Why go local? — privacy, cost, offline

Local (or on-device) AI means the model runs on your machine or your own server, instead of sending every prompt to a company’s default cloud. Open-weight models are weights you can download and run under a licence—but “open” does not mean “automatically safe”; it means inspectable rules, not zero risk.

This lesson stays high level. You will write your own comparison, not install anything yet.

Note — Vendors and versions move fast. Model names, free tiers, and install steps change. Bookmark official docs; read licences and pricing at the source before you rely on a product.

Do this — two lists (10 minutes)

List A — When local / open-weight feels attractive

Write two bullets in your own words (examples you can use or mix):

• Privacy: prompts and file paths stay off a public cloud if your setup truly runs locally (always verify—some “local” apps still phone home).
• Cost: after hardware is paid, no per-token bill for hobby use (electricity still exists).
• Offline: patchy Wi-Fi on a train or rural site—if the model is already loaded.

List B — When cloud is still fine

Write two bullets:

• You only need quick answers and a work-approved chat.
• Your laptop cannot run a useful model (too slow or too little RAM).
• Your org requires a specific cloud for security logging.

Do this — two personas (5 minutes)

Persona	Local-first?	One-line reason
Traveller, bad Wi-Fi, practice hobby site	Yes / No / Maybe	You write:
Employee, strict data rules, work laptop	Yes / No / Maybe	You write:

There is no “exam answer”—there is judgement.

Concept — one paragraph to remember

Fit-for-purpose beats ideology. Local tools are powerful for learning and for sensitive drafts when configured correctly. They are not mandatory to ship a static HTML page—and they add setup time. Your capstone can be built with any workflow your employer allows; this module teaches you when local is worth the effort.

Check

• You can explain “local” to a colleague in two sentences without saying “blockchain.”
• You named one risk of assuming “local = 100% private” without checking the app’s settings.

Key takeaways

• Privacy, cost, offline are the usual reasons people explore local—each has caveats.
• Cloud tools remain valid for many learners; this module adds options, not a new religion.

What’s next

Next up: Open-weight families — Qwen, DeepSeek, MiniMax-class (high level)