Levia · Guides

AI journaling: what it is and how to keep it private

By the Levia team · Aleana AG · Published July 14, 2026 · Updated July 14, 2026

What is an AI journal?

An AI journal is a journaling app with a companion built in: you write, and the AI responds — with a reflective question, a gentle summary, or a quiet observation about something that keeps returning in your entries. The writing is still the heart of it; the AI is there to help you see your own words more clearly.

It's just as important to name what an AI journal shouldn't be. It isn't an audience — you're not writing to impress it, and it has no opinion on whether your day was good enough. And it isn't a therapist: it can't diagnose, treat, or replace a human who is trained to help. A well-built companion knows its lane — questions, summaries, patterns — and stays in it.

What can an AI companion add to journaling?

Three things, mostly. It asks the question you didn't think to ask: you write that the week felt heavy, and instead of a generic reply you get "what made Thursday different?" It remembers threads: the project you were dreading in March, the friend you meant to call — and gently picks them up when they matter again. And it turns weeks of scattered entries into visible patterns: the way your mood dips after short nights, or lifts on days you spend outside.

What it adds, in short, is a second reading of your own words. What it never replaces is the writing itself — the untangling happens while you type, not while the AI responds. If a companion did the writing for you, the whole point would be gone. The writing stays yours.

What are the privacy risks?

The core risk is simple: for an AI to respond to your words, something has to read them — and where that reading happens matters more than any feature. A journal is among the most sensitive text you will ever produce. It holds what you wouldn't say out loud: health worries, relationship doubts, half-formed fears, other people's names.

The risks worth knowing are mostly structural, not sinister:

• Cloud processing by default. Many apps send every entry to servers just to generate a response — without asking first.
• Vague retention. If a policy can't say clearly how long your words are stored, assume it's longer than you'd like.
• Training on user content. Some services improve their models on the text users type in. A journal entry is text users type in.
• Accounts that link words to identity. An email address sitting next to your most private sentences turns a journal into a profile.
• Analytics and ads SDKs. Even when entry text stays private, third-party trackers can reveal a lot about when, how often, and how long you write.

None of this makes AI journaling reckless — it makes the details decisive. The same feature can be built carefully or carelessly, and in the screenshots the two look identical. That's why it's worth asking specific questions.

How do I evaluate an AI journal's privacy?

Start with where the AI runs — almost every other answer follows from that one. Then work down this list; a trustworthy app answers all of it in plain words, without making you dig:

• Where does the AI run — on-device or on servers?
• If the cloud is involved, is that opt-in — or silent?
• What is the retention policy, stated in plain words?
• Is your content used to train AI models?
• Is an account required to use the app?
• Can individual entries be encrypted beyond the app lock?
• Can you export everything, any time?
• Does the privacy policy name the AI provider it relies on?

Vagueness is itself an answer. If it takes a lawyer to work out whether your entries train a model, decide accordingly.

How does Levia answer these questions?

Since this guide lives on Levia's site, here is Levia's own answer sheet — the same questions, taken in order.

Where the AI runs: on iPhones with Apple Intelligence (iPhone 15 Pro and newer, on iOS 26 and later), the companion runs entirely on-device through Apple's Foundation Models. Your entries are processed on your phone and never leave it. On other devices, Levia can use cloud AI instead — but only if you turn it on. It is strictly opt-in and off by default, runs with zero retention, is never used for training, and is never linked to an identity, because there is none to link: Levia has no accounts.

Beyond the app lock: individual entries can be sealed in the Secure Enclave — the iPhone's hardware security chip — so they open only with your Face ID and are never sent anywhere, in either mode. Backups are encrypted with AES-GCM, with keys derived via PBKDF2 at 600,000 iterations, and everything can be exported as PDF or Markdown whenever you want. There are no ads and no trackers.

The full detail is in the privacy policy; the short version lives in the privacy section of the homepage. Judge Levia by the same checklist you'd apply to anyone else — that's what it's for.

Is my journal used to train AI?

For Levia: never, in either mode. On-device processing means your words don't go anywhere at all, and the opt-in cloud mode runs with zero retention and no training — your entries don't improve any model, Levia's or anyone else's.

For AI apps in general, the honest default is to assume yes until a policy says otherwise in plain words. "We may use your data to improve our services" is not a no. Look for the explicit sentence — your content is not used to train AI models — and if you can't find it, treat the silence as the answer.