AI that runs on the phone, laptop or other device itself instead of on a company's servers.
It works without a connection and keeps data on the device, but it is limited by the device's chip and memory.
Because local models keep prompts and files on the machine, the tool gives Mac users a privacy-focused option alongside optional cloud models billed by provider.
The engine behind LiteRT now runs on Windows and Linux desktops as well as macOS, and Google is retiring the legacy TFLite GPU delegate.
The biggest change is that agents can now run entirely on a personal machine, keeping user data on the device.
The update keeps AI workflows on-device for eligible Macs, giving subscribers more control over private data and model spend.
Tom’s Hardware says the Windows on Arm laptop is the fastest system it has tested for local agentic AI workloads.
An opt-in protection mode will scan unknown senders and links, while Google extends scam detection beyond Pixel to more Android devices.
Meta says the system will let glasses handle more advanced AI features in the cloud without direct access to sensitive data.
The update keeps more code on-device while shifting only a small part of agent work to the cloud.
The Stockholm startup is looking to expand from hundreds of units to several thousand as it sells AI systems that keep animals away from infrastructure.
Foresight keeps meeting audio, notes and files on the computer, giving Mac users an offline option for AI note capture and follow-up questions.
Microsoft’s new AI-focused PCs and Windows 11 features point more work toward local, on-device use.
The system is designed to let health workers gather key pregnancy information without a sonographer or reliable power and internet.
The model is built for local search and retrieval, letting more data stay on device.
SharpEmu can now run real game code, while Relapse widens the range of PS5 units that can be jailbroken.
The new lower-capacity model ships Oct. 23 and gives teams a cheaper entry point while still allowing two units to be clustered for larger models.
Demo shows compressed, open‑weight models handling real‑time visual queries fully on-device, but no products are announced yet.