Skip to main content

Frequently Asked Questions

Platform and Training​

Do I need a data science background to use Hubosmart?
No. The training interface is designed for domain experts — people who know what they need to detect — not machine learning engineers. If you can identify what you want the model to recognize and provide example images, you can train a model.

What types of AI tasks does Hubosmart support?
The platform currently supports image classification tasks: assigning an image (or a region of an image) to one of several categories you define. This covers the majority of visual inspection, monitoring, and sensing use cases in industrial and commercial deployments.

How many images do I need to train a model?
A functional initial model typically requires 20–50 images per class. More images generally improve accuracy, particularly for classes with high visual variation. Models can be improved by adding new training data and retraining at any time.

Can I use my own images, or do I need to collect them specifically through the platform?
You can upload images from any source. For best results, images should be representative of the actual conditions where the model will run — same lighting, camera angle, and environment.

How long does training take?
Training jobs are processed on Hubosmart's cloud infrastructure and typically complete in 2–5 minutes for standard classification tasks. Retraining an existing model with new data takes the same amount of time.

Deployment​

Does the deployed model require internet connectivity to run?
No. Once deployed, inference runs entirely on the edge device. No internet connection is required for detection. Only Agent Workflow actions that call external services (sending alerts, API calls) require connectivity — and those can be queued for the next connectivity window.

What hardware does Hubosmart run on?
Hubosmart targets affordable microcontroller-class hardware suitable for battery-powered and field-deployed applications. See the hardware overview for the list of supported devices and their capabilities.

Can multiple models be deployed to the same device?
Currently, each device runs one active model. Model updates replace the previous model on the device.

What is the difference between the three deployment paths?

  • Hot-Swap: Fast model swap for testing. No access control, single device, no compilation.
  • Personal SDK: Encrypted model packaged with a credit-based access key. For distributing AI capability to individual customers.
  • Enterprise Batch: Custom SDK with batch deployment tooling. For B2B scenarios, OEM products, and multi-site fleet rollouts.

See Deployment Paths for detailed comparison.

Access Control and Credits​

How does the credit system work?
SDK keys are issued on a credit basis. Each key you issue for a Personal SDK or Enterprise Batch deployment costs credits from your account. Credits give you predictable distribution cost and the ability to control and revoke access when needed.

Can I revoke access to a deployed SDK?
Yes. SDK keys can be revoked at any time through the model management interface. Revoking a key disables the associated deployment without affecting other keys or customers.

Is the trained model protected against extraction?
Personal SDK and Enterprise Batch deployments use model encryption. Customers receive a functional SDK — they cannot inspect or extract the underlying model weights.

Agent Workflow​

Do I need to write code to set up Agent Workflow?
No. Agent Workflow is configured through the platform interface. You define the conditions and the responses in plain terms — no firmware development or integration code required.

What systems can Agent Workflow connect to?
The OPENCLAW execution layer can connect to any system that accepts standard HTTP webhooks or API calls. This covers most messaging platforms, dashboards, ERP systems, and custom APIs.

Does Agent Workflow run on the device or in the cloud?
The inference and decision logic run at the edge. Only the final action output — a webhook call, a database write — leaves the local environment.

Getting Help​

If your question is not answered here, contact support through your account dashboard.