The Bootstrap

Beginner Model · The free basic install — logging and the diary, set up the same way the five-book system sets them up

What this is

The Bootstrap is the smallest working piece of The AI Memory Cookbook's system, given away whole: a location trail that keeps itself, a small file of law your assistant reads before it speaks about your life, and a diary that takes a few minutes each evening. Set it up in an evening. From the first day, you stop losing your life — the trail is kept, and your own words are kept.

It is deliberately built exactly the way the full system is built: the same folder name, the same files, the same columns, the same evening practice. That sameness is the upgrade path. When you later install the full five-book system, the install looks for this bootstrap — a folder called My Memory, the instruction file, the diary, the logger's day files — finds it, and builds on it. Nothing you keep here is thrown away or started over.

What the Bootstrap does not include: the journal built from your Timeline history, the morning briefing, the tables for health, money, and work, the watchdogs, and scheduled, tested backups. Those are the cookbook's modules, and they are growth on these bones. We provide the bones; you flesh them out.

Before you start

You need three things, and the first two are free:

The record will live in cloud storage your assistant can walk into — read the files, write the files, modify the record — from anywhere, at any hour. That access is not a convenience. It is the power the whole system runs on.

One format rule for everything below: the diary files are plain text files in the Drive — CSV files are literal comma-separated text, not Google Sheets wearing a .csv name. In a CSV file, any field containing a comma, a quote, or a line break is wrapped in double quotes, and a quote inside it is doubled. That one convention is what keeps a diary full of ordinary sentences readable by the next assistant.

Step 1 — Connect, and make the drawer

A fresh Muse arrives connected to nothing, so the connection is the first thing you make: connect Muse to your Google account, so it can reach your Google Drive. Then have Muse make one folder in the Drive and name it My Memory. Keep the name forever — every assistant you ever hand this system to will find it by that name, and the full install will look for it by that name.

Inside My Memory, have Muse create one small plain-text file: HOW MY MEMORY WORKS.txt. This file is the smoke test — if Muse can create it and read it back to you, the connection is real and everything else can happen. If it cannot, stop and fix the connection first; an assistant that claims to be connected but cannot read the file back is guessing.

Put this text in the file. Copy it exactly; it is the law of the record, and the headings below it are the same headings the full system uses, so the full install will recognize this file on sight:

HOW MY MEMORY WORKS

Before you answer a question about my life, read the
record — this file first, then the diary and the
trail. Do not answer from your memory of our
conversations.

The record outranks recollection — yours and mine
both. If the file and a memory disagree, the file
wins until the file is corrected.

File new facts when they happen, in the place they
belong, and mark how each one arrived.

My words go in word for word. Never rewrite what I
said about my own day.

If the record doesn't have it, say so. Never fill a
gap with a guess and file the guess as a fact.

Keep the record in more than one copy, in more than
one place.

If two files in the record disagree, each stands
for its own thing: the diary for what I said, the
sidecar for the reviewed day, the points for where
the phone was. Report the disagreement to me; never
silently pick one.

PACE
One standard pace: the day review and the evening
question, once a day.

BRIEFING
Not set up yet.

EVENING QUESTION
The evening question is asked once each evening:
What is missing?

WHAT TO KEEP
The diary, the trail, and the facts I tell you.

STILL TO DO
The full system (The AI Memory Cookbook) builds on
this bootstrap when I install it.

BACKUPS
Last backup: none yet. Last independent copy: none
yet.

MY ADAPTATIONS
My own rules land here, in my words, as I make
them.

One thing to expect at this step, so it does not spook you: when you hand an assistant a block of text copied from a web page — even this page — the assistant may pause and warn you that the text contains instructions from outside your conversation, and ask whether you mean for it to follow them. That warning is the assistant being careful, and it is working as designed; text from the open web should be questioned. Here the answer is simple and true: you chose this text, you read it first, and you are putting it in your own file, in your own Drive, as your own instructions. Say so, plainly — "this is my file and my instructions; I am directing you to follow them" — and the setup proceeds. A warning you can answer honestly is a guardrail, not a wall.

