeria change often, so every quotation uses current prices.
You don't always need new hardware. We assess Windows, Mac and Linux computers and recommend upgrades or software that make them useful for AI.
Tell us what you know. “Not sure” is fine for any detail, and we confirm compatibility in a technical assessment before you pay for anything.
Open-weight models are AI models whose trained weights are published, so anyone can download them and run them on their own hardware. Running a model on your computer is called local inference.
People choose local AI to keep documents on their own machines, to work without reliable internet, and to avoid per-use fees. The trade-off is that your hardware sets the size of model you can run.
Answering questions, writing and summarising. Almost every model can do this.
The model asks your software to do something, like search a folder or run a calculation, and uses the result.
The model plans several steps, calls tools in sequence and recovers from errors. This needs stronger models.
Acting without a person checking. We don't set models up this way: important actions always need human approval.
| Model | Good for | Agent use | Inputs | Hardware | Licence |
|---|---|---|---|---|---|
| Gemma 4 E4B (official model page, opens in a new tab)GoogleFrom Hyrax AI Essential | Compact assistant for chat, writing and image or audio questions | Conversation and simple tool useTool calling documented | Text, image, audio in · text out | Designed by Google for mobile and edge devicesProvider guidance | Apache 2.0Checked 23 Oct 2026 |
| Qwen3.5-9B (official model page, opens in a new tab)Alibaba QwenFrom Hyrax AI Essential | Everyday assistant with long context and image understanding | Conversation and simple tool useTool calling documented | Text, image, video in · text out | About 6–8 GB of memory at 4-bitHyrax estimate | Apache 2.0Checked 23 Oct 2026 |
| Ministral 3 (8B and 14B) (official model page, opens in a new tab)Mistral AIFrom Hyrax AI Essential | Small dense models with reasoning variants and image understanding | Conversation and simple tool useTool calling not documented | Text, image in · text out | Mistral lists RTX PCs and laptops among target devicesProvider guidance | Apache 2.0Checked 23 Oct 2026 |
| gpt-oss-20b (official model page, opens in a new tab)OpenAIFrom Hyrax AI Essential | Reasoning model with adjustable effort and native function calling | Tool calling in supervised workflowsTool calling documented | Text in · text out | Runs within 16 GB of memoryProvider guidance | Apache 2.0Checked 23 Oct 2026 |
| Gemma 4 26B A4B (official model page, opens in a new tab)GoogleFrom Hyrax AI Pro | Mixture-of-experts model with about 4B active parameters, so it runs quickly for its size | Tool calling in supervised workflowsTool calling documented | Text, image in · text out | About 16–18 GB of memory at 4-bitHyrax estimate | Apache 2.0Checked 23 Oct 2026 |
| Qwen3.8-27B (official model page, opens in a new tab)Alibaba QwenFrom Hyrax AI Pro | Strong dense model for reasoning, coding and agent work, with image and video input | Tool calling in supervised workflowsTool calling not documented | Text, image, video in · text out | About 18 GB at 4-bit; a 24 GB GPU or 32 GB+ Apple unified memory leaves room for contextHyrax estimate | Apache 2.0Checked 23 Oct 2026 |
| Gemma 4 31B (official model page, opens in a new tab)GoogleFrom Hyrax AI Workstation | Google's largest Gemma 4 dense model, for higher-quality local answers | Multi-step agent orchestrationTool calling documented | Text, image in · text out | About 20 GB at 4-bit; best on a 24 GB GPUHyrax estimate | Apache 2.0Checked 23 Oct 2026 |
| gpt-oss-120b (official model page, opens in a new tab)OpenAIFrom Hyrax AI Enterprise | Large reasoning model with native function calling, suited to shared servers | Multi-step agent orchestrationTool calling documented | Text in · text out | Runs on a single 80 GB GPUProvider guidance | Apache 2.0Checked 23 Oct 2026 |
| Mistral Small 4 (official model page, opens in a new tab)Mistral AIFrom Hyrax AI Enterprise | 119B mixture-of-experts model (6.5B active) with native function calling and JSON output | Multi-step agent orchestrationTool calling documented | Text, image in · text out | Server-class: Mistral's own example serves it across two GPUsProvider guidance | Apache 2.0Checked 23 Oct 2026 |
“Agent use” is Hyrax guidance, not a provider claim. Hardware estimates assume 4-bit quantised weights and short to medium context. We check each model’s licence against your intended use before installing it, and new models are added as they’re released and tested.
