Safe for the Mac that gives. Careful with what students ask.

Two people trust Tokens for Good: the person who lends a Mac and the student who asks it a question. This is how each of them is protected today, and what is still being built.

Stronger isolation, being builtIn progress

Today the model runs on the donor’s own computer, through Ollama, and our agent only passes questions in and answers out. The computer handles that text in plain text. An extra layer of isolation, to protect the donor’s machine and to keep what students ask private, is being built now.

The app never shows donors the questions

Point at the heart in the notch and its panel shows that your Mac is helping and how many requests it has answered. Your dashboard shows how much it did and which causes it helped. Neither shows what was asked, or what the answer said.

The text is still processed in plain text on the donor’s computer, so its owner could in principle read it. That is why there are two tiers, and each school or cause works in one of them.

Open tier

Any donor’s computer
Only for information that is not sensitive

Trusted tier

Only donors under contract and a data processing agreement
For more sensitive work

Students never see whose Mac answered

To a student it is simply the AI tool their school already uses. Nothing in an answer names the person who lent the computer. When a donor is matched with a school or cause, that organisation sees the donor’s name on its dashboard.

Nothing on the Mac is opened up

The agent only connects outwards, to Tokens for Good, and listens on no port, so nothing on the internet can reach your Mac through it. Ollama, which runs the model, by default only accepts connections from the Mac itself. Our app gives the model no access to files, apps or tools. It receives the text of a question and sends back an answer, and that is all we ask of it.

Counts are kept, never content

What is kept is enough to run budgets and show the impact, and nothing that says what anyone asked.

Kept

Which school or cause sent a request, and when
Which Mac answered it, with which model
How long the question and the answer were, in tokens

Never kept

What a student asked
What the answer said

Our tests send a marked question through the whole system and check that it never turns up in our database or our logs.

Questions people ask The safeguards on your Mac

If you have a MacDonate computeYour Mac answers their questions while it is plugged in.If you are a school or a causeApply for computeFree AI for your students or your team, on open models.