PRIVACY BOUNDARY

Can local AI really be private?

Yes, local model inference can happen entirely on hardware you control. But “the model is local” does not automatically mean every part of the app or workflow stays local.

Start with the model runtime

If a runtime loads a downloaded model and processes prompts on your machine, that inference can stay local. Check its networking and telemetry settings.

Then check the interface

Open WebUI and similar interfaces can connect to both local and remote providers.

Tools can cross the boundary

Web search, remote MCP servers, hosted embeddings, cloud speech recognition and third-party APIs can send selected data away from the machine.

Local files need their own security

Disk encryption, user accounts, backups, browser access and network exposure still matter.

Offline is the clearest test

If the workflow continues to work after disconnecting the network, that is strong evidence the required processing is local.

Compare the wider trade-offs in local vs hosted AI.