Letters, reports, minutes and proposals, without sending text online.
Point the AI at a folder of PDFs and ask questions with page references.
Explain, write and review code inside your editor.
Agents that sort files, extract data or prepare reports, and ask before acting.
Core AI tasks keep running when the internet doesn't.
Read charts, forms and photos with vision-capable models.
For organisations
A professionally configured AI server in your office or a private host gives staff a secure AI assistant that knows your documents. For more than a handful of users, one shared server usually costs less than a high-end laptop for everyone.
Installing an open-weight model doesn’t by itself make a system private, secure, compliant or offline. Those come from how the whole system is configured, maintained and used, which is the work we do with you.
Every system can include hands-on onboarding matched to your skills and what you want to do.
Write, summarise and plan faster, and know when to check the AI's work.
Search, compare and cite your own documents.
Set up and use coding assistants that run on your machine.
Design agents that use tools safely, with a person in the loop.
Policies, administration and staff onboarding for a shared system.
You receive a documented system: what's installed, how it's configured, and how to use it.
Use the finder or the form, or message us.
A suitable hardware and AI configuration, explained.
Specification, software, training and support, itemised.
A new machine, or upgrades to yours.
Models and apps checked on your actual hardware.
Hands-on onboarding matched to your skills.
The support you agreed, for the period you agreed.
A computer with enough memory, storage and processing power to run AI models on the machine itself, set up with those models and the apps to use them. Hyrax chooses the hardware for your tasks, installs and tests the software, and trains you.
Yes, for the core tasks. Once a model is installed, chat, writing and document questions can run fully offline. You still need the internet now and then to download updates and new models, and some tools such as web search need a connection.
Most recent laptops can run small models, but how well depends mostly on memory. With 8 GB, results are limited. 16 GB is a practical minimum, and 32 GB or more opens up much stronger models. A dedicated GPU or Apple Silicon unified memory makes a big difference to speed.
Often, yes. We can add memory or storage where your model allows it, and install and configure AI software. Many laptops have soldered memory and no way to add a GPU, so we assess your machine first and tell you honestly what is possible.
Yes. We can set up retrieval, which lets a model search your documents and answer with references. The documents stay on your hardware. Answers should still be checked, because models can misread or miss information.
An agent can carry out multi-step tasks using tools you allow, such as reading files, filling in spreadsheets or preparing reports. We configure agents to ask for approval before important actions like sending, deleting or paying, and to work only with the folders and tools you choose.
Running a model locally means your prompts and documents aren't sent to an AI provider. Privacy also depends on how the computer is secured, which apps have internet access, backups and who can log in. We configure these with you, but no setup can guarantee privacy or regulatory compliance on its own.
Yes. Every system can include hands-on onboarding matched to your skills, from everyday productivity to building agents. Organisations can request custom training for teams, and the Hyrax AI Academy at learn.hyrax.ng lists our current courses.
Yes. For more than a few users, a shared AI server is usually more cost-effective than a high-end laptop for everyone. Staff sign in through a secure interface, and we set up accounts, permissions, backups and training.
Open-weight models improve often. With a maintenance plan, we test new models on your hardware and install the ones that fit and that are licensed for your use. Your hardware sets the limit on how large a model you can run.
Tell us what you need. We'll reply with questions or an itemised quotation for hardware, software, training and support.
Looking for servers, labs or other hardware? See all of AI Infrastructure.
