Facilitator console · 3-hour build session

Your AI is not forgetful. You built it with nowhere to remember.

I ran this on nanoGPT. Graphify found 73 nodes across 76 edges, grouped them into 19 communities, and put the GPT class in the middle where the code said it belonged.

Run the room See what they will own

Before people arrive, confirm the laptop has terminal access and Obsidian is installed. They also need 10–30 files they can safely test.

From rented context to owned memory

Raw
folderGraphifyObsidian
vaultAI
hands

CaptureMapAskWrite back

Stop renting memory one prompt at a time.

Build the place it lives.

Graphify exposes the structure, and Obsidian keeps it local. The agent works inside what you own.

Explain the problem before the tools

These people are new to agents. Make the reset real before you ask them to install anything.

Start with the moment they know

You explained the project yesterday. Today a fresh session asks what you are working on… so you paste the same background again. Beginners understand that immediately.

An agent is AI that can use tools. It can inspect files and take actions.
A new session cannot recover yesterday's decision unless that decision was saved somewhere it can reach.

Do not lead with Graphify.

Lead with the reset. Then explain that the map and vault give the agent a place to return to.

Exercise · 10 minutes

Make them feel the reset

Ask for the last time they had to explain the same work twice. Then choose the folder their agent keeps rereading.

What did you have to explain to AI more than once?

Pull the system apart

Before anybody installs a package, show them where the memory lives and who controls it.

01Raw folderMessy source material

02GraphifyMaps nodes and edges

03Knowledge graphFinds communities

04Obsidian vaultMakes links navigable

05AI agent skillsRead, write, maintain

06Memory loopRebuilds over time

Node

A concept, file, function, topic, person, or idea.

Edge

A named relationship between two nodes.

Community

A cluster of closely related nodes.

God node

A central concept that many other ideas connect through.

Extracted

Found directly in source files.

Inferred

Reasoned from available evidence.

Ambiguous

Uncertain and queued for review.

Choose the host, then install

Ask where the agent runs before anybody copies a command. A Mac mini is one machine. A VPS is a split setup with Obsidian on the participant's desktop.

Route A · Mac mini

Agent and vault on one machine

Install the agent, Python, Graphify, and Obsidian on the Mac mini. A remote Mac mini still needs Screen Sharing for Obsidian.

Mac mini · install agent

curl -fsSL https://claude.ai/install.sh | bash
export PATH="$HOME/.local/bin:$PATH"
claude --version

Checkpoint… claude --version works and Obsidian reaches the vault screen on that Mac.

Route B · VPS

Agent remote, vault local

Install the agent, Python, and Graphify on the VPS. Keep Obsidian on the participant's Mac or laptop. The exported vault comes back over SCP, Git, or an approved sync method.

Local · connect to VPS

ssh <username>@<server-ip>

VPS · install and verify

sudo apt update
sudo apt install -y curl git python3 python3-pip python3-venv pipx
curl -fsSL https://claude.ai/install.sh | bash
export PATH="$HOME/.local/bin:$PATH"
claude --version
python3 --version
python3 -m pip --version
git --version

Local · copy exported vault

# Run locally after Graphify exports the vault
scp -r <username>@<server-ip>:~/ai-second-brain-demo/<exported-vault-folder> ~/Documents/

Checkpoint… SSH reconnects, every version command works, and two linked notes open after the vault is copied locally.

Ask every person where their agent runs.

Write “Mac mini” or “VPS” beside their name. Do not let the room discover the split architecture during the export.

Before Graphify

Get Python 3 working

People cannot install Graphify until Python works. Send them to the official download page. Mac users run the macOS package. Windows users run the Windows installer and select “Add python.exe to PATH” if the option appears.

Open the official Python download ↗

Verify Python · Mac

python3 --version
python3 -m pip --version

Verify Python · Windows

py --version
py -m pip --version

Do not move on until Python and pip both return version numbers.

Before the export

Get Obsidian open

Send participants to the official Obsidian download page. Mac users move the app into Applications. Windows users run the installer. Linux users choose the package made for their distribution. VPS users do this on their desktop, never on the headless server.

Open the official Obsidian download ↗

The checkpoint is simple… every participant can open Obsidian and reach the vault screen.

Check the official Claude Code setup page before class ↗

Do not teach from memory. Package spelling can change. Verify the current Graphify installation command before the room opens.

Terminal · install

pip install Graphifi
Graphifi install

# If pip permissions fail on macOS
pipx install Graphifi

# Confirm the current command
graphify --help

pip not found

Install Python and pip, then restart the terminal.

Permission error

Use pipx so the package installs outside the system Python.

Command not found

Restart the terminal and verify the Python scripts directory is on PATH.

Make the machine show its work

Start with 10–30 safe files. That is enough to expose structure without turning the room into a loading screen.

