Local AI runs a model on a computer or server you control. With the model and its dependencies installed, some workflows can run without an internet connection.
Where it can be useful
- Working with documents on a controlled local system
- Coding and drafting with a locally hosted model
- Experiments that need configurable models and predictable hardware access
Local does not automatically mean private
Privacy depends on the entire setup. Applications, plugins and integrations may still send data elsewhere. Review network access, telemetry, permissions and storage, and test the system offline if that is a requirement.
Choose hardware for the workload
Model size, memory, context length and the number of users affect speed and quality. Hardware, electricity, maintenance and licences are part of the cost; running locally is not always cheaper than a hosted service.
Build a useful pilot first
Start with a defined task and representative data. Compare output quality and response times, then decide whether to keep the workflow local, hosted or a combination of both.
