Question 1

What is AI-in-a-Box?

Accepted Answer

AI-in-a-Box is an on-premise appliance that runs the full Understand Tech stack inside your own network. It ships in two configurations, Team on NVIDIA GB10 and Organization on NVIDIA GB300 Blackwell Ultra, and includes the platform (Understand Studio app builder, assistants, multi-LLM portals, RAG, APIs), local storage, container monitoring and update services designed for restricted environments.

Question 2

Is it a hardware product or a software product?

Accepted Answer

It's software-defined (the Understand Tech platform + local inference stack), deployed on NVIDIA-certified hardware. Depending on your commercial model, hardware can be:
• Bundled/provided as part of the offer, or
• Purchased by the customer from NVIDIA / partners, while Understand Tech provides the full software stack and deployment support.

Question 3

What problems does AI-in-a-Box solve?

Accepted Answer

• No cloud required: run AI entirely inside your perimeter
• Air-gapped / offline ready for regulated or classified environments
• Data sovereignty: documents, embeddings, and logs stay local
• Enterprise operations: multi-user, audit logs, RBAC, admin control
• Production deployment: pre-packaged stack + support + update mechanism

Question 4

Does any data leave the customer environment?

Accepted Answer

By default: no. AI-in-a-Box is designed for zero external network calls (depending on your configuration). Data (documents, embeddings, metadata, logs) remains stored locally.

Question 5

Can it run fully offline / air-gapped?

Accepted Answer

Yes. AI-in-a-Box supports offline operation, including offline update procedures (e.g., updates delivered via approved media/workflow).

Question 6

What security controls are included?

Accepted Answer

Typical capabilities include:
• RBAC (role-based access control) for users/admins
• Audit logs for sensitive actions
• Local encrypted storage for documents, embeddings, and logs
• Secure service-to-service communications within the stack
• Optional integration with KMS/HSM depending on the deployment

Question 7

Does it support enterprise identity providers (SSO)?

Accepted Answer

Yes. AI-in-a-Box can integrate with enterprise identity systems (e.g., Microsoft Entra ID, Okta, Google Workspace) depending on the configuration.

Question 8

Is it SOC 2 / GDPR compliant?

Accepted Answer

AI-in-a-Box is built on the same platform foundation as Understand Tech. In on-prem deployments, the customer controls the infrastructure, which is important for compliance ownership. Visit our Trust Center for exact scope details.

Question 9

Which LLMs can run on AI-in-a-Box?

Accepted Answer

AI-in-a-Box supports local open-source models compatible with the included inference stack and the available GPU capacity. The exact list depends on:
• GPU memory / compute
• Model size and quantization
• Your approved model catalog and deployment policy

Question 10

Can users switch models?

Accepted Answer

Yes, within your allowed model catalog. Many customers run:
• A "fast" model for day-to-day tasks
• A "stronger" model for complex reasoning
• Optional domain models for specific workflows

Question 11

Does it require internet access for LLM inference?

Accepted Answer

No. Inference can run entirely locally.

Question 12

Can we use cloud LLMs too?

Accepted Answer

If your organization allows it, AI-in-a-Box can be configured to enable/disable external LLM access at the admin level. In restricted environments, external access is typically disabled.

Question 13

Can we fine-tune models on-prem?

Accepted Answer

Possible, depending on your hardware, the chosen model, and your governance/security policy. Many customers start with RAG first (fastest path), then evaluate fine-tuning if needed.

Question 14

What's included in the platform?

Accepted Answer

AI-in-a-Box includes:
• Understand Studio, our no-code app builder, included in the box
• Assistants, multi-LLM portals, RAG, chat with documents and CSV
• Bundled applications for regulated industries
• Admin console (users, teams, permissions, audit logs)
• APIs for integration and automation
• Local storage for documents and vector indexes
• Monitoring of platform usage and appliance health (GPU, CPU, memory, network, I/O)

Question 15

Can we build our own applications on the box?

