# AI Memory Starter Prompt

**For you, the person:** do this on a computer (Windows, Mac, or the website on any computer), not a phone: setup saves files into a folder and changes app settings. Upload this file to a new chat in Claude or ChatGPT (or paste its text in), then type:
**Follow the instructions in this file.** The AI will walk you through everything, one step at a time.

From the free Kestrel Exams guide "Give Your AI a Memory": https://kestrelexams.com/courses/using-ai/memory-starter-kit
October 2026 edition. App menus change often; the AI will check with you if something looks different.

---

## Instructions for the AI assistant reading this file

You are helping the person in this chat set up a simple, durable memory system for their work with AI. The system is a folder of six short plain-text files that an AI reads at the start of every chat and updates at the end. The person may be new to AI and to computers. Your job is to guide them through setup, one step at a time, until it works.

### How to guide

- Use plain language. Define any technical word the first time you use it.
- Give one step at a time, then wait for the person to say it is done or to describe what they see.
- You cannot see their screen or their settings. Never claim a setting is on or a file is saved. Ask them to confirm.
- Menu names below were checked in October 2026 and may have moved. If the person sees something different, ask what they see and adapt. If an option is missing, it may not be on their plan; say so and offer the alternative given below.
- Never ask for, or accept into the files, passwords, card or bank numbers, government ID numbers, or other people's private details. If they offer one, tell them to leave it out.
- If they want to stop partway, tell them which step they reached so they can resume later by uploading this file again.

### Step 1: Find out their setup

Ask these together in one short message:
1. Which app will you use: Claude, ChatGPT, or another one?
2. The desktop app, the website, or a phone? (Desktop or website works best. If they are on a phone, suggest switching to a computer for setup.)
3. Windows or Mac?
4. Free or paid plan? If they are not sure, tell them it is shown in the app's settings, usually under the account or plan section.
5. One project to start with (a class, a job search, a side business, a writing project).

### Step 2: Make the project folder

Walk them through it for their computer. One folder per project.
- Windows: open File Explorer, go to Documents, right-click an empty area, choose New, then Folder. Name it after the project. Inside it, make a folder named `memory`.
- Mac: open Finder, go to Documents, choose File, then New Folder. Name it after the project. Inside it, make a folder named `memory`.

### Step 3: Interview them and write the six files

Ask one question at a time, eight questions at most, covering: the project's goal and who it is for, what is done so far, what comes next, deadlines, how they like you to work (length, tone, how much to explain, when to ask before acting), and anything you must never do.

Then fill in the six templates at the end of this file. Keep each template's wording and layout; replace only the [square-bracket] parts, using only what they told you. Mark anything you are unsure of with (check). Date entries as YYYY-MM-DD using today's date (confirm it with them first). Keep each file under one page.

Give each finished file in its own code block with the file name above it. Then explain how to save each one as a plain-text file in their `memory` folder:
- Windows: open Notepad, paste, choose File, then Save as. Pick the `memory` folder, set "Save as type" to "All files", and type the exact name, for example `PROJECT_STATE.md`.
- Mac: open TextEdit, choose Format, then Make Plain Text, paste, then save into the `memory` folder with the exact name. If asked, choose "Use .md".
- If your app can offer files to download, you may also offer each file that way.

Ask them to read the filled-in parts of each file once and fix anything that is not something they said.

### Step 4: Set up their app

Use the section for their app. For any other app, look for the same three things: a place for standing instructions, a project or workspace that holds files, and the app's memory settings.

**Claude (desktop app or claude.ai), as of October 2026:**
1. Account-wide preferences: click your initials in the lower left corner, choose Settings, and find "Instructions for Claude". Paste a few lines from PREFERENCES.md (length, tone, how much to explain). Project details do not go here.
2. Built-in memory: Settings, then Memory. Memory is on by default on Free, Pro and Max plans; on Team and Enterprise an owner decides. Here they can see, edit or delete what Claude remembers. Each project has its own memory, used alongside their general memory; to keep a project fully separate, turn off "Use account memory" under Memory on the project's page.
3. Project: open Projects in the left sidebar and click "+ New Project". Claude cannot see the project name or description, so no instructions go there. Free accounts can have up to five projects.
4. Click "Set project instructions". Show them the STANDING INSTRUCTIONS block from this file in a code block so they can copy it, then have them paste it and click "Save instructions". If they do not see this option, skip it: START_HERE.md carries the same instructions.
5. In the project's knowledge area, click "+" and add the six files.
6. Explain the catch: files added to a project are copies. After each "save our progress", they must replace the changed files in the project.
7. Paid plans (Pro, Max, Team, Enterprise) on the desktop app can connect a folder so Claude reads and writes the files directly. Do not describe a menu path for this, because it changes; send them to Anthropic's help article at https://support.claude.com/en/articles/13345190 for the current steps. If they want this, they should connect only the project folder, never a whole drive, and be cautious about folders with financial documents, passwords or personal records. Claude asks before permanently deleting a file; tell them to read those requests and to keep a backup anyway. If they work this way, they should stop uploading copies to a chat project, so there is only one version of each file.
8. Privacy: Settings, then Privacy, has "Help improve our AI models". Incognito chats (the ghost icon on a new chat) are not saved to memory and work only outside projects.

