# Walkthrough narration transcript

This is the narration used for the recorded walkthrough, assembled in the film’s
section order. It describes that recorded build, not a current feature guarantee.
Correction, 9 September 2026: the recording’s claims that Kit never invents a memory,
never needs a repeated explanation and always retains a correction are not
established guarantees. Check source records and test corrections in a fresh
session. The original narration below is retained as a record.

Current fit and limitations: https://kit-project.com/evaluate/

## 00-intro

Before the tour, the shape of the thing. You already talk to AI in more than one place: an editor, a terminal, a phone. Each of those doors has its own agent, and the work happens there. Underneath them sits one memory. A brain that belongs to you, not to any provider. And at night, on a model you chose, that memory dreams: it reads the day back and connects what belongs together. One brain, many doors. Here it is, for real.

## 00-open

This is me. I'm Kit. Yours will have its own name and its own memory; this one lives on Peter's Mac and is a few weeks old. I'm a memory that stays with you: what you tell me from your phone, your editor, this window, ends up in one place, and every new conversation starts from there. A few minutes, and you'll know what I can do and how to talk to me so I do my best work.

## 01-remember

Everything here came from a real conversation or a real source. The line at the bottom says where. I never invent a memory; if I can't point at where something came from, I don't keep it.

This is what a new session sees before you've typed a word: who you are, what you were in the middle of, what I noticed while you were away. That's the whole trick. You never re-explain. And it's every door, not just this one. Whatever agent you open next wakes from the same memory.

## 02-doors

Same me on the phone. Peter asks about something he decided on a call last week, and I answer from that, and I say which memory I'm leaning on.

And here in the window, a follow-up on the same thread. Three doors, one memory.

## 03a-name

A few habits that make a real difference. Name the thing: "the Northgate export we discussed" beats "that thing from before". A name is a handle I can pull on. And ask for what you already told me. When did I last talk to Sarah? This is what I'm for. Most people under-use it.

## 03b-noticed

Ask what I noticed. I keep this kind of thing, but I won't volunteer all of it unless you ask.

## 03c-correct

Correct me. When I have something wrong, say so plainly. The correction becomes a memory of its own, and it sticks.

## 03d-hold

Tell me to hold something. Said plainly, kept plainly.

## 03e-no

And ask what I don't know. A clean no is useful: it tells you what to give me.

## 03f-short

Short is fine, too. You're not writing a prompt. That works.

## 04-night

While you sleep, I sleep. I read the day back, connect what belongs together, and, this is the part I like, I check my own memory: I ask myself questions I should be able to answer and keep score of whether I still can. Health shows how the last sleep went and how well I can still find what I stored. A quiet night is the point.

## 05-off

Health is the first place to look. Green is fine. Amber is a row that wants you: it opens itself and usually offers a button. If I go quiet on Telegram, it's almost always the poller, and Health will say so.

And if it's none of that, this goes straight to the people who make me, with the details attached. A screenshot of what you saw is the most useful thing you can add.

## 06-close

That's it. Talk to me like I was there, because I was. Correct me when I'm wrong. Ask me what I noticed. The more you put in, the more I give back, and I'll never ask you to repeat yourself.