Notes

Local AI or cloud AI? Start with what must stay private

“Should we use local AI or cloud AI?” sounds like a technology question. For a small business it is mostly an information question: what does this job need to see, and where is that information allowed to go?

Answer that first, and the choice of where a job runs usually gets much easier.

Sort the information before you pick a tool

Take one job you want help with and list what the assistant would need to read to do it. Then put each item in one of three piles:

  • Fine to share. Things that are already public or harmless: your hours, your menu or service list, your FAQs, a draft post.
  • Share with care. Ordinary business details that should not be spread around, but that you already trust to online services: customer names, bookings, schedules, internal notes.
  • Must stay in house. Passwords and account logins, payment details, health, legal, or financial records about people, anything a client asked you to keep confidential, and anything a contract, license, or rule you work under says must not leave your control.

Most everyday jobs only touch the first two piles. Those are good candidates for a cloud assistant. The third pile is where the question gets serious.

What cloud AI is good at

A cloud assistant runs on the provider’s computers. That usually means the most capable models, nothing for you to install or maintain, regular improvements, and access from your phone or any computer.

The trade is that what you share travels to someone else’s system and is handled under their terms and settings. That is fine for a lot of work, but you should know what those terms say before you share anything from the second pile, and you should not share the third pile by default.

What local AI is good at

A local setup runs on a computer you control, in your office or home. The information it reads can stay on that machine, and some setups keep working without an internet connection.

The trade is responsibility. You need suitable hardware, someone has to set it up and keep it updated, and backups and security are on you. Models that run on a single machine are often less capable than the largest cloud models, so it pays to test whether one is good enough for the specific job.

Questions to ask before deciding

  • Does this job actually need anything from the “must stay in house” pile, or can it be done without it?
  • Can you remove names, account numbers, or other details before the assistant sees the information?
  • For a cloud service: what do its terms and settings say about storing your information and using it to improve their models, and can you turn that off?
  • Do any contracts, licenses, or professional rules you work under say where client information may be kept? If you are unsure, ask the person who advises you on those rules before you connect anything.
  • For a local setup: who looks after the machine, its updates, and its backups, and what happens when it breaks?
  • How smart does the job really need to be? Sorting and summarizing are different from careful writing or tricky reasoning.

Often the answer is both, or neither

You rarely have to pick one for the whole business. A sensible setup often splits the work: a cloud assistant for the jobs in the first two piles, and a local setup, or plain manual work, for the jobs that touch the third.

Sometimes the simplest answer is no AI at all for a particular job. If a private task takes ten quiet minutes a week, a careful person and a good folder may be the right tool.

A hypothetical example

This is an illustration, not a client story.

Picture a small bookkeeping practice. Its newsletter, its answers to common seasonal questions, and its website updates use no client details, so a cloud assistant drafts them and the owner reviews them. Client financial records are in the third pile, so they never go to the cloud assistant. After checking what its obligations allow, the practice tests a local model on an office computer for one narrow job: sorting incoming client documents into the right folders. If the test is not good enough, that job stays manual.

Where Grok Bot fits

Grok Bot is a cloud assistant, and it is usually where we start right now for the everyday jobs in the first two piles: drafting replies and posts, research, summaries of what you choose to share, and tidying documents, with the same limits we describe in the off switch note. For work that must stay in house, we look at a local setup or keep it manual. Other tools come in when they fit a job better. Where the information is allowed to go decides the tool, not the other way around.

Start with the piles

Before anyone talks about models or hardware, take one job and sort what it needs into the three piles. That short list tells you more about where the job should run than any comparison chart.