Cynobi AI Academy
Course 1 · AI Foundations · Lesson 07

Memory, Projects & System Prompts

You've typed the same introduction fifty times. Three mechanisms end the ritual — and all three turn out to be the same trick: text loaded onto the desk before you type.

Free · Course 1 ~5 min video Type · Practical shift

The one mental model

Every new chat starts with an empty desk (Lesson 04) — by design. The system prompt, memory, and projects are all the same trick: text loaded onto the desk before you type. No magic, no overnight learning, no secret brain. The machine didn't change — the desk just stopped starting empty.

Key terms

System prompt
Hidden stage directions the app loads before your first word — tone, rules, persona. It speaks first in every chat.
Custom instructions
Your piece of the system prompt: standing orders (who you are, how you like answers, what to avoid), written once in settings, loaded into every new chat automatically.
Memory
A notebook about you. Durable facts get saved as notes and pasted back onto the desk in each new chat. You can read, edit, or wipe it — and you should skim it occasionally.
Project
A room for a job: your files plus instructions, set up once. Every chat inside starts pre-briefed. ChatGPT and Claude both offer Projects by that name.

The misconception to drop

✕
“When it remembers me, the model learned about me overnight — it's retraining itself on my data.”
✓
Nothing retrains. Your notes and files just ride along on the desk, chat after chat. That's also why you stay in control: delete the note, empty the room — gone for good.

Put it to work — your setup, once, this week

1
Write your custom instructions — who you are, who your work is for, the tone, the format.
2
Tell it what to remember — and skim the notebook monthly, deleting what's stale.
3
Build one project for your most repetitive job, files included.
4
For anything critical, still paste it back in — memory is a convenience, not a guarantee.

Ask the AI Tutor

Pause the video and ask anything from this lesson — the tutor answers from this lesson’s material.

What's the difference between memory and a project? What should I put in custom instructions? Is the model learning about me over time? How do I see what it remembered?