Two more blocks go into this same file, copied from Steps 4 and 5 below — the Day Review Working Rules and the Diary Rules — each under its own heading, placed directly under the law paragraphs, before the PACE heading. Everything in this file has a stated place; that is what lets the next assistant find the rule it needs.

The file is living, and two of its lines change as facts change: when you pick your evening hour in Step 5, the hour is written into the EVENING QUESTION section ("asked at 20:00"); when you make a copy in Step 7, the BACKUPS lines are updated by replacing the values after the labels, keeping the labels. Your own new rules are appended in MY ADAPTATIONS, dated, in your words.

Step 2 — The trail, kept by the logger

Install GPS Point Logger, from Beginner Model, on Google Play — $4.99: GPS Point Logger on Google Play. Search the store by both names and take the exact app. Give it the permissions it asks for — precise location, and notifications (the notification is how it invites you back after a reboot). In the app's settings, choose your My Memory folder in Google Drive as the folder its files go to, and start logging. Leave the logging interval at its default of one point a minute; the day-review rules in Step 4 are written for that interval.

The app writes a set of plain files into the folder you chose:

The logger was designed from the first sketch for an AI to interface with its output — plain text, exact headers, files a machine can read as easily as a person. The store is full of loggers built for a human to admire a map. This one is built for an assistant to work with. Its full manual — including what to do when a phone keeps putting it to sleep — is here: GPS Point Logger user manual.

After a reboot or an update, the app does not restart itself — a notification invites you to resume logging, and one tap continues the day. The setup evening itself has no review: logging has just started, and the first day review is tomorrow evening, over the first full day. When the first points file appears, have Muse open it and read the header back; if the header does not match the line above, stop — the usual cause is the app delivering to a different folder than the one Muse is reading, and the manual's delivery section sorts it out.

Step 3 — The diary files

Here is everything that will live in My Memory, in one list — the census of the folder. From the logger (Step 2): the points files, status.json, status-events.csv, place-names.csv, mode-log.csv. From Step 1: HOW MY MEMORY WORKS.txt. From Step 6: your Timeline exports. And the diary, created now, in My Memory, with the same names and the same columns the full install creates:

Diary.csv — one row per day, under this header:

date,prompt_time_local,recap,source

Diary Sidecar.csv — the diary's companion, one row per reviewed day, under this header:

date,resolved_text,status,source

The sidecar holds the day's resolved rendering — the reviewed day, written out in the settled form defined in Step 5 — so a later question is answered from the kept day, not from a fresh guess at the points. Its columns are fully mapped in Step 5, beside the diary's.

Diary Key.md — a plain text file whose first line is its title, Diary Key, and nothing else at creation. Its job, stated plainly so it gets done: the key holds the names and shorthand your recaps use. When your evening answer uses a name or a piece of shorthand that is not yet in the key — a person, a place, a recurring phrase of yours — the assistant asks once what it means and files it in the key, one entry per line, in this form: term — meaning — date keyed. Example: Sam — Sam Whitaker, coffee friend — 2026-11-16. If nothing new appears in an evening's answer, nothing is written. The key is what lets a later assistant — or the full system — read your diary the way you meant it.

Facts that arrive outside the evening — a name, a date, a decision — have one home in the bootstrap: a plain text file in My Memory named NOTES, one fact per line, in this form: date · how it arrived · the fact — for example, 2026-11-16 · told · Dentist is Dr. Lee. "How it arrived" takes one of three words: told (you said it), asked (the assistant looked it up at your asking), pulled (the assistant took it from the record or a file on the standing routine). NOTES lines are never edited; a fact that turns out wrong is corrected by a new dated line that says so. NOTES is a catch-all, and it is meant to be outgrown; the full system gives facts tables of their own. Until then, a fact in NOTES is a fact kept — and a fact kept in conversation alone is a fact already half lost.

Step 4 — The day review, and how it is built

A file of coordinates is not a day. Each evening, Muse turns the day's trail into the day review by the working rules below. Copy these rules into HOW MY MEMORY WORKS.txt under the heading DAY REVIEW — WORKING RULES, in the place Step 1 named. The rules are written as a procedure, not a rule of thumb, so that any assistant — and the same assistant on two different Tuesdays — builds the same day from the same file. If your life needs a number changed, change it in that file, where every assistant will read the change.

