Sovereignty & compliance
Regulators and boards need to know where data lives. With the Box, it lives on hardware you own, in a room you control.
Flagship product · EntrepriseAIBox
Your private
AI workforce.
A private AI appliance, custom-built from a detailed assessment of your AI workload. It runs open models and ready-made agents inside your walls, without sending sensitive data anywhere.
24 TOPSOn-device NPU for local inference (edge model).
6 agentsReady on first boot. Custom agents easily added.
CustomSized to your workload by our automated compute design tool.
0 bytesOf sensitive data leave your network unless you allow it.
Why a box
Cloud AI wasn’t built
for your constraints.
Connectivity & cost
Local inference keeps working when bandwidth doesn’t, and turns per-token cloud bills into a predictable appliance.
Skills & time-to-value
No MLOps team required. Agents, models, dashboard and updates arrive pre-integrated and managed.
Built-in agents
Six agents on day one.
Custom agents, easily added.
Request → router → agent → audited answer00:12
Triage, draft and summarise threads.
Calendar
Schedule, prepare and follow up meetings.
Documents
Search, compare and extract from files.
Support
Answer staff and customer questions.
IT
Resolve tickets and guide fixes.
Reports
Turn data into briefs and dashboards.
Hybrid by design
A smart router decides
where every request runs.
Sensitive or routine requests stay on local NPU or GPU compute. Complex, non-sensitive ones can use a cloud model, but only if your policy allows it.
InputUser or system requestChat, voice, email, document or API call, with user identity and role.
Smart routerPolicy + classifierScores sensitivity and complexity, checks RBAC and data-residency rules, then picks a model.
Local · defaultOn-box NPU / GPUOpen models, e.g. Llama · Qwen · DeepSeek · Kimi
Cloud · optionalApproved providerOnly for non-sensitive, complex tasks. Can be fully disabled.
Security & governance
Enterprise controls,
not afterthoughts.
RBAC
Roles decide which agents, data and models each person can use.
SSO
Plugs into your identity provider. No new passwords to manage.
Encryption
Data encrypted at rest and in transit, keys under your control.
Audit logging
Every prompt, route, tool call and answer is logged and exportable.
Technical overview
What’s inside.
| Compute (edge) | Dedicated NPU, 24 TOPS · 8 GB memory · compact, low-power appliance |
|---|---|
| Compute (enterprise) | GPU servers and multi-GPU clusters for larger models and high-volume workloads |
| Local models | Any open model sized to your hardware, e.g. Llama, Qwen, DeepSeek, Kimi, Whisper (speech) · swappable via managed updates |
| Routing | Policy-driven smart router: local by default, optional cloud fallback |
| Agents | Email, calendar, documents, support, IT, reports · custom agents easily added · industry agent packs |
| Software stack | NPU & GPU runtimes + Docker Compose · OpenClaw agentic framework · React admin dashboard & REST API · hardened Debian 12 base |
| Voice & vision | 4-microphone array · HDMI 4K input |
| Security | RBAC · SSO · encryption at rest & in transit · audit logging |
| Design & setup | Workload assessment · automated compute design · first-boot wizard for network, SSO and roles |
| Operations | Managed updates · works with cloud fallback disabled (air-gapped) |
Form factors
From a branch office
to a national data centre.
EntrepriseAIBox Edge
A compact appliance for teams, branches and field sites. Local NPU inference with the full agent suite.
- 24 TOPS NPU · 8 GB
- Voice & vision inputs
- Ideal for teams, branches and field sites
EntrepriseAIBox Server
GPU servers and clusters for larger models, heavy document and geospatial workloads, and organisation-wide rollout.
- Multi-GPU, scale-out
- Runs Paper-to-Digital, Unium and Map2See at volume
- Same dashboard, policies and agents
The platform
Everything we build
runs on the Box.
Start with the agents. Add products as your needs grow. Each one inherits the same security, routing and operations.
Paper-to-Digital
Industrial OCR and structured extraction.
Unium
Unified, queryable data fabric.
Map2See
Agentic geointelligence from imagery.
YourVoice · Intello
Local-language voice AI and predictive models.
Custom by design
Designed for your workload.
Not just plug-and-play.
STEP 01
Assess
A detailed assessment of your AI workload: use cases, users, data volumes, languages, latency, connectivity and budget.
STEP 02
Design
Our automated compute design tool sizes the Box: NPU or GPU, memory, storage, models and agents.
STEP 03
Build & install
We build the Box to that design, install it on your network and connect SSO, roles, mail and document stores.
ONGOING
Operate & extend
Managed updates, re-sizing and new agent packs or products as your needs grow.
FAQ
Questions we hear often.
Does any of our data leave our premises?
Not by default. All inference runs on your local NPU or GPU hardware. Cloud fallback is optional, policy-controlled and can be disabled entirely for air-gapped operation.
Which languages are supported?
The bundled models handle English and French well out of the box. Through our YourVoice work we add African-language speech and text capabilities as agent packs.
Can it connect to our existing systems?
Yes. The Box exposes a REST API and ships connectors for mail, calendars, file shares and identity providers. Custom connectors are built as part of your deployment.
What if we need bigger models?
Move to the Server / Cluster form factor, or allow the router to use an approved cloud model for non-sensitive tasks. Your agents and policies stay the same.
Is it plug-and-play?
Setup on site is simple, but no two Boxes are the same. Each one is designed from a detailed assessment of your AI workload, sized with our automated compute design tool and configured with the models, agents and connectors you need.
How is it priced?
Each Box is priced from its design: hardware sized to your workload, a software licence and optional agent packs. Contact us for a quote.
Book a demo
See EntrepriseAIBox
on your own data.
We’ll run a live demo, assess your AI workload and send a Box design sized to your needs.
PlatformExplore the full stack →