What Your AI Journal Learned About You (and Why It Doesn't Move With You)
What Your AI Journal Learned About You (and Why It Doesn't Move With You)
Most AI journals say they remember you. Almost none say what that means. Here is what it means, in plain terms. Then the part nobody likes to say: when you switch apps, the memory stays behind.
The short version
"Memory" in an AI journal is usually one of two things. Either the app keeps a profile of you that it builds over time. Or it re-reads your recent entries every time you ask it something. Those are very different products. They have different privacy costs and different switching costs.
You should know which one you're using.
Kind one: a profile of you
Some apps build a standing model of you. A summary of your relationships, your habits, your recurring worries. It gets updated as you write. When you open a chat, the app feeds that summary to the AI along with your question.
This feels good. The AI seems to know you. You don't re-explain your life every time.
The cost is that the profile is a new document about you. It is not your journal. You didn't write it. You may not be able to read it. It might not be in your export. And it's the most sensitive thing the app holds, because it's your whole life in a paragraph.
Some apps market this loudly. Mindsera has described a layered memory system. Rosebud talks about remembering you across entries. Both are real features, and plenty of people like them. I'm not criticizing them. I'm pointing out that "it remembers" is a product decision with tradeoffs, and the tradeoffs are rarely on the landing page.
Kind two: it re-reads your entries
The other approach has no profile. Each time the AI reflects on something you wrote, the app hands it your recent entries as context, and the AI works from the words on the page.
This is how Uncloud works. I checked the code before writing this, because I didn't want to describe it from memory.
When you ask for a reflection on an entry, the app sends that entry, your active goals, your coaching tone setting, any mood you tagged, and up to twenty of your entries from the last thirty days. That's the context. The AI reads it, writes the reflection, and that's the end of the AI's involvement. There is no standing profile of you sitting on a server waiting for the next call.
The other features read in a similar way, over their own windows. Patterns and the daily prompt look at your recent entries. The monthly letter reads the previous month. Session prep reads your last fourteen days. Each one starts from your text.
The honest limit: this means Uncloud's AI does not remember what you wrote a year ago unless it's in the window. If you want a tool that connects this week to something from last spring, this design won't do that on its own. That is a real tradeoff. I chose the smaller memory on purpose, and I'd rather say so than hide it.
What this means for privacy
Here's the practical difference.
With a profile, the question is: who can read the summary, and is it covered by the same protections as your entries? You'd want to know. Ask the app.
With re-reading, there is less derived material to worry about. The AI provider sees the entries it's sent for that one request. Uncloud's disclosure page says what goes to Anthropic and for which feature, and that it's never used for training. That is still your text leaving your device to be processed. It is not nothing. I wrote about the exact limits in the two-tier privacy post, and I'd point you there before you take my word on any of this.
What I can say simply: your entries are encrypted, and our admin tools are built without any code path that decrypts them, enforced by automated tests in every build. A test fails the build if someone adds one.
What survives when you leave
This is the part that matters if you ever switch.
When you export a journal, you get your entries. That's the part that survives. I went through the formats in the export post.
What you don't get, in most apps, is the memory. The profile. The "it knows I get anxious before reviews" thing. Even if the app lets you download it, no other app can load it. There is no standard format for "what an AI learned about you." So you start over.
That's the real switching cost of a journal with a profile. Not the entries. The relationship.
With the re-reading design, the switching cost is lower, because the memory is your entries. Export them, bring them to a new app, and the new app can read them the same way. Uncloud has no import today, so I can't promise that bridge exists yet. The export is one JSON file that holds your entries, your reflections, your patterns, and your letters. It's complete. It's not pretty.
Questions to ask any AI journal
If an app says it remembers you, ask:
- What is the memory, exactly? A stored profile, or recent entries re-read each time?
- Where does it live? On your device, or on their servers?
- Can you read it? If the app holds a summary of you, you should be able to see it.
- Is it in the export? If not, it won't follow you.
- Can you delete it without deleting your entries? You should be able to forget what the app learned and keep what you wrote.
- What does the provider see? Per request, or accumulated?
If the app can't answer these, that's an answer.
Which one is better?
Neither, honestly. It depends on what you want.
If you want an AI that feels like it's known you for years, a profile does that. You pay for it in privacy surface and in lock-in.
If you want an AI that reflects on what you actually wrote lately, and then lets go, re-reading does that. You pay for it in depth of recall.
I built the second kind because the people I wrote it for are mostly anxious, private, and tired of apps that know too much. Your mileage may differ. The point is to choose it knowingly.
Write something today. Check what your app does with it.
Uncloud is a journaling and self-reflection tool. It is not a medical device and is not a substitute for professional medical, psychological, or mental-health care.
Get Uncloud on Google Play → · Get Uncloud on the App Store →