What Is Brain-Box? Local AI Processing for Drones, Docks, Cameras and Robots

Why a dedicated local AI device is useful

Video analytics combines sustained data ingest, model inference, metadata, storage and operator access. A general laptop can demonstrate parts of that workflow, but a field deployment needs predictable interfaces, mounting, power, software lifecycle and recovery behavior. Brain-Box packages the compute layer specifically for Spectro AI applications.

Keeping the device near a controller, dock, robot or camera network shortens the data path and gives the organization direct control over where footage and models reside. It also allows core local functions to continue when external connectivity is unavailable, subject to the design of the complete system.

• Run Spectro AI detection models on live RGB and thermal feeds.
• Store video, images, detections and structured metadata locally.
• Support SAI-HUB mission, video and robot workflows.
• Connect selected third-party sensors or systems through APIs.
• Send alerts through configured local or connected channels.

Brain-Box M and Brain-Box D serve different installations

Brain-Box M is the flexible format for portable field use, controller-connected drone operations, mobile robots and video systems. Brain-Box D is designed for fixed DJI Dock deployments and a weather-protected installation approach. The software layer and algorithms are selected around the job.

The letter is not a performance score. It indicates the intended deployment. A needs assessment should consider the SAI-HUB configuration, model workload, number and resolution of streams, storage period, network interfaces, enclosure and site power.

What “on-premises” means in practice

On-premises processing means inference and primary storage occur on infrastructure controlled at the operating site or within the organization’s environment. It does not mean a device can never connect outward. Remote viewing, SMS, updates or integrations can be added through explicit network paths while the core media pipeline remains local.

A security review should still cover user accounts, encryption, ports, update provenance, backups, physical access and retention. Local hardware changes the control boundary; it does not replace governance. Ask for a data-flow diagram and test traffic during normal, offline and recovery states.

Algorithms, streams and operational outputs

Every Brain-Box deployment includes a set of Spectro AI algorithms selected for local RGB and thermal detection. The catalogue covers common targets such as people, vehicles, animals, fire and smoke, safety gear, foreign object debris and solar panel hotspots. Custom-trained models can address a specific asset or scene after representative validation.

With SAI-HUB VID, Brain-Box can process connected RTSP or ONVIF video streams and store structured detection metadata. With SAI-HUB RC or SAI-HUB DD, the same local layer supports drone mission evidence and live detections.

• Detection class, confidence and bounding-box coordinates.
• Timestamp, project and source stream.
• Location metadata where the source and workflow support it.
• Original or annotated clips for verification.
• Exports or API events for surrounding systems.

How to specify and validate a Brain-Box deployment

Begin with workload rather than a generic hardware specification. List each camera or drone feed, resolution, frame rate, RGB or thermal format, target models and acceptable alert latency. Define how much original and derived data must be retained and who needs access.

Then conduct a representative test. Run the intended model and streams, interrupt internet access, fill a realistic amount of storage, export detections and perform an update or rollback. Document the limits and monitoring that operations teams will use after handover.

1. Select SAI-HUB RC, DD, VID or ROB based on the operating environment.
2. Measure the real inference workload and end-to-end latency.
3. Define local retention, backup and deletion policies.
4. Review network, identity and physical-security controls.
5. Validate alerts, exports and failure recovery with the responsible users.

Frequently asked questions