Glossary

Client-side inference

3 min read
In short

Client-side inference runs a machine learning model in the user's own browser or device rather than on a server, so raw input never leaves the device. Only derived values are transmitted.

What it is

In a browser this typically means TensorFlow.js or ONNX Runtime Web executing a model against webcam frames or event data locally. Frames are processed in a canvas buffer and discarded; only numerical scores are sent.

The trade-off is model size and device performance — client-side models are usually smaller and less accurate than server-side equivalents.

Why it matters

It substantially narrows the data protection footprint. If images are never transmitted or stored, most of the GDPR analysis around biometric data becomes simpler.

A common confusion

Client-side processing does not affect the AI Act prohibition. Article 5(1)(f) concerns the inference, not where it is computed — a prohibited practice is prohibited on-device. This is one of the most common misunderstandings in the category.

See also biometric data, emotion inference and air-gapped deployment. See the full glossary for the rest, or our security and data handling page for the wider context.

See where your content loses people

Book a walkthrough and we will show you the engagement data on your own content.

Request a demo
Get Started

See what completion rates can't tell you

Find out exactly where your content works, where it fails, and what disengagement looks like before people leave.

Request a Demo