DEFINITION
On-Device AI
Running model inference or pre-processing entirely on the user's machine instead of sending data to a cloud service.
In depth
On-device AI keeps the model — or, in privacy-layer use, the detection and transformation logic — on the user's hardware. No prompt, embedding, or intermediate result is transmitted to a remote server. On-device AI sidesteps cloud-provider data-retention policies, regulatory cross-border transfer issues, and the entire class of breaches that target central servers. It is constrained by what fits in local memory and how fast local CPUs and GPUs can run, but for pattern detection and lightweight transforms the constraint is generally not binding.
Examples
- Local large language models running through Ollama or LM Studio.
- Apple Intelligence on-device features for summarisation and rewriting.
- Browser-side detection of sensitive patterns before a prompt is sent to a remote model.
How Locke handles on-device ai
Locke is fully on-device. Detection patterns are bundled with Locke and execute against the text in your composer using local APIs. There is no remote inference, no telemetry, no fallback path that sends data off-machine.
See the productRelated terms
- Zero-Knowledge Processing
An architecture where the service provider has no technical ability to read user data, because all sensitive processing happens on the client.
- Privacy Layer for AI
A control point between a user and an AI service that detects, transforms, or blocks sensitive data before it reaches the model.
- Contextual Matching
Detecting sensitive data by combining structural patterns with surrounding context, instead of relying on regex alone.