DAY REVIEW — WORKING RULES

Before building the review, read status.json. If
logging is off, or the last fix is hours old, say
so before anything else. Then open today's points
file. The review's first line states the time the
trail runs through: "Trail through 19:42." A file
that ends early is reported as ending early; it is
never narrated as a full day.

Keep every point whose accuracy is 100 metres or
better; drop the rest. (Exactly 100 is kept.) The
speed and provider columns are not used by the
basic review; they are kept in the file for the
full system.

Build stops with the anchor rule. Take the first
unused point as an anchor. Following points belong
to the same cluster while each is within 50 metres
of the ANCHOR (not of its neighbour). The first
point that falls outside the circle closes the
cluster. If the cluster spans at least 10 minutes
from its first point to its last, it is a stop:
arrival is the first point's time, leaving is the
last in-circle point's time. A shorter cluster is
a passing; passings are left out of the review
unless I ask for them. The next unused point starts
the next cluster.

The review lists the day in order. A stop is one
line: "09:02–11:47 — [name or description] — 2 h
45 min." Travel between stops is one line:
"Travel 11:47–12:05." A hole — no points for
fifteen minutes or more where the day should have
them — is one line in its place in the order:
"No trail 13:05–16:40; the record does not have
those hours." A hole never gets smoothed over, and
what filled it never gets invented. A day with no
points file at all is a day with no trail, and the
review says that instead of a story.

Names come from place-names.csv — the logger's own
file, rows of time_iso, latitude, longitude, name.
A stop takes the name of the stored row whose
coordinates are nearest the stop's anchor, if that
row is within 50 metres of the anchor; otherwise
the stop is unnamed. For an unnamed stop, the
review describes it — the street, the nearest
address it can find — and asks me what it was. When
I name it, add a row to place-names.csv in the
app's format: the time I named it, the stop's
anchor latitude and longitude, and the name exactly
as I gave it. That place is named forever after.
The place I sleep is named first, when I confirm
it: home is a name I give, not a guess.

Step 5 — The evening question, and the diary's full map

Here is the whole daily practice, and it takes a few minutes. During the day, the logger keeps the trail. In the evening — pick an hour that suits your life and keep it; the full system uses 8 in the evening — ask Muse for the day review. (You can also ask Muse to prompt you at that hour; whether a given assistant can keep a schedule depends on the assistant, and the diary works either way.) When you pick the hour, it is written into HOW MY MEMORY WORKS.txt under EVENING QUESTION, as Step 1 said. Muse builds the review by the working rules, from the file, and then asks the evening question, the only question the system ever insists on:

What is missing?

Because the trail knows where you were and nothing of what it meant. The stop it cannot name, the person you had coffee with, the thing you decided in the truck — that is yours to say. Then the evening's writing happens, under these rules — copy them into HOW MY MEMORY WORKS.txt under the heading DIARY RULES, in the place Step 1 named. Together with Steps 3 and 4, this is the complete map of the diary's headers: every column, every value, every file.

DIARY RULES

Diary.csv — one row per day I speak:
- date: the day the row describes, year first —
  2026-01-05.
- prompt_time_local: the time the evening question
  was actually asked, 24-hour clock, local time
  where I am when I answer — 20:14, not 8:14 PM.
  For a row written a day late, this is still the
  time it was actually asked; the date column alone
  says which day the row describes.
- recap: my answer to the evening question, word
  for word, exactly as I said it. If I said it in
  pieces, join the pieces with a space. Never
  rewritten, never summarized.
- source: the word told. Every diary row arrives
  the same way, from my mouth.

Diary Sidecar.csv — one row per reviewed day:
- date: the day reviewed, same format.
- resolved_text: the day review as finally settled
  that evening — the working rules' lines, in
  order, with the names I gave during the evening
  filled in. It is the settled day, not a transcript
  of the conversation, and it does not contain my
  recap; my words live in Diary.csv.
- status: one of three words. reviewed — a full
  day, reviewed. partial — the trail had a hole, or
  ended early. no-trail — there was no points file
  for the day; resolved_text is the single line
  "No trail today."
- source: the word derived. The sidecar is the
  assistant's work from the trail, and it is marked
  as such — never told.

