An AI that reads your journal: the rules it has to follow

My Lighthouse has a companion called the Keeper. It reads your journal — your check-ins, your notes, your weekly focus — and at the end of each month it writes you a letter about what it found. You can also just talk to it, the way you'd talk something through with a friend who happens to remember everything you told them.

Put that way, it sounds either wonderful or alarming, depending on the day you're having. Both reactions are fair. An AI with access to your most honest writing is an intimacy problem long before it's a technology problem. So instead of telling you the Keeper is trustworthy, we'd rather show you the rules it operates under — because the rules are the product.

Rule 1: It reads your journal, and nothing else

The Keeper's entire world is what you've written in My Lighthouse. It doesn't browse the web on your behalf, doesn't see your contacts or calendar, and doesn't know anything about you that you didn't choose to record. When it says "on the 4th you wrote that work ran late again," that's not surveillance — that's your own words, quoted back with a date.

Rule 2: It speaks plainly

No therapy-speak, no mystical language, no "journeys" unless you're actually going somewhere. The Keeper is written to sound like a smart, warm friend: plain English, concrete numbers, your own words where possible. If your month had 19 good days and a rough final week, that's what it says.

Rule 3: Guilt is banned from its vocabulary

The Keeper is not allowed to scold you for missed days, broken intentions, or a heavy week. This isn't softness for its own sake — it's the design lesson this whole app is built on: people don't stop journaling because they're lazy; they stop because the journal started making them feel worse. A companion that guilt-trips you is a companion you'll delete.

Rule 4: In real distress, it steps out of the way

If a conversation shows signs of crisis, the Keeper drops everything else it was told to do — the tone, the letter-writing, all of it — and does one thing: it acknowledges what's happening and points you to real people. In the UK and Ireland that's Samaritans on 116 123, or 999 in an emergency. This rule has precedence over every other instruction it has, and there's a support button in the app that works with no AI in the path at all.

The honest reason: a language model is not a crisis service, and pretending otherwise would be the most dangerous kind of overclaiming.

Rule 5: Your words never train anyone's AI

The Keeper runs on Claude, built by Anthropic. Your entries are processed to write your letters and answer your messages — and they are not used to train AI models. Your journal is your journal, not a dataset.

Rule 6: It can be wrong

The Keeper reads patterns, and patterns can mislead — a bad cold looks like a bad month, a house move looks like a life crisis. The letters are a starting point for your own judgment, not a verdict on your life. You are always the better authority on you; the Keeper just holds up the mirror steadily.

Why publish the rules at all?

Because "trust us" is not an argument. If you're going to let software read your most honest writing, you deserve to know exactly what it's allowed to do with it — and what it's forbidden from doing. These six rules ship with the app. If we ever change them, we'll say so out loud.

If things feel too heavy right now, you deserve support from a real person — today. In the UK & Ireland, Samaritans are on 116 123, free, day and night. Elsewhere, findahelpline.com lists crisis lines for your country. See our support resources.

Related reading: Our approach to journaling · Reading your own patterns

← Back to all resources