100% on premise
BrainBox is installed in the customer's data center and runs without third-party endpoints.
IMBox AI
Sovereign AI for defense, security and public administration.
IMBox AI brings models, orchestration and applications inside the customer's perimeter. No cloud dependency, no data leaving the organization and a native experience inside the secure IMBox platform.
400k+
active deployments
200k+
public employees using AI
300+
public bodies
8 wk
from zero to production
Sovereignty by design
Your AI runs where your data lives.
Inference hardware, models, orchestration and applications are deployed inside your data center. Sensitive information never leaves the perimeter.
Zero exfiltration
No telemetry, no anonymized samples and no external training on customer information.
Air-gapped environment
Editors, viewers, players and sandboxing live inside IMBox to remove external dependencies.
Institutional scale
Built for public administration, security and defense, with traceability and operational control from day one.
BrainBox
BrainBox brings AI into your own data center.
BrainBox is the inference node for IMBox AI: an autonomous compute unit for running quantized open source models inside your infrastructure, with full control over data, performance, and scaling.
BrainBox cluster
Linear capacity, standard rack deployment
4x
NVIDIA RTX 5090
128 GB
VRAM agregada
1.500
TFLOPS FP16
32C
AMD Genoa CPU
192 GB
RAM de sistema
N+1
escalado modular
GPU BOUNCER
The software that turns your hardware into a production AI system.
Without it, it's just raw hardware. With GPUBouncer, everything runs from day one.
ModelsAPIOrchestrationScalingSecurityHigh availability
The platform integrates everything needed to consume AI in production from day one without taking on the technical complexity of maintenance and scaling: optimized models, API, orchestration, security, monitoring and high availability.
GPUBouncer is the layer that makes this infrastructure work as a single service. It routes each request to the right model, balances load across available instances, preserves sessions to improve latency, monitors model health and applies retries or fallback transparently. When new compute capacity is added to the cluster, the platform brings it into the service so it can keep growing without redesigning integrations.
API
/v1/chat/completionsActive
/v1/embeddingsActive
Models
T2TL
I2TL
A2T
OCR
Orchestration
Active models3/5
Running inferences24
Load balancingAutomatic
GPUBouncer
Capacity
68%
Performance
2.4k t/s
Scaling
Clusters online10/10
Available capacity75%
Automatic scalingActive
Security
Access levelsEnabled
DataOn-premise
AuditTraceable
High availability
Active instances10/10
Health checksOperational
IncidentsRecovered
One API
An OpenAI-compatible entry point for connecting current or new applications without redesigning integrations.
Plug and play
Add new BrainBox units and capacity grows modularly, with automatic resource detection and enrollment.
High availability
GPUBouncer monitors model health, isolates failing instances and redirects requests transparently.
Sovereignty and security
Inference runs on the customer's on-premise infrastructure, keeping data, traceability and control inside the corporate environment.
Model Foundry
Open-source models, quantized and tuned in house.
The IMBox engineering team maintains a curated catalogue of model families, evaluated against institutional workloads and deployed through a controlled operational cycle.
CodePurposeGPUsVRAM
T2TLText to text, large reasoning4128 GB
I2TLImage to text, large264 GB
A2TAudio to text132 GB
OCROptical character recognition132 GB
EMBEmbeddings132 GB
T2TFText to text, fast132 GB
I2TFImage to text, fast132 GB
Applications consume stable capabilities such as T2TL, I2TL, A2T or OCR. IMBox keeps the best available models running underneath.
No provider lock-in
Integrations target a capability, not a specific version that can become obsolete.
Continuous SOTA updates
The catalogue evolves with new stable open-source models without redesigning existing applications.
MLOps handled by IMBox
Research, quantization, testing, deployment and the operational cycle are managed inside the platform.
Multiple AI capabilities. One platform your teams already use.
Users interact with AI in natural language inside IMBox: no tool switching, no data leaving the perimeter, no unnecessary complexity.
On-premise agentic workflows
Agents with 30+ specialized tools to chain actions, cross data sources and complete complex tasks.
tools and traceability
Governed assistants
Create generalist or specialist assistants, feed them internal documents and publish them by user, group or organization.
RAG and evaluation
Native recorder
Record meetings, interviews, statements or debriefs and send them into the transcription and analysis flow.
custody and context
Transcription with diarization
Turn audio and video into text, identify speakers and generate summaries, tasks and review-ready minutes.
local speech to text
Translation and rewriting
Translate documents, summarize at different lengths and rewrite with configurable tone and style.
languages and style
Document generation
Create and edit Word, Excel and PowerPoint in conversational mode using IMBox Drive context.
secure office work
Document OCR
Read scanned files, images and annexes to turn them into searchable and reusable content.
scanned documents
Knowledge graph
Connect fragmented data and internal relationships so teams can query operational knowledge naturally.
contextual search
Database queries
Allow AI to query corporate data under permissions, auditing and limits defined by the organization.
controlled data
Sovereign API
Expose AI capabilities through a compatible interface for integrations and custom clients.
fast integration
Assistant memory
Assistants remember preferences, context and relevant patterns within security boundaries.
operational continuity
Administration panel
Central control over models, quotas, users, audit, BrainBox fleet and service health.
unified operations
Deployment
From zero to production AI in eight weeks.
A predictable process to install BrainBox, activate GPUBouncer, validate models and publish capabilities to teams.
1
Week 1
Architecture, security and integration requirements.
2
Week 2
Rack, networking and initial BrainBox commissioning.
3
Week 4
Models, assistants, auditing and real-workload tests.
4
Week 8
Production launch, training and operational follow-up.
IMBox AI
Bring sovereign AI inside your perimeter.
A private intelligence layer for organizations that cannot afford to lose control over their data.
Talk to IMBox