Local where it matters
Updates
Ollama, LM Studio and Jan release updates every few weeks: speed improvements, support for new models, security fixes. Keep the app updated from inside the app or the official site. If you run a local server that other programs use, read the release notes before updating, as defaults sometimes change.
Disk space and tidying
Each model is several gigabytes, and each app keeps its own copy. Every few months, list what you've downloaded and delete what you don't use. Keep one good everyday model, maybe one small fast one and one larger one for harder jobs.
New models
Better open models appear every few weeks. You don't need to chase them. Every quarter, re-run your three test tasks on one promising new model; switch only if it's clearly better on your work. Check the licence again if you use it for business.
Your hybrid plan
Write it down: private documents, client data and sensitive questions stay local; current events, hard reasoning and large creative jobs go to a cloud tool your workplace approves, with nothing confidential in them. Many people find this split gives them most of the benefit of both.
Sort your typical AI tasks into two lists: ones involving private material, and ones that don't.