Accepted Answer

Yes. Teams build their own applications in Understand Studio without code, and they run on the same appliance, inside the same security perimeter. No new procurement cycle for each new use case.

Question 16

What can we monitor?

Accepted Answer

Two levels, from one place. Platform usage: prompts, applications, workflows and models per team, plus unique users over time. Appliance health: GPU, CPU, memory, network and I/O per container. All metrics stay on the appliance.

Question 17

Can we mix local models and external providers?

Accepted Answer

Yes. You choose your AI providers, including hybrid setups: run local models on the box, add your own provider keys (for example OpenAI or Claude), or combine both. Admins control what is allowed, and restricted environments simply keep external access disabled.

Question 18

Is it multi-user?

Accepted Answer

Yes. It's designed for enterprise multi-user access with admin governance, RBAC, and audit logs.

Question 19

Can it connect to our internal systems?

Accepted Answer

Yes. AI-in-a-Box can integrate with internal systems for identity (SSO), logging/monitoring, and approved data sources, depending on your connector policy and network architecture.

Question 20

Does it include a UI and APIs?

Accepted Answer

Yes. Users can access the web interface, and teams can integrate through APIs for automations, workflows, and embedding AI into internal apps.

Question 21

Which use cases are already running in production?

Accepted Answer

Live and active deployments include:
• Silicon test automation: agents draft chip test cases, run them on real hardware, and evaluate results under human supervision
• Accounting and legal (CerFrance, on NVIDIA GB300): document processing, figure extraction, drafting and knowledge search on the firm's own records, plus custom apps built in Understand Studio
• IP and design teams: analysis stays on the appliance, with no token costs, no per-seat metering, and IP never leaving the perimeter

Question 22

How long does installation take?

Accepted Answer

About 30 minutes. The journey has five steps: unbox, install, configure (AI providers, DNS, authentication and teams), launch, and monitor. An installation guide and a setup video are provided, and prerequisites are hardware provisioned, credentials available and network policy validated.

Question 23

How is the stack deployed?

Accepted Answer

Usually via a containerized stack (Docker Compose) with a reverse proxy, platform services, local LLM service, database, cache/queue, and worker services.

Question 24

How are updates delivered?

Accepted Answer

Updates are delivered over the air (OTA) by Understand Tech:
• Platform releases and model updates
• Security patches and integrity/health checks
• No manual rebuild on your side
For air-gapped or isolated sites, the same validated packages are delivered offline through your approved transfer process.

Question 25

What is included in Update Services?

Accepted Answer

Commonly includes:
• OTA platform upgrades and security patches
• LLM lifecycle management (approved versions)
• Health checks and platform monitoring guidance
• Support workflow for incident handling
(Exact SLAs and cadence are defined in your contract.)

Question 26

Who operates the infrastructure day-to-day?

Accepted Answer

In on-prem deployments, the customer typically operates the infrastructure (as the admin), and Understand Tech provides software + support. Some customers request a fully managed model depending on constraints.

Question 27

Which configurations are available?

Accepted Answer

Two configurations, both running the same software stack:
• Team (NVIDIA GB10): compact desktop appliance for a team, pilots and departmental deployments
• Organization (NVIDIA GB300 Blackwell Ultra): datacenter-class appliance for organization-wide rollouts, larger models and higher concurrency

Question 28

Is the GB300 available today?

Accepted Answer

Yes. We are among the first in Europe to ship AI-in-a-Box on the NVIDIA GB300 Blackwell Ultra platform, with live customer deployments running fully on-premise, including CerFrance.

Question 29

How do we choose between GB10 and GB300?

Accepted Answer

Sizing is based on:
• Concurrent users and requests
• Target model size and quantization
• RAG indexing volume and refresh frequency
• Availability and redundancy requirements
Teams typically start on GB10 and move to GB300 when concurrency or model size grows.

Question 30

Do you provide the hardware?