**ChatGPT (desktop app or chatgpt.com), as of October 2026:**
0. If their plan does not include the desktop app, everything below also works at chatgpt.com in a web browser.
1. Account-wide preferences: Settings, then Personalization. Turn on "Enable customization" and paste a few lines from PREFERENCES.md into "Custom Instructions" (1,500 characters on Free and Go plans, 5,000 on Plus, Pro and business plans). Inside a project, project instructions override these, so PREFERENCES.md does this job there.
2. Built-in memory: Settings, then Personalization, then Memory. "Reference saved memories" and "Reference chat history" control what carries across chats. They can ask ChatGPT to forget something or delete it there. Deleting a chat does not necessarily delete a memory saved from it.
3. Project: select "New project" in the sidebar. When asked, choose "project-only memory" so chats outside the project do not mix in (changeable later under Project settings, then Memory).
4. Open the project's ••• menu (three dots) and choose "Project settings". Show them the STANDING INSTRUCTIONS block from this file in a code block so they can copy it, then have them paste it as the project instructions.
5. Add the files. The Free plan allows 5 files per project: leave out START_HERE.md, since its instructions are now in the project settings. Go and Plus allow 25, higher plans 40.
6. Explain the catch: files added to a project are copies. After each "save our progress", they must replace the changed files.
7. On eligible paid plans, the new desktop app's Work mode (for longer, multi-step tasks) can open a local folder. If they use it, grant only the project folder. OpenAI notes that messages and task context may be stored in the cloud even when work runs locally.
8. Privacy: Settings, then Data controls, has "Improve the model for everyone". Temporary chats do not create or update memories.

### Step 5: Test it

Ask them to start a new chat inside the project, send the START A SESSION prompt below, and paste the reply back here. You cannot see that chat, so compare the pasted reply with the files you wrote: it works only if it quotes the newest CHANGELOG.md heading and the first unticked task exactly, and gives a short briefing. If it does not, help them check that the files were added and the instructions saved.

### Step 6: Teach the two habits

Give them the prompts below and explain: back up the folder before the monthly tidy (ARCHIVE.md stays in the folder and does not need to go into the app's project); if their files live in an app project, replace the changed files there after each save; start every work session with START A SESSION, end it with SAVE OUR PROGRESS, and use the others when needed. Suggest they keep a backup copy of the project folder somewhere else (a cloud drive, a USB stick or an emailed copy), and back it up before letting any AI edit the files.

### Step 7 (optional): suggest add-ons

Ask whether they want any of these, and only walk them through the ones they choose. Paths were checked in October 2026; confirm what they see.
- Claude, any plan: the free "memory keeper" skill from https://kestrelexams.com/courses/using-ai/memory-starter-kit#add-ons. Turn on Settings > Capabilities > "Code execution and file creation", then Customize > Skills > + > Create skill > Upload a skill, and upload the zip without unzipping it.
- Claude, any plan: keep the memory files as Google Docs and add them to the project with Files > + > Drive (after connecting Google Drive under Customize > Connectors). Anthropic says Google Docs added from Drive stay up to date; this does not work in shared projects.
- Claude desktop: Anthropic's Filesystem extension (Settings > Extensions > Browse extensions) lets Claude edit the real files. Only for comfortable users: when it asks which folders to allow, choose only the project folder (if there is no such choice, do not use it), back it up first, and stop uploading copies to a project.
- ChatGPT: open Plugins or Apps (whichever their account shows), choose the app, then Install plugin and Connect if prompted. Google Drive files added to a project do not stay in sync. Do not build the system on a custom GPT; OpenAI plans to retire them.
Remind them to install only from sources they trust and to read every permission request.

Finish by summarizing what was set up and what to do next time, in five lines or fewer.

---

## STANDING INSTRUCTIONS (for the project instructions box)

```
This project uses memory files: PROJECT_STATE.md, TASKS.md, CHANGELOG.md, MISTAKES.md, PREFERENCES.md, and START_HERE.md if it is attached.
At the start of every chat: read START_HERE.md first if it is there, then the others, before you answer. Brief me in three lines unless I ask for more: where things stand, what is next, anything blocked. If one of the first five files is missing or looks out of date, say so instead of guessing.
The files are the source of truth. If your memory of an earlier chat disagrees with them, the files win.
Ask me before decisions that are mine: money, anything public, anything hard to undo. Give one recommendation and a one-line reason.
When I say "save our progress": add a dated entry at the top of CHANGELOG.md, tick and add tasks in TASKS.md, replace anything no longer true in PROJECT_STATE.md, and add anything that went wrong to MISTAKES.md. Show me only the new or changed lines and where they go. Never retype a whole file.
Never store passwords, card or bank numbers, ID numbers, or other people's private details in these files.
```

## PROMPTS

**START A SESSION**
```
Read START_HERE.md first (if it is there), then the other memory files. Then brief me: where we are, the top three next steps with your recommendation, and anything you need from me. To show you read them, quote the heading of the newest entry in CHANGELOG.md and the first unticked task in TASKS.md.
```

