THE STEALTHBOX SYSTEM

One runtime. Four deployment classes.

StealthBox is not a single product on a shelf. It is one governed local AI runtime — compute, models, knowledge, agents, applications and integrations — delivered in four deployment classes. The hardware scales. The runtime scales. The control plane stays the same.

The family

The class tells you the shape. The architecture tells you what fits inside.

Each StealthBox class is a deployment shape — not a fixed retail SKU. Systems are sized against your real workloads, concurrency, models, context, data volume, latency, availability, security requirements, integrations and expansion plan.

View StealthBox Edge
View StealthBox Core
View StealthBox Enterprise
View StealthBox Sovereign
StealthBox Edge

Compact local AI for a site, facility or field environment.

Intended environment
A single site, facility, field operation or remote office where data must not leave the location.
Workload class
Focused workloads — assistant Q&A, document review, local knowledge retrieval, bounded automation.
Deployment model
Single-node local runtime, on-device inference, optional controlled connectivity.
Relative scale
One team or one site. Low concurrency, tight latency budget.
Local / private AI role
Keep inference and knowledge at the point of work — no round-trip to a remote cloud.
Knowledge & agent use
Local document corpus, assistant interfaces, bounded agents for routine site tasks.
Integration pattern
Lightweight local integrations; optional controlled sync to a central environment.
Security & control level
Local identity, local permissions, local audit. Data stays at the site.
Deployment & support path

Discover → Design → Size → Validate → Deploy → Operate, with on-site support.

StealthBox AI OS

One control plane across the whole family.

StealthBox AI OS is the common software foundation — model serving, knowledge, agents, permissions, audit and monitoring — that runs across every StealthBox class. The hardware scales. The runtime scales. The control plane stays the same.

Model serving & registry
Knowledge & retrieval
Agent orchestration
Identity & permissions
Audit & monitoring
Human approval gates

Same control plane, different footprint — from a single edge node to a sovereign data-centre deployment.

Sizing

Systems are sized against the workload — not picked from a shelf.

A StealthBox environment is sized against your real workloads, concurrency, model and context requirements, data volume, latency target, availability, security requirements, integrations and expansion plan. The class tells you the deployment shape. The architecture tells you what fits inside it.

Workloads & concurrency
Models & context requirements
Data volume & residency
Latency target
Availability
Security requirements
Integrations
Expansion plan
Comparison

Compare the deployment classes.

A class is a shape, not a final architecture. Use this to understand which class fits your environment — then we size the system to the real workload.

DimensionStealthBox EdgeStealthBox CoreStealthBox EnterpriseStealthBox Sovereign
Intended environmentA single site, facility, field operation or remote office where data must not leave the location.A department, programme or mid-size organisation that needs one governed AI environment.A large organisation running AI across many functions, systems and locations.National, regulated or classified environments where no external dependency is acceptable.
Workload classFocused workloads — assistant Q&A, document review, local knowledge retrieval, bounded automation.Shared knowledge, governed agents, copilots and workflow automation across teams.Mixed workloads — knowledge, agents, copilots, custom apps and operational intelligence at scale.Sovereign workloads — national data, regulated processes, classified or high-control operations.
Deployment modelSingle-node local runtime, on-device inference, optional controlled connectivity.Clustered local runtime, shared knowledge base, agent orchestration, identity-scoped access.Multi-cluster runtime, federated knowledge, role-based access, observability and lifecycle.Air-gapped where technically supported, sovereign model and weight management, local identity, no external telemetry.
Relative scaleOne team or one site. Low concurrency, tight latency budget.One organisation, many users. Moderate concurrency, shared context.Many teams, many sites. High concurrency, federated context, shared governance.Sized to the mandate — site, programme or national footprint. Isolation over scale.
Local / private AI roleKeep inference and knowledge at the point of work — no round-trip to a remote cloud.A single governed runtime for the whole organisation — models, knowledge and agents under one policy.One governed estate, many workloads — shared control plane, distributed inference.No external dependency. Sovereignty by architecture — data, models, identity and audit all inside.
Knowledge & agent useLocal document corpus, assistant interfaces, bounded agents for routine site tasks.Enterprise RAG, role-aware assistants, bounded agentic workflows, document intelligence.Federated knowledge, cross-function agents, secure copilots, custom applications and integrations.Sovereign knowledge, controlled agents, classified-document intelligence, human-gated decisions.
Integration patternLightweight local integrations; optional controlled sync to a central environment.Identity federation, API and event integration with core enterprise systems.Enterprise integration layer — APIs, events, data connectors and legacy system bridges.Isolated or controlled-boundary integration only; no external API calls on sovereign data.
Security & control levelLocal identity, local permissions, local audit. Data stays at the site.RBAC, audit trails, human approval gates, policy enforcement, data residency.Enterprise RBAC, federated identity, centralised audit, policy governance, monitoring.Maximum control — local identity, sovereign models, full audit, human approvals, no external telemetry.
Deployment & support pathDiscover → Design → Size → Validate → Deploy → Operate, with on-site support.Discover → Design → Size → Validate → Build → Integrate → Assure → Deploy → Operate.Full method with staged rollout, observability and lifecycle support across sites.Full method with assurance-first staging, air-gap validation where supported, and sovereign lifecycle.
Bring us the workload

Tell us what must stay local.

Bring the workload, the data constraints and the operating requirements. We start at discovery and size the right class to the real mission.

Request Architecture Session