Terminal · prepare folder

mkdir ai-second-brain-demo
cd ai-second-brain-demo
mkdir raw

Agent · run graph

cd path/to/ai-second-brain-demo

# Inside your compatible AI coding agent
/graphify .

ai-second-brain-demo/

raw/

meeting-notes.md

product-ideas.md

research-summary.md

architecture-notes.md

screenshot.png

sample-paper.pdf

First-run receipts

Number of nodesNumber of edgesNumber of communitiesTop god nodeMost surprising connectionMost ambiguous connection

Keep private material out of the workshop. Documents may take longer when they need AI processing, while code can map faster because its structure is already explicit.

Break15 minutes

Compare the graphs. Help the people who got stuck… then get everybody back in the room.

Put the memory somewhere they own

The export turns the report into local notes. Participants can inspect the links and keep the work after the session ends.

Terminal · export

graphify export Obsidian

01
Locate the export
Wait for the process to finish and note the generated vault folder.
02
Open folder as vault
In Obsidian, choose “Open folder as vault” and select the export.
03
Inspect the graph view
Click through concept notes, properties, wiki links, honesty tags, and the generated canvas.

Success criteria

You can open the exported vault, click between linked concepts, and identify the most connected note.

Give the agent access… not ownership

The graph supplies the structure. Obsidian skills let the agent work inside a vault the participant controls.

Agent · install skills

/plugin marketplace add Kapono/Obsidian skills
/plugin install install

Verify before teaching. The source transcript shows inconsistent repository spelling. Confirm the current marketplace command before the workshop.

Obsidian markdownWiki linksBasesCanvasCLIClean web capture

Prompt · first write-back

Read the graph report. Build one permanent Obsidian note that links every god node. Separate what the files prove from what the graph inferred. Put uncertainty in an Ambiguous section. If the evidence is not there, say so. Do not make the graph sound smarter than it is.

Expected permanent note

Overview summary
Links to god nodes
Extracted relationships

Inferred relationships
Ambiguous review queue
Recommended next questions

Make the work survive the chat

One graph is a demo. Rebuilding after useful work is what turns it into owned memory.

01Capture

02Map

03Ask

04Write back

05Rebuild

06Repeat

Daily

Capture a source

Terminal · capture URL

graphify add <url>

After useful analysis

Write insight back

Save a permanent note with wiki links. Label connections extracted, inferred, or ambiguous.

Weekly

Review the graph

Rebuild, inspect new god nodes, review ambiguous links, save a weekly summary, and back up the vault.

Monthly

Audit the system

Find emerging themes, duplicates, outdated notes, evidence gaps, and useful outputs the graph can support.

Field notes

Prompts that make the graph prove itself

Do not ask for vibes. Ask where the evidence came from… and make useful answers permanent.

Prompt · Find the center

Identify the god nodes in this graph. For each one, explain why it is central, what communities it connects, and what I should do with it next.

Prompt · Find surprising connections

List the most surprising connections in this graph. Explain why each matters and whether it is extracted, inferred, or ambiguous.

Prompt · Review uncertainty

Find all ambiguous connections. Explain what evidence would confirm, reject, or preserve each one as ambiguous.

Prompt · Trace a path

Show me the shortest path between [[Concept A]] and [[Concept B]]. Include every intermediate node and label each relationship.

Prompt · Write the weekly update

Review the latest graph changes and write a weekly Obsidian note with new nodes, changed god nodes, new communities, surprising connections, and ambiguous links needing review.

Prove it for seven days

The workshop is not the win. A system that still works next week is the win.

Day 1

Build small

Use 10–30 files. Run the graph, export it, and write the first summary note.

Day 2

Add sources

Add 5–10 sources and clip one useful webpage into clean markdown.

Day 3

Ask better

Find god nodes, surprising connections, and ambiguity.

Day 4

Write back

Save one insight note, link concepts, and rebuild.

Day 5

Clean

Review duplicates, vague names, and missing context.

Day 6

Apply

Create one article, memo, product idea, report, or code explanation.

Day 7

Review

Ask what changed this week and what the system needs next.

What they leave owning

NOT RENTED. BUILT.

A mapped knowledge graph
An Obsidian vault with linked concepts
Honesty-tagged relationships
An AI-readable, writable memory system
A weekly maintenance ritual

Your AI is not forgetful. You built it with nowhere to remember.

Build the place it lives.

Explain the problem before the tools

Start with the moment they know

Make them feel the reset

Pull the system apart

Node

Edge

Community

God node

Choose the host, then install

Agent and vault on one machine

Agent remote, vault local

Get Python 3 working

Get Obsidian open

pip not found

Permission error

Command not found

Make the machine show its work

First-run receipts

Put the memory somewhere they own

Locate the export

Open folder as vault

Inspect the graph view

Give the agent access… not ownership

Make the work survive the chat

Capture a source

Write insight back

Review the graph

Audit the system

Prompts that make the graph prove itself

Prove it for seven days

Build small

Add sources

Ask better

Write back

Clean

Apply

Review

NOT RENTED. BUILT.