The memory note template for your AI assistant
Here's the memory note template you asked for. Save it as a small text file in a memory/ folder, one file per fact.
---
name: client-acme-deadline
type: project
---
[The fact, in one or two sentences, with an absolute date.]
**Why:** [the reason it matters]
**How to apply:** [what the assistant should do differently because of it]
And the index file, memory/INDEX.md, with one line per note:
- [About me](about-me.md) — role, languages, answer style
- [Acme deadline](client-acme-deadline.md) — live before 15 Mar 2027 (trade fair)
- [No morning routines](feedback-no-mornings.md) — I work nights
Then add one line to your rules file: "Read memory/INDEX.md first. Before saving a new note, check whether one already covers it and update that instead."
Below: how to fill the template, the four kinds of notes, and what to keep out.
Want the whole system?
Build Your Own AI Assistant System · $5 — about a coffee
The rules file, memory, sub-agents and automations — with copy-paste templates. Bonus: the Note tool (Windows) that makes your AI fix only what you point at.
Why one fact per note
Built-in "memory" features save whatever the tool decides is important. Your own folder saves what you decide matters, in a form you can read, fix and delete.
One fact per file keeps each note short and easy to update. The index lets the assistant see everything at a glance and open only the notes it needs, instead of reading a giant document every time.
The three lines that make a note work
- The fact. Short and specific. "Acme's new site must be live before 15 March 2027" beats "Acme is in a hurry."
- Why. Without the reason, the assistant can't judge edge cases. With it, it can.
- How to apply. The change in behavior you want. This is the line that turns a stored fact into a better answer.
Four kinds of notes
| Type | What it holds | Example |
|---|---|---|
| About me | Who you are, how you like to work | "I prefer short answers. I'm a beginner at spreadsheets." |
| Feedback | Corrections to the assistant's behavior | "Don't suggest morning routines; I work nights." |
| Project | Facts about ongoing work | "Acme deadline is 15 March 2027 because of the trade fair." |
| Reference | Where things live | "Invoices are in Drive › Finance › 2027." |
Put the type in the header (type: feedback) so you can sort and review notes later.
Feedback notes are the most valuable. Every time you correct the assistant, say: "Save that as a feedback note, with the reason." A month later it has learned your taste, and you've stopped repeating yourself.
What not to save
- Anything already written in your project files. Memory is for what isn't written anywhere else.
- Secrets. Passwords, card numbers, ID numbers, health details you wouldn't want sitting in a file.
- Things that only matter today. Memory is for next month, not this afternoon.
Common mistakes
- Relative dates. "Next Friday" is meaningless in three weeks. Write "Friday 20 November 2026".
- Several facts in one note. When one fact changes, the whole note goes stale. Split it.
- Skipping the index. Without it the assistant has to open every file, or it forgets some exist.
- Saving duplicates. That's why the rule above says to check for an existing note first.
Keep it fresh
Once a month, ask: "Read every note in memory/. Merge duplicates and list anything that looks outdated. Don't change anything without asking me." Ten minutes keeps the whole system trustworthy.
Test it
Create two notes and the index, open a fresh chat, and ask: "What do you know about me and my current projects?" The answer should come from your notes. If it doesn't, check that the rules file tells the assistant to read the index first.
Memory is one of five parts of the system. The rules file, project folders, helpers and a scheduled morning brief are in the full guide, with every template ready to copy.
Want the whole system?
Build Your Own AI Assistant System · $5 — about a coffee
The rules file, memory, sub-agents and automations — with copy-paste templates. Bonus: the Note tool (Windows) that makes your AI fix only what you point at.