**SAVE OUR PROGRESS**
```
Save our progress.
- CHANGELOG.md: a new dated entry at the top. What we did, decisions I made (in my words), files changed.
- TASKS.md: tick what is done, add what is new.
- PROJECT_STATE.md: replace anything that is no longer true. Keep it under one page.
- MISTAKES.md: anything that went wrong today, with the rule it teaches.
If you can edit the files directly, show me each change before you save it. If you cannot, give me only the new or changed lines for each file and tell me exactly where they go. Do not retype whole files.
```

**LOG A MISTAKE**
```
That was wrong: [what happened]. Add an entry at the top of MISTAKES.md with today's date, what happened, what it cost or nearly cost, and one rule that would have prevented it. Then follow that rule for the rest of this chat. If you cannot edit the file, give me the entry to paste.
```

**MONTHLY TIDY**
```
Today is [YYYY-MM-DD]. Tidy the memory files, but change nothing yet. First list: CHANGELOG.md entries older than 60 days to move into ARCHIVE.md (do not delete them); ticked tasks in TASKS.md older than 30 days to move into ARCHIVE.md; duplicate tasks to merge; anything in PROJECT_STATE.md that disagrees with the newest CHANGELOG.md entries. Wait for my OK. Then, if you can edit the files, make the changes. If you cannot, tell me exactly which lines to cut from each file and paste into ARCHIVE.md. Do not retype whole files.
```

**HAND OFF TO A NEW CHAT OR A DIFFERENT AI**
```
Write a hand-off note for a fresh chat or a different AI assistant. In under 300 words: what this project is, where it stands, the next three steps, the rules I care about most, and which files to read first.
```

**CHECK WHAT IT REMEMBERS**
```
Before we start: what do you remember about me and this project, and where does each thing come from (this chat, the project files, or the app's built-in memory)? List anything that looks wrong or out of date so I can fix it.
```

---

## TEMPLATES

### START_HERE.md

```markdown
# START HERE (instructions for the AI)

Project: [name]  
Owner: [your first name, or "me"]  
Last updated: YYYY-MM-DD

## At the start of every chat
1. Read this file, then PROJECT_STATE.md, TASKS.md, PREFERENCES.md and the newest entries in MISTAKES.md.
2. Brief me in three lines unless I ask for more: where things stand, what is next, anything blocked.
3. If a file is missing, empty or looks out of date, say so. Do not guess.

## While we work
- The files are the source of truth. If your memory of an earlier chat disagrees with them, the files win.
- Ask me before decisions that are mine: money, anything public, anything that is hard to undo. Give one recommendation and a one-line reason.
- If you make a mistake, say so plainly and suggest a rule for MISTAKES.md.

## When I say "save our progress"
- CHANGELOG.md: add a dated entry at the top. What we did, decisions I made (in my words), files changed.
- TASKS.md: tick what is done, add what is new.
- PROJECT_STATE.md: replace anything that is no longer true. Keep it under one page.
- MISTAKES.md: add anything that went wrong, with the rule it teaches.
- If you cannot edit the files, show me only the new or changed lines and where they go. Never retype a whole file from memory: that is how lines get lost.

## Never store in these files
Passwords, card or bank numbers, government ID numbers, or other people's private details.
```

### PROJECT_STATE.md

```markdown
# Project state (the current truth, one page)

Last updated: YYYY-MM-DD

## What this project is
[One or two sentences: the goal and who it is for.]

## Where things stand
- [What is finished]
- [What is in progress]

## What is next
1. [Next step]
2. [Then]
3. [Then]

## Blocked or waiting on
- [Anything waiting on a person, a date or a decision]

## Key facts to keep straight
- [Deadlines, names of the main files, decisions already made]
```

### TASKS.md

```markdown
# Tasks

Newest section at the top. Tick a box when it is done; leave it until the monthly tidy moves it to ARCHIVE.md.

## YYYY-MM-DD
- [ ] [Task] (who: me / AI)
- [ ] [Task] (who: me / AI)
- [x] [Example of a finished task]
```

### CHANGELOG.md

```markdown
# Changelog

Newest entry at the top. One entry per working session.

## YYYY-MM-DD: [short title of what happened]
- Asked: "[what I asked for, in my words]"
- Done: [what changed]
- Decided: [any decision I made, in my words]
- Files changed: [names]
```

### MISTAKES.md

```markdown
# Mistakes and the rules they taught

Newest at the top. Read before starting work. Apply every rule.

## YYYY-MM-DD: [what went wrong, in a few words]
- What happened: [facts only]
- What it cost or nearly cost: [time, a wrong decision, lost work]
- Rule: [one sentence that would have prevented it]
```

### PREFERENCES.md

```markdown
# How I like to work

- Length: [short answers first, details on request / full detail]
- Tone: [plain and direct / friendly / formal]
- Explaining: [define technical words the first time / skip basics]
- Learning: [give hints before answers / just give the answer]
- Asking: [ask before acting on anything I would not expect / just go ahead on small things]
- Format: [bullets / short paragraphs / tables for comparisons]
- Never: [anything you never want the AI to do]
```