How do I export my Day One journal?

Open Day One on the desktop app (the export is fuller than the mobile version). Go to File → Export → JSON. You'll get a ZIP with one JSON file per journal and a media folder. The JSON preserves every entry, timestamp, location, and tag.

Is it weird to feed my journal to an AI?

It's a fair question and the honest answer is: depends on the AI. Dropping years of journal entries into a public chat is a privacy disaster. Loading them into a memory vault you control, where they're encrypted and never used to train anything, is closer to talking to a therapist who has read your file. Different surface, different stakes.

What does an AI actually learn from a journal?

The interesting stuff isn't the events — it's the patterns. The same fight you keep having. The same career anxiety that surfaces every September. The same hope you express the morning after a hard week. A decade of journaling shows the shape of you in a way no five-question onboarding ever could.

Will the AI judge me?

No. Models don't have opinions in that sense. But they can mirror back patterns you've half-noticed about yourself, and that mirror can feel intense the first time. Most people find the value is in the recognition, not the analysis. The model isn't smarter than you about your life — it just remembers more of it at once.

What if I don't want some entries used?

Any serious memory vault should let you mark entries private — kept in the vault, excluded from extraction. The grief entries, the relationship entries, the entries from the week you don't want to revisit. Privacy at the per-entry level is the only model that respects how journals actually work.

Does Konshus support Day One imports?

Yes, on Founding and above. Export the JSON from Day One, drop it in, and we extract atoms across the timeline. The full entries stay in your vault, encrypted, never used to train anything. The persona file is short enough to paste into any AI and reflects the shape of you over the years you've been writing.

A field note · ~7 min read

Day One journals + AI: what 10 years of you can teach any model

A long-running journal is the most honest dataset you'll ever own. Day One users with five, ten, fifteen years of entries are sitting on a corpus more revealing than any questionnaire could ever be. The question isn't whether your journal can teach an AI who you are. It's whether you want it to — and on what terms.

What's actually in there

Patterns, not events. The fight you keep having. The career anxiety that arrives every September. The relief that follows the morning runs.
Your voice over time. How you wrote at 26 vs 36. The phrases that come back. The hedges that disappeared.
People who matter. The names that show up across years — friends, partners, mentors, the brother you mention every birthday.
The shape of your seasons. What February usually looks like for you. What you do at the end of a project. How you write the day after a hard conversation.

A model that has read this and distilled it doesn't know more facts than you about your life. It just holds them all in mind at once — which you mostly can't.

The privacy bargain that matters

Dropping years of journal entries into a public ChatGPT conversation is a category error. The same data, loaded into a vault you control — encrypted, never used to train external models, with per-entry private flags — is a different bargain entirely. One is broadcasting your interior life. The other is teaching a private mirror.

The features that make this safe aren't sexy: encryption at rest, per-artifact private mode, hard-delete with a grace window, full export available forever. They're the plumbing that makes feeding your journal to an AI a considered choice rather than a leap.

The export, in 3 steps

Open Day One on desktop. File → Export → JSON.
Pick which journals to include (you can leave out the ones you'd rather keep separate).
Save the ZIP somewhere encrypted. The file has every entry, every timestamp, every tag, every location.

Short answer

A decade of journal entries can teach an AI more about you than any onboarding ever could — patterns, voice, the people who matter, the shape of your seasons. The question is which AI you trust with it. Export Day One as JSON, load it into a vault you control with per-entry privacy, and use the distilled persona with whichever AI you're using on any given day.

The portable answer

Konshus has a Day One importer on Founding and above. The JSON export goes in once; the distilled persona is what gets used. Entries are encrypted at rest, never used to train anything, and the per-artifact private flag lets you keep the entries you don't want analyzed out of the extraction pipeline entirely.

See also the lost AI memory problem for why decade-long context is so rare — and what happens to your AI when you die if you've ever wondered where the journal goes after.

Frequently Asked Questions

The journal already remembers. The AI should too.

Konshus turns a decade of Day One entries into a portable persona — encrypted, private by default, and yours forever.

Meet Konshus