Why keyword search fails your journal

Here is a test you can run on any journal app you already use. Think of something that genuinely happened to you, then search for it the way you would describe it today.

Most people come up empty. Not because the entry is missing — because you did not write it in the words you are now using to look for it.

The vocabulary problem

When you write in a journal, you write from inside the experience. When you search for it later, you search from outside it, often years later, with different words.

You search “burnout”. What you wrote at the time was that you were tired of the project and did not want to open your laptop.

You search “when I met Anna”. What you wrote was about a loud dinner where you sat next to someone who had strong opinions about films.

You search “that cozy evening”. You have never written the word cozy in your life.

Keyword search compares strings. Your memory does not store strings. That gap is the whole problem, and it gets worse the more you write — a bigger journal makes the failure more frustrating, not less, because you know the thing is in there.

Searching by meaning instead

Mento indexes your entries as embeddings — numeric representations of what a passage is about, rather than which characters it contains. Two passages that describe the same kind of experience land near each other, even with zero words in common.

So the search for “that cozy evening” can surface the night you wrote about the rain, the slow dinner, and not wanting the conversation to end. No shared keyword. Same meaning.

This is also why you can ask questions instead of guessing at search terms:

• “How was I feeling about work in March?”
• “What did I say about that trip?”
• “Have I written about this before?”

You do not need to remember what you wrote. You only need to remember roughly what it was like — which is exactly the part human memory holds onto longest.

The part that matters most: it is built not to make things up

An AI that searches your life has an obvious failure mode, and it is not a small one. If it finds nothing and answers anyway, it invents a version of your past that never happened — and you have no way to tell, because you came here precisely because you could not remember.

Mento is designed hard against that. Whether a search actually returned anything is decided in code, not left to the model to judge: when it comes back empty, the companion is held to saying so — “I could not find anything about that” — and often suggests asking a different way, since a memory can be filed under a date you are not expecting.

It is a smaller-sounding promise than “AI that knows your life.” It is the one that makes the rest trustworthy.

What this feels like in practice

You stop maintaining your journal like a filing system. No tags to invent, no folder structure to keep consistent, no discipline about titling entries so future-you can find them.

You write. Later, you ask.

Mento is on the App Store — your journal becomes searchable from the first entry you write.