Local AI vs cloud AI: where the model runs and who bills you

What is local AI, and what is cloud AI?

Local AI is an AI model that runs on your own computer, so your question and the answer never travel to a company’s servers. Cloud AI is a model that runs on a company’s servers: you send the question over the internet, the company runs the model and sends the answer back. Both are the same kind of program, a large language model; the difference is where it runs and who pays for the computing.

Cloud models are usually the largest, and local models are the ones you can keep entirely under your own roof. Minda Super Intelligence, the personal AI agent app for your computer, offers both, so you can choose per chat.

How do the three ways work in Minda?

You pick the model in the chat’s model menu, and the menu groups the models by where they run: Minda AI at the top, then Ollama and each provider you added a key for. Keep SI on this computer is a switch above all three: while it is on, Minda refuses every cloud call before anything is sent, and only a model in Ollama answers.

What is an API key, and where do you get one?

An API key is a long secret string of letters and digits that tells an AI company which account to bill when software, such as Minda, asks its model a question. You get one from the provider’s own developer site, usually after you create an account there and add a payment method, and you paste it into Minda once.

Minda takes keys for OpenAI, Anthropic (Claude), Google (Gemini), DeepSeek, OpenRouter, which offers models from many vendors behind one key, and Alibaba Qwen. The key stays on your computer, sealed with the operating system’s key store, and it never syncs to your other devices, so each computer needs its own. Treat it like a password: anyone who has it can spend your balance at that provider.

An example: a Salt Lake City accountant picks per task

Dana runs a small bookkeeping practice in Salt Lake City and uses all three ways in one day.

1. Morning, client ledgers: she turns on Keep SI on this computer in Settings → AI → Privacy and asks a model in Ollama to explain odd entries. The request stays on her computer, and the AI activity log shows "This computer" as where it went.
2. Late morning, a question about a new state filing rule: she picks Allow cloud models in the model menu, which turns Keep SI on this computer off, and chats on her own OpenAI key. OpenAI bills her, not Minda.
3. Afternoon, a long research report on payroll software: she picks a Minda AI model and spends credits from her Free plan. Its price in credits showed before she picked it.
4. Evening: she opens the AI activity log and finds three kinds of rows, one for This computer, one for the provider’s host and one for Minda AI, each with the model and the character counts, never the words.

What can you use each way for?

• Local, with Ollama: material you would rather keep on your computer, such as client notes, contracts or a journal, with no credits spent.
• Your own key: daily chat billed by a provider you already use, on any model that provider offers, with no Minda credits spent.
• Minda AI: no setup and one balance across the SI chat, the SI agent, deep research, voice and pictures, with premium models for hard tasks.
• Compare: ask 2 or 3 models from the menu at once and keep the answer you like.
• A mix: each chat can use a different model, so you can choose by task, price and how private it must be.

How to set each one up in Minda

1. Minda AI: sign in to your Minda account, press Alt+Q for the SI chat, open the model menu and pick a model from the Minda AI group. Its price in credits shows at the end of its row.
2. Your own key: in the model menu choose Add API key, or open Settings → AI → Cloud providers, and paste the key in the provider’s row. Its models then appear in the menu, in a group marked "your key".
3. Ollama: install it from ollama.com (the chat’s Get Ollama button opens the page), pull a model and leave Ollama running. Settings → AI → On this computer shows Running and the number of models; they appear in the menu under Ollama (on this computer), marked free.
4. To keep everything local, open Settings → AI → Privacy and turn on Keep SI on this computer. Cloud models are listed as off, and Allow cloud models turns the switch off again.
5. To check what happened, open Settings → AI → Privacy → AI activity and press View: each row shows the provider, the model, where the request went and what went along with it.

What does each one send, and what does the log show?

Every AI request sends your question, the chat so far, and the pages, files, notes and pictures you add, plus what the SI agent’s tools read for it. Where that goes depends on the way you picked.

• Minda AI: to Minda’s server and then to the vendor of the model you picked. The server keeps only the model, the credits and the time for 92 days, never your words.
• Your own key: direct to the provider you picked, under that provider’s own terms. Minda cannot say what a provider keeps, so read its policy.
• Ollama: the request stays on the computer, because the model runs there.
• The AI activity log: the provider, the model, the host, character counts and page titles for every request, never the words. It is kept on your computer, and a private window’s requests stay in memory only.