InferTrust™ integrates through a lightweight SDK that wraps your existing inference calls. Minimal code changes are required. The SDK intercepts model output at
InferTrust™ is designed to integrate with existing AI pipelines with minimal disruption. The core integration point is the InferTrust™ SDK, which wraps your model's inference call. Instead of calling your model directly, your application calls the InferTrust™ SDK, which in turn calls your model, captures the output at the inference boundary, signs the decision record, and returns the result to your application.
The integration requires wrapping your inference call with the InferTrust™ SDK. Your model code, training pipeline, and data processing logic remain unchanged. The SDK operates as a thin layer between your application and your model's output, adding cryptographic signing without altering the inference itself.
A typical integration involves three steps:
The SDK is framework-agnostic. It works with TensorFlow, PyTorch, ONNX Runtime, custom inference engines, and third-party model APIs. The SDK does not need to understand your model's internals. It operates on the output, capturing the result and associated metadata at the moment inference completes.
InferTrust™ supports cloud deployments (containerized services, serverless functions), edge deployments (on-device inference on mobile, embedded, or IoT hardware), and hybrid architectures where inference runs on-device but results sync to a central server. The signing always happens locally, at the point of inference, regardless of where the model runs.