Private team chat
Draft, summarise and analyse in a browser.
Your team works with AI on a machine you own, on your own network.
Rack-mount AI server for private, on-premises AI
Private AI for the server room: shared across your teams, managed by IT and kept on your own network, with RBOS Studio ready on day one. Renderboxes can help you put powerful local AI compute to work.
The possibilities with private, local AI
Ask your own documents
Your files answer back, with sources.
Contracts, briefs, manuals and project files, searched and summarised without leaving the building.
Private AI agents
Agents and coding assistants that stay in-house.
Agentic AI workflows and a coding assistant for your developers, working on your files without sending them anywhere.
Meetings, transcribed in-house
Recordings into notes and actions.
Transcription and summaries without uploading calls to a third party.
The newest open models
Cutting-edge open-weight models, on your hardware.
Run the latest open models, which now compete with the leading cloud services, or a fast everyday model alongside a flagship.
Hybrid, on your terms
Local first, cloud only when it counts.
Send only the hardest tasks to a frontier cloud model, with names, numbers and confidential details masked before anything leaves.
Local AI capabilities, model support and performance vary by model, configuration, software and workload.
Protect your business
We can help you build AI workflows that protect your business.
- Help keep your data secure. Prompts, documents and answers are processed on your own hardware, on your own network. Nothing is retained by an AI provider and nothing trains someone else’s model.
- Your identity, protected. Personal and confidential details can be masked automatically at every exit, so only what is safe to send ever leaves the building.
- A record you own. Every request and action can be logged to a tamper-evident audit trail on your own disks, which helps you meet client confidentiality and GDPR obligations.
Reduce token costs
PREDICTABLE COSTS FOR THE WHOLE TEAM
Replace per-seat AI subscriptions and usage-based token fees with a fixed cost for on-premises compute your whole team can share. Workflows can be set up to check your own knowledge before a paid cloud call, and any cloud use can be capped, with a kill switch that you control.
Ready on day one: RBOS and RBOS Studio
Skip the weeks of set-up. RBOS and RBOS Studio can come preloaded, configured and tested together, so your team starts on the work, not the software.
- One-click workflows. Runbooks set up team chat, document search, coding assistants, transcription and model serving with step-by-step guidance.
- Your team gets simple, IT keeps control. Everyone works in a browser on your network, while admins decide who can use what and every action is logged.
- Change it as your work changes. Add or remove workflows at any time, with no endless tweaking.
- See every GPU at work. Assign workflows to GPUs and see what each one is doing.
Works with the software you use
We can set up, test and support popular AI software on your machine before it ships, including:
Open WebUI · Ollama · llama.cpp · vLLM · SGLang · Docker
Using something else? Tell us and we will help you make it work.
Built on NVIDIA RTX PRO Blackwell
These workflows rely on the NVIDIA software ecosystem, so we build them on NVIDIA RTX PRO Blackwell GPUs, with up to 96GB of VRAM on each card.
- The broadest software support for creative, training and research AI tools.
- Workstation-class GPUs built for sustained load, all day.
- Room to grow from one GPU to eight as your work does.
Recommended builds
Rack-mount private AI – NVIDIA
Atom Rack · 2x NVIDIA RTX PRO 6000 Max-Q 96GB
192GB VRAM for flagship open models, long documents and more people at once.
- CPU: AMD Ryzen Threadripper 9970X 32 Core
- RAM: 4 × 64GB RDIMM ECC DDR5 6400
- PCIe slots (16x): 2 × Nvidia RTX Pro 6000 Blackwell Max-Q 96GB
- NVMe M.2 Drives (Gen 5): Samsung 9100 Pro Gen 5 2TB
- NVMe M.2 Drives (Gen 5): Samsung 9100 Pro Gen 5 8TB
- Operating System: RBOS Ubuntu LTS
£51,506.37 inc VAT
£42,921.99 ex VAT
Built to order · dispatched in 10 working days
Each recommended system is a starting point. Select Customise to add components or change the specification before you order.
Inside the Atom Rack
- Two GPUs in a rack-mount chassis for the server room.
- AMD Ryzen Threadripper, ECC memory and fast networking.
- Cooling engineered for sustained AI work, around the clock.
The machine underneath
- 1In the largest build here, with NVIDIA RTX PRO 6000 Blackwell GPUs. Total VRAM depends on the GPUs specified.
- 2Compared with a single-GPU Photon fitted with the same GPU. Real-world performance varies by model, configuration, software and workload.
- 3Plus 1-year return-to-base labour and 3 years of technical support.
Scale up as your team grows
2-GPU private AI
The Atom: 2× the GPUs and VRAM of a Photon4, for a team getting started with private AI.
Explore
4-GPU private AI
The Nano Pro: 4× a Photon, desk-side, for heavier models and more people at once.
Explore
8-GPU AI compute node
The Molecule Air: 8× a Photon, for whole-company AI, agents and fine-tuning.
Explore
- 4Compared with a single-GPU Photon fitted with the same GPU, counting GPUs and total VRAM. Real-world performance varies by model, configuration, software and workload.
Why Renderboxes
- Engineered for demanding AI work. Built and stress-tested in the UK for generative AI, agents, document search and fine-tuning, running all day.
- Software ready. RBOS and RBOS Studio prepared for your AI work before delivery.
- Secure and reliable. Your models and data on hardware you own, supported by the people who built it.
- Future-proofed. Upgrade GPUs, memory and storage as your work grows, or move up to four or eight GPUs.
Frequently asked questions
Tell us your rack depth, power and cooling, and we will confirm the fit and the right configuration before you order.
No. RBOS Studio runs in a browser, and Runbooks guide you through set-up step by step. We can also set the machine up for your workflows before it ships.
Not unless you choose to. Everyday work runs on the machine. If you use a cloud model for harder tasks, personal and confidential details can be masked before anything is sent.
It can. You could keep the machine for private and everyday work and use a cloud model only for the hardest tasks, with spending capped. We can help you set that up.
That depends on your workflows and how heavily they are used. Tell us about your team and we will size the machine for it.
Many of the tools behind these workflows are built for NVIDIA first, so we build them on NVIDIA RTX PRO GPUs for the widest software support.
Yes. We keep developing new workflows, and if you need one we do not list, talk to us and we will help you make it work.
Yes. Select Customise and the configurator opens with this build already selected.
Tell us what you’d like your team to do with private AI
Tell us what your teams want to do with private AI, how many people will use it, and your rack depth, power and cooling, and we will recommend the right build.
Talk to someone at Renderboxes
Not sure which build is right? Speak to our team and we will help you choose the right configuration for your work.
Send us an enquiry
Tell us what you are working on and we will recommend a configuration.
Contact us