Fully offline operation

Neural Cube runs entirely on-device. Every model, tool, and software lives on the robot, with no required round-trip to a cloud service, vendor API, or remote inference endpoint. It can be disconnected from the internet and continue to work.

What this means for the robot

Privacy

Cameras, microphones, and conversation history can never leave the device, so that whatever the robot sees and hears stays on the robot. There is no telemetry pipeline, third-party transcription service, or vendor sitting between the user and their robot, which makes it suitable for homes, clinics, classrooms, and other settings where sending raw sensor data to an outside server is not acceptable.

Reliability

The robot should not stop working when the Wi-Fi drops and is not dependent on upstream API rate-limits, vendor changes to a model's behaviour, or subscription lapses. In this case, latency is bounded by local compute and not by network conditions. This makes it usable in environments where cloud-dependent robots cannot operate such as internet-denied and off-the-grid environments. The robot will work on a home network, on a hotspot, or with no network at all.

Security

There are no API keys to leak, no cloud accounts to compromise, and no inbound endpoints exposed by default. Additionally, no vendor-side breach can expose conversation history or sensor data, because no data is ever uploaded to any server. By removing the network dependency, we are trying to ensure that data stays on the robot as much as possible unless otherwise desired by the user.

*Online features (such as web search) are optional and are simply disabled when internet is not accessible.