Skip to main content

Deployment Paths

A model that works in training is only useful if it reaches the hardware where it needs to run. Hubosmart offers three deployment paths, each designed for a specific distribution scenario.

Hot-Swap Deployment

Best for: Prototyping, internal validation, single-device testing

Hot-swap deployment gets a trained model onto hardware in minutes without any compilation step. The model is packaged into a deployable format and loaded onto the device via USB or local network. No firmware build required.

Use hot-swap when you need to:

  • Test a newly trained model on real hardware before wider deployment
  • Validate detection accuracy in actual operating conditions
  • Iterate rapidly during the development phase

Hot-swap models can be replaced by any new model at any time. There is no credential or access control — this path is appropriate for internal use, not for distributing to end customers.


Personal SDK

Best for: Independent developers building AI-powered products for individual customers

The Personal SDK path packages a trained model into an encrypted, key-protected format for distribution to individual end users. Each SDK contains the model you trained — your customer cannot inspect or extract the model weights.

How access control works:
SDK keys operate on a credit basis. When you issue a key to a customer, it consumes credits from your account. Keys can be revoked if a customer relationship ends. This gives you predictable distribution cost and control over who can use your model.

What customers receive:
A self-contained SDK that activates the AI capability on their hardware. They do not need to understand model training or deployment — they install the SDK and the capability works.


Enterprise Batch Deployment

Best for: B2B scenarios, large device fleets, OEM hardware products, multi-site rollouts

Enterprise Batch creates a custom SDK for a specific customer or deployment context, then distributes it to an entire device fleet in a single operation using the batch-flash tooling.

Custom SDK per customer:
Each enterprise customer receives a SDK built specifically for their deployment. The SDK contains their trained model — not a shared model — and is cryptographically bound to their authorization context.

Batch flash operation:
The batch-flash tool accepts a device list and SDK package and handles the entire fleet update: connecting to devices, flashing firmware, verifying deployment, and logging results. Updating 10 or 1,000 devices requires the same number of operations.

Typical use cases:

  • Manufacturer deploying visual inspection across multiple production lines
  • Retail chain rolling out shelf monitoring to every store
  • OEM shipping devices with custom AI built in

Path Comparison

Hot-SwapPersonal SDKEnterprise Batch
Compilation requiredNoNoCustom build
Access controlNoneCredit-key per devicePer-customer authorization
Distribution scaleSingle deviceIndividual customersFleet / B2B
Model protectionNoneEncryptedEncrypted + custom
Typical useDev / testingIndependent developerEnterprise / OEM