“Local AI” usually means that some or all of the model processing happens on a computer or device you control instead of being sent to a hosted AI service. That can be an important distinction, but the label by itself does not explain what every part of a particular setup does.
An application may have more than one mode, connect to online services, or rely on other software around the model. The computer, network, account, and settings matter too. If privacy is part of the reason for choosing a local setup, check the documentation for the exact program and features you plan to use.
What I would check
- Which model is being used, and where does it run?
- Does the application connect to online services for any features?
- What computer resources does the chosen model need, and is performance acceptable on your equipment?
- How are the program and models updated, and where are prompts or files stored?
- Can you use the setup comfortably, or does maintaining it become more work than it saves?
As one product-specific example, Ollama’s privacy policy describes its local and cloud-hosted model options separately. That is the provider’s description of its own service, not a rule for every local-AI program. Check the current documentation for whichever software you choose.
This page began as notes for a possible Smart Tech for Real People topic. It is not a book announcement, and I may never publish a book about it. See the books currently available for confirmed titles.