Accepted Answer

Depending on your offer:
• Yes, the appliance can be provided as part of AI-in-a-Box, or
• The customer purchases the NVIDIA system from approved vendors and Understand Tech delivers the software plus deployment support.

Question 31

What about NVIDIA warranties and terms?

Accepted Answer

NVIDIA hardware, drivers, and firmware remain subject to NVIDIA's applicable warranties and terms. Understand Tech provides the software stack and associated support scope (as defined in the commercial agreement).

Question 32

How much does AI-in-a-Box cost?

Accepted Answer

AI-in-a-Box is a fixed annual subscription per deployment, never per token, per prompt or per usage:
• Team (NVIDIA GB10): $25,000 per year
• Organization (NVIDIA GB300 Blackwell Ultra): from $150,000 per year
See the pricing page for the full breakdown.

Question 33

Is it cheaper than building our own stack?

Accepted Answer

In most cases, yes. A comparable in-house GB300 build typically runs around €515,000 in year one and €985,000 over three years once hardware, software and development, maintenance and internal engineering time are counted. AI-in-a-Box removes most of that, with savings up to 80%, and removes the delivery risk of assembling the stack yourself.

Question 34

What does the yearly subscription include?

Accepted Answer

The yearly subscription includes everything needed to run and operate the solution:
• Full Understand Tech platform license (apps, assistants, workflows, RAG, APIs)
• Maintenance: software updates and security patches
• Support: troubleshooting and deployment assistance (per the agreed support level)
• Ongoing platform improvements and compatibility updates

Question 35

Are there any usage limits (tokens, prompts, API calls)?

Accepted Answer

No token-based pricing. The subscription is fixed price per box, designed for predictable budgeting.

Question 36

Is support included or optional?

Accepted Answer

Support is included in the annual subscription (no separate mandatory support add-on).

Question 37

Do we need to buy additional licenses for new users?

Accepted Answer

Typically no. AI-in-a-Box is priced per deployment, so adding users does not trigger token costs.

Question 38

What drives the price if it's fixed?

Accepted Answer

Pricing depends on the configuration you choose, Team (NVIDIA GB10) or Organization (NVIDIA GB300 Blackwell Ultra), and on deployment requirements such as performance, redundancy and support level. Never on tokens.

Question 39

Can we run a pilot before buying?

Accepted Answer

Yes. Most customers start with:
• A Proof-of-Value pilot on a single use case, with a defined scope and success criteria
• A production rollout on Team (GB10) or Organization (GB300) once sizing and governance are validated

Option	Hardware	Software + dev	Maintenance	Human resources	3-year totalover 3 years
Build your ownHardware is a one-off cost. Software, maintenance and people are per year, multiplied by 3.3 to 12 months to deploy, no SLA.	Hardware k€ One-off	Software + dev k€ Per year	Maintenance k€ Per year	Human resources k€ Per year	€239k
AI-in-a-Box · NVIDIA GB10All-in subscription, live on day one.Signed air-gapped updates, vendor SLA.	Included	Included	Included	Included	€75,000€25,000 / yearSaves 69%

AI-in-a-Box

Everything your teams need ships inside the box

From delivery to production in 30 minutes

Unbox the appliance

Two generations: GB10 and GB300

GB10

GB300

Full hardware comparison

What's included

Hardware

Software & License

How it works

Resources & documentation

GitHub Repository

Get Started / User Guide

Product Leaflet

Datasheet

Trust Center

Installation & Setup Guide

How companies are using it today.

Silicon Test Automation

Accounting & Legal

Secure Internal Assistants

Critical Infrastructure

Code Generation

Patents & R&D

What does it really cost?

Pricing at a glance

Why AI-in-a-Box

Technical specifications

AI-in-a-Box FAQ

General

Security, Privacy, and Compliance

Models and LLM Capabilities

Platform, Features, and Integrations

Deployment, Operations, and Updates

Hardware and Sizing

Pricing and Subscription

Ready to deploy AI inside your perimeter?

Read more