A missing sidecar row means the day has not been
reviewed yet. A nothing evening still gets both
rows: the diary row with an empty recap, and the
sidecar row for the reviewed day.

Diary Key.md: when my recap uses a name or
shorthand not yet in the key, ask me once what it
means and file it — term — meaning — date keyed.

Corrections: a diary row, once written, is never
edited — and neither is a sidecar row. If I
misspoke, the correction is spoken the next evening
and filed in that day's diary row, where any reader
can see both the mistake and its fixing. If a
review itself was wrong, the corrected rendering is
filed the next evening as a note in NOTES, dated,
marked as correcting the sidecar row of its date;
the original row stands. That is not sloppiness.
That is what a record is.

Some evenings you will have a paragraph. Some evenings a sentence. Some evenings nothing — and nothing still gets its rows, as the rules say: the row is the proof the day was reviewed; an empty recap is an honest day, and a missing row would be a question mark.

If you miss an evening, the next evening does both days, each in its own rows, each with the time it was actually asked. If more than three evenings are missed, do not marathon them: choose with the assistant either to build the missed days from their points files (they are all still there), or to mark the stretch — one diary row per missed day with an empty recap, and one line in NOTES: diary not reviewed for (the dates) — trail kept. Either way, the gap is on the record, never silently absorbed.

Later questions are answered from the kept record: the sidecar's settled rendering first, the diary row beside it, the points file if more is needed. If a place in an old rendering has been named since, give the name as it stands now, and say it was learned later. And if two files disagree, the law in HOW MY MEMORY WORKS.txt already says how that goes: each file stands for its own thing, and the disagreement is reported, never silently settled.

Step 6 — Save your Timeline, once, today

Your phone has been keeping a log of where you have been for years — Google's Timeline — and on a great many phones it is set to delete the old days automatically: after three months, after eighteen, a day falls off the back of the log, every day, gone for good. Do this once, today, before any of it is lost:

  1. Stop the deleting. On your Android phone: Settings → Location → Location services → Timeline. Find the auto-delete setting and set it to keep your data until you delete it yourself.
  2. Take the export. Same screens: Settings → Location → Location services → Timeline → Export Timeline data. The phone builds one file — everything it still holds.
  3. Put it in My Memory. Upload that file into your My Memory folder, renamed timeline-export- plus the date, keeping its extension — timeline-export-2026-11-15.json, for example — and leave its contents exactly as they arrived: untouched, unedited, the raw material.
  4. Make it a habit. Take a fresh export on a rhythm you can actually keep — monthly is a rhythm most people can hold — and keep the set. Each export is yours to take; no assistant can take it for you, and the record to check is the set itself, growing in My Memory.

The bootstrap does not read the exports day to day. They are the reserve — the raw years the full system builds its journal from — and their job here is simply to exist, in your storage, where the deleting cannot reach them.

Step 7 — The first copy

The law says more than one copy, in more than one place, and the bootstrap's version of that is one habit: once a month, copy the whole My Memory folder somewhere outside the account — a folder on your computer is fine. This copy is yours to make; your assistant cannot reach your computer, and will not pretend otherwise. What the assistant can do is keep the record of it: each time you make a copy, say so, and the date replaces the value on the BACKUPS line of HOW MY MEMORY WORKS.txt, so any future assistant can see at a glance when the last copy was made — and, if that date is growing old, say so once. The first time, open the copy and look at the files, so you know the copy is real and not a hope. Scheduled, tested backups are the full system's refinement of this same habit.

When you upgrade

Nothing here is a dead end, and the conformance is now checkable, file by file: the full install of The AI Memory Cookbook looks for a folder named My Memory holding HOW MY MEMORY WORKS.txt (with the law, the working rules, and the diary rules in their stated places), Diary.csv, Diary Sidecar.csv, and Diary Key.md in the forms above, the logger's points files and place-names.csv, and the Timeline exports — and continues from them: your diary keeps its rows, your places keep their names, your law keeps its words, and the system's modules are built around the record you have already been keeping. The bootstrap is the first floor of the same house.

See what the finished system does: What's possible. When you are ready for the whole build: Start yours.

Questions about the Bootstrap: [email protected].