Choose what goes in.
Your school approves the models, documents and users. Start with selected teaching resources, not an unrestricted drive full of sensitive files.
Brightlab
brightlab.Made for curious mindsbrightlab.Made for curious mindsBrightlab
NVIDIA DGX Spark. Managed by Brightlab.
Private AI for your school. Real models, useful tools and hands-on lessons. Set up and kept up to date by Brightlab.
Run selected models on your own network. Keep prompts and documents local, with access and retention rules agreed with your school.
See how local privacy worksYour school network
Local modelsYour school approves the models, documents and users. Start with selected teaching resources, not an unrestricted drive full of sensitive files.
Local inference does not require sending prompts to a cloud model. Logs, retention and any external integrations are reviewed as part of the deployment.
Updates and remote support use an agreed access process. Local hardware still needs sensible security, human review and school policies.
Start with a real job. Then explore the AI behind it.
Search a selected collection of school resources and trace answers back to their source. Teachers choose the documents. People check the result.
Explore AI investigationsIllustrative workflow · not a live model
What should we bring for the field trip?
Bring a hat, water and a notebook.
Year 8 field guide · p. 2
Excursion checklist · p. 1
These are example workflows we can scope for your school. Model output needs human review; the examples shown here are illustrations, not a live AI service.
A managed service that grows with your school.
Hardware supply, setup, local models and a first useful workflow.
We work with your IT team on installation, staff and student access, approved data sources and a practical starting point.
Planned maintenance, compatibility checks and recovery options.
Operating system, drivers and the model-serving software are maintained together. Update windows and recovery arrangements are agreed with your school.
Evaluate new releases before agreed rollouts.
We assess suitable open-weight models for usefulness, licensing, hardware fit and your use cases. Updates are tested before they replace a working setup.
New lessons, Marimo notebooks and teacher onboarding.
Connect the hardware to Brightlab investigations, add new learning activities, and help teachers turn a working model into something students can understand.
Enough capability for real local AI work: running open-weight models, searching documents, developing assistants and exploring how models work. We select for useful performance on your tasks, not just the largest model that fits.
NVIDIA hardware specificationsHardware specifications: NVIDIA DGX Spark User Guide, checked 19 September 2026. Storage and supplied configuration are confirmed in your quote. Model size, response speed and simultaneous users depend on the workload. Power, ventilation and a suitable school network are part of the installation plan.
Tell us what you want to teach, build or make easier. We’ll shape the hardware and service around it.
Start with a short setup brief. It stays in your browser until you choose to download or share it.
For configured local workflows, model inference, prompts and your document collection can stay on the school network. Model downloads, software updates and authorised support need separate connections. Optional cloud tools change the data path and must be explicitly agreed. We document access, logs and retention as part of the setup.
Students can access an agreed local interface from their existing devices. The number of simultaneous requests depends on the model, context length and task. We plan for your lesson and test the expected workload; one box does not mean unlimited concurrent users.
The proposal separates hardware supply and setup from ongoing maintenance, model updates, learning materials and support. It defines the update cadence, support hours, remote-access permissions, backup responsibilities and hardware warranty arrangements before you commit.
Supported local workflows can run after their models and dependencies are installed. Downloading new models and lessons, software updates, remote support and any cloud integrations require a connection. Offline use is planned and tested for the tools you choose.
BrightLab Spark is our collection of research investigations and companion notebooks. Managed Spark is the hardware and service offering on this page. They work together, but you can explore the research lessons without buying hardware.
The hardware configuration, installation and ongoing service are scoped to your school. A proposal confirms pricing, availability, support and delivery. There is no checkout or automatic subscription on this page.
Brightlab provides the managed service described in your proposal. NVIDIA DGX Spark is NVIDIA hardware; this page does not imply NVIDIA endorsement or partner status.