An AI journal that remembers what you wrote
By The Architect team · Last updated
We make Architect. Everything this page says about how its memory works was checked against the app itself on October 17, 2026.
Yes. Architect is a private AI journal whose mentor reads each new entry next to a dated selection of what you wrote before: today's entries first, then up to ten recent ones, plus short dated notes and lines from older pages. When something you write echoes an earlier entry, the mentor can bring that line back, with its date, and set it beside today.
It does not reread your whole journal for every reply, and it can miss or misread something; you can correct it in the conversation. Memory works the same way on the free plan and the paid plan.
How it remembers, in plain words
- It reads today first. Every reply is written with today's entries first, then up to ten recent entries, each with its date.
- It keeps a short note of who you are. From your own entries, the mentor keeps a brief note of the people in your life, your work, projects, places, practices and values. It is encrypted like your entries.
- It marks the moments that matter. After each reply, Architect notes what the entry recorded: a decision, a commitment, a belief, a turning point, a fear, a hope. Each note keeps your own words and the date, encrypted with the same key as your journal.
- It finds the old line when it matters. While you write, your words are matched against those notes, so the mentor can reach a page from weeks or months ago. To find related past entries, what you are typing may be sent as you type; it is never stored.
- It shows you the receipt. When a reply quotes an earlier entry, the quote appears in a gold card with its date. The words in the card are copied from your entry, not reworded, and a tap opens the original page. Two small marks under it let you say That's right or Not quite. If a quote can't be matched to your entry, no card appears.
What it looks like
This is the app, with a made-up person called Alexandra. On Monday she wrote, "I'll start after the launch." On Friday she wrote that the launch had shipped, that she still hadn't started walking again, and that a recruiter wanted an answer by Monday. The reply opened with Monday's line and its date, then said what she had not: "That isn't laziness. 'After the launch' gave you a date to wait for, and the date has passed."
The mentor can do this because Monday's entry left a dated note with her exact words. Nothing in the reply had to be invented, and the card shows only what she actually wrote.
What it doesn't remember
- Not every page, every time. It does not reread your entire journal for each reply. Older entries come back through the dated notes and lines it kept, when they fit what you are writing now.
- Not perfectly. It can miss context or get something wrong. Tell it in the conversation, or tap Not quite on a card.
- Not on a page you can edit. The short note it keeps about who you are has no page to view or edit. You see memory when it is used: in the reply, with the date.
- Not without your key. Your journal and its memory are encrypted with a key that stays with you. If you lose your twelve recovery words (or your private password) and your session ends, the history can't be recovered, by design.
Free and paid
On the free plan, memory works the same way. Writing is free, and every entry saves. The free plan includes 11 full mentor replies, then one reply every 7 days. No card is needed.
The paid plan is $33 a month, $299 a year, or $9.99 a week. It adds mentor replies to every entry with no daily cap, Ask The Architect, which answers questions from your whole journal (up to 10 a day), and Insights, the patterns that keep coming back across your entries.
Want the category view first? What memory in an AI journal actually means explains the three kinds of memory and a five-minute test you can run on any app, this one included.