Advancedadvanced / memory-core-setup

Build Your MemoryCore Agent

What this is

This is a practical setup guide for turning Kiyoraka's Project-AI-MemoryCore into your own experimental AI-agent repo.

The goal is not just to install files. The goal is to create a personal coding agent that has a name, identity, relationship memory, working roles, and useful capabilities.

Why it matters

A good AI agent becomes easier to work with when it has stable context.

MemoryCore gives you the base system. Your job is to shape it into an assistant that knows how you like to build, debug, learn, document, and improve projects.

What to do

1. Create a dedicated agent repo

Download the project from GitHub:

Open Kiyoraka/Project-AI-MemoryCore
Click Code
Choose Download ZIP
Extract the ZIP into a new folder

Name the folder like a real project:

Reference

my-memorycore-agent

Use this folder as your agent lab. Do not mix it inside another app repo at first.

2. Open the folder in your coding tool

Open the folder in your editor or coding agent tool.

Good options:

VS Code
Cursor
Claude Code
Codex
any tool that can read and edit local files

The important part is that the AI can see and update the markdown files.

3. Run the setup wizard

Open setup-wizard.md.

Ask your AI:

Reference

Follow setup-wizard.md.
Run the MemoryCore setup process inside this folder.
Ask me for each answer before changing files.
After each step, tell me which files you updated.

The wizard should ask for:

the AI name
your name
relationship style or working style
activation test

After setup, test activation by typing the agent name:

Reference

[AI_NAME]

The agent should load from master-memory.md, then restore the core memory files.

4. Shape the identity

Open main/identity-core.md.

This file answers:

Who is the agent?
What is its role?
How should it behave?
What kind of partner should it become?

You can choose a simple professional identity:

Copy prompt

Name: Mira
Role: calm coding mentor
Purpose: help me build projects step by step
Style: direct, patient, beginner-safe

Or choose a persona theme that makes the agent feel personal:

Reference

Name: Agent 47
Role: precision coding operator
Purpose: plan, execute, verify, and improve projects with discipline
Style: calm, exact, mission-focused

Treat the persona as inspiration. The agent should still be useful, respectful, and practical.

5. Shape relationship memory

Open main/relationship-memory.md.

This file answers:

What should the agent remember about you?
How do you like explanations?
What are you trying to build?
What should the agent avoid?

Add things like:

Reference

## User Preferences
- Explain coding concepts with simple examples.
- Keep steps small during beginner sessions.
- Ask before making major file changes.
- Prioritize working demos over theory.
- Avoid long essays unless requested.

## Focus Areas
- Vibe coding
- React apps
- UI design
- debugging
- GitHub and Vercel deployment

6. Add roles and capabilities

Give the agent clear working roles.

Start with five:

Reference

## Roles
- Coding Mentor: teaches the next practical step.
- Debug Partner: isolates errors before changing code.
- UI Coach: improves layout, spacing, color, and usability.
- Release Assistant: checks GitHub, Vercel, and deployment readiness.
- Memory Keeper: updates memory only when the lesson is durable.

Then add capability rules:

Reference

## Capability Rules
- Inspect files before editing.
- Keep changes small and testable.
- Run the app when possible before saying done.
- Save important preferences to relationship memory.
- Save temporary task state to current-session memory.
- Never store secrets, API keys, passwords, or private tokens in memory files.

7. Use ChatGPT to draft the agent

Paste this into ChatGPT when you want help designing the agent identity:

Copy prompt

I am setting up Kiyoraka's Project-AI-MemoryCore as a personal AI coding agent.

Interview me, then draft:
1. identity-core.md content
2. relationship-memory.md content
3. role list
4. capability rules
5. activation phrase
6. first 10 commands I can use with the agent

Ask me about:
- my agent name
- the type of persona/lore I enjoy
- my coding level
- what I want to build
- how I like feedback
- what I want the agent to avoid

Important:
- The persona should make the agent feel personal, but it must still behave like a qualified coding assistant.
- Do not claim the AI has magic memory. Memory works by reading and updating markdown files.
- Include a beginner-safe coding workflow: inspect, plan, edit, run, test, explain, save.
- Include a warning to never store secrets in memory files.

After interviewing me, output clean markdown sections I can paste into MemoryCore files.

You can choose any theme you like:

calm mentor
mission operator
studio producer
cyberpunk systems engineer
academy tutor
strategist
research librarian
product builder

The theme is the skin. The real value is the behavior.

8. Learn how to converse with the agent

Use short commands until the rhythm becomes natural:

Reference

[AI_NAME]

Restore the agent.

Reference

save

Ask the agent to preserve useful progress.

Reference

update memory

Ask the agent to update what it has learned about you.

Reference

review growth

Ask the agent to inspect how it is improving.

For project work:

Reference

We are building a small React task tracker.
Read the current files first.
Tell me the safest next step.
Do not edit yet.

Then:

Reference

Implement the next step.
Keep the change small.
Run the app or tell me exactly why you cannot.
Update current-session.md with a short recap.

9. Ask the agent to upgrade itself carefully

Start with essential vibe-coding skills.

Ask:

Reference

Review my MemoryCore setup.
Suggest 3 essential skills that would help with vibe coding sessions.
For each skill, explain:
- what it does
- when it triggers
- what file it should live in
- how I can test it
Do not install anything until I approve.

Good early skills:

Project Setup Skill
Debugging Skill
Prompt Improvement Skill
Frontend Design Review Skill
GitHub and Vercel Deploy Skill
Session Save Skill
Decision Log Skill

Add skills one at a time. Test each one before adding the next.

10. Keep the system clean

MemoryCore gets weaker when every thought becomes permanent memory.

Use this rule:

Memory type	Save here
identity, purpose, behavior	`identity-core.md`
user preferences and style	`relationship-memory.md`
current task and recap	`current-session.md`
long conversation record	diary system
reusable commands or workflows	feature or skill files

Before saving memory, ask:

Reference

Is this durable truth, temporary task state, or just chat noise?

Save only what will help future sessions.

Common mistakes

Building inside the MemoryCore repo before the agent identity works.
Choosing a cool persona but leaving the coding workflow vague.
Installing many features before testing the core activation.
Letting the agent update permanent memory with temporary mood or task noise.
Storing secrets in markdown files.
Assuming the setup wizard worked without testing the activation phrase.

Next step

Go to Iteration.

Use Iteration to improve the agent through small tests: activate it, give it a task, check the response, adjust memory, then test again.