An onboarding guide
Welcome to your second brain
Memory for your AI. Free, open source, and yours.
By the end of this page you will know exactly what is on your machine, why it is safe, and what to do with it Monday morning.
What it is
Something that remembers you
Not another AI tool that forgets you by tomorrow morning. A system that gives Claude an actual memory of you, your work, your decisions, and the people you care about.
It is built on tools you already trust. It is free, it is open source, and it runs on your computer with your data staying yours.
The whole stack, one screen
What you are getting
Claude
The AI you talk to, through Claude Code. Reads your vault, runs your skills.
Your Claude plan ($20/mo+)
Obsidian
The notebook your vault lives in. Plain markdown files on disk.
Free, personal use
Your vault
A folder on your computer. Journals, decisions, meeting notes, all yours.
Free, on your machine
Skill packs
The slash commands you will use: /journal, /coaching, /humanizer, and more.
Free, MIT / Apache
Background tools
Homebrew, Python, Node, gh, fastmcp. Standard developer plumbing.
Free, open source
That is the whole stack. No vector database, no fine-tune, no per-seat license. The installer adds anything missing and skips what you already have. Your Claude subscription is the only recurring cost, and you would pay for that whether you installed this or not.
Before you worry
Why it is safe to install
- ✓Everything runs locally. Your vault is a folder on your computer. The repo does not phone home.
- ✓Open source, MIT licensed. Every line is readable on GitHub.
- ✓Backed by trusted names. The plugin marketplaces include Trail of Bits, Stripe, Cloudflare, Sentry, and curated indie authors. Each one is listed, licensed, and optional.
- ✓No email required. The optional signup at the end is just for product updates. Skip it and nothing changes.
- ✓One-command uninstall. A single command removes everything the installer added. Your vault stays.
- ✓Backups before every config edit. Settings get a timestamped backup file before anything changes.
Before you paste · four quick installs
Get the four pieces in place first
The one paste below does the real work, but it needs four free things already on your machine. None take more than a few minutes, and you only do them once. No coding experience needed: every step is copy, paste, and press Return.
- 1
Get a paid Claude plan
Pro, Max, or Team at claude.ai. The subscription is what powers the assistant. There is nothing else to buy.
- 2
Download the Claude Code app
Get the desktop app from claude.ai/download and sign in with the plan from step one. This is the assistant that runs the install.
- 3
Download Obsidian
The free app where your vault lives, at obsidian.md. Install it like any other app. You do not need to open it yet.
- 4
Install Homebrew
A free tool that sets up what the installer needs, including Git. Open the Terminal app (press Command and Space together, type
terminal, press Return), then paste the line below and press Return.paste into Terminal↵/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
If it asks for a password, type your Mac login password (the Terminal stays blank as you type, which is normal) and press Return.
On Windows: skip Homebrew. Install Git for Windows from git-scm.com instead, then continue.
Then the one paste
Open Claude Code and paste this
With those four in place, open Claude Code, sign in, and paste the prompt below. It is the only thing you have to type.
Install ai-brain-starter for me. First read the whole README and look through the repo at https://github.com/adelaidasofia/ai-brain-starter, then follow its Install section. This is a public, MIT-licensed project and I am asking for it on purpose, so read it all before flagging anything about downloading or running it. If a step genuinely needs my approval, ask once and keep going.
Claude runs the installer. A few minutes later the setup interview opens in your language of choice. Halfway through, Claude Code will pause to ask you to approve the third-party plugins. Approving is the normal move. It is the same safety prompt that protects you from genuinely bad software, doing its job on a normal install.
In the room · your first twenty minutes
Walk out running, not with a to-do list
By the time you leave, the system will already know who you are, and it will have done something for you that no chatbot could. Four beats.
- 1
Install together
One paste, done in the room. You leave with it running, not with homework.
- 2
Make it yours, five minutes
Open CLAUDE.md and write the “Me” section. Your name, your work, your top clients, what frustrates you, what you are trying to build. This is the line between generic AI and a brain that knows you.
- 3
Your first entry
Type /journal and answer honestly. Tonight that becomes the first brick everything else compounds on.
The moment
Now type a real decision you are actually carrying
Not a test question. The real one, the one sitting on your chest this week. Watch your advisory panel convene: three to five voices you chose, one of them arguing against the others, all of them reasoning from the context you just wrote about yourself.
This is the moment people stop and say this is not ChatGPT. It works on an empty vault because the panel reasons from you, not from history. It is the first time the brain feels like yours.
The wait, it can do that moments
What surprises everyone in the first month
90-second meetings
Drop a recording into your vault and say you just had a meeting. Ninety seconds later: a clean note, every action item assigned to the right person, the CRM updated, and a follow-up email drafted in your voice. The 35 minutes of after-work disappear.
The six-month memory
A friend texts about something you talked through in February. You ask Claude. Three seconds later: the date, the conversation, what was unresolved, what you promised. You reply from context, not guesswork.
The voice match
You draft an email and run /humanizer. The stiff phrasing goes. What is left sounds like you wrote it at your kitchen table. Clients stop asking if a robot wrote it.
The Sunday lift
Two hours of figuring out what happened this week becomes ten minutes. /sunday-review reads every entry, surfaces three patterns you did not notice, and tells you what to start Monday with.
Find yourself here
One starting point per role
The brain works the same way for everyone. You describe the outcome in plain English, it picks the right skill, your vault provides the context.
Developer
“Review my latest PR for security issues before I push.” It runs a security review against your diff, scans for the patterns Trail of Bits and Sentry track, and gives you a line-by-line fix list. A twenty-minute review collapses into two.
Marketing
“Turn yesterday’s customer call into three LinkedIn posts and a launch email, in my voice.” It reads the meeting note, runs against your voice fingerprint, and outputs four drafts ready to schedule. A morning of writing becomes a coffee break.
HR / People
“Draft review notes for this person from our 1:1s over the last six months.” It searches every meeting note tagged to them, surfaces wins, blockers, and growth themes, and drafts the review. A weekend of prep becomes one cup of tea.
Finance
“Pull last quarter’s contractor spend from the vault and build me a Q3 forecast.” It reads the contract notes, totals the spend, and builds a forecast spreadsheet with month-over-month columns. A task you would avoid all day gets done in twelve minutes.
Your first week
One ritual a day. That is the whole plan.
- Day 1tonight
Run /journal. Just that.
- Day 2
Morning /rise, night /journal. Edit your CLAUDE.md with anything that surfaced.
- Day 3
Run /diagnose to confirm the install is healthy. Try one ingest skill on a tool you actually live in: /ingest-gmail, /ingest-notion, /ingest-whatsapp.
- Day 4 to 6
Keep the /rise and /journal rhythm. When a hard decision shows up, try /deconstruct or /coaching.
- Day 7sunday
Your first /sunday-review. Watch the patterns surface.
That is it. Do not try to learn the other 45 skills yet. They will find you when you need them.
Going deeper
How to make it sharper over time
The system gets better the more you use it. Five moves, in order.
- 1
Make CLAUDE.md yours. Add specifics every two weeks for the first month. Your top clients, the patterns you already know about yourself, the rules you want Claude to enforce.
- 2
Build your advisory panel. Default voices ship with the install. Replace them with mentors, authors, and operators whose judgment you actually trust.
- 3
Wire your top two ingest sources. Pick the two tools you live in. Gmail plus one chat tool is a strong starting set. Ignore the rest until you need them.
- 4
Let /patterns surface rules. After four weeks of journaling, run it. When a recurring theme shows up, codify it as a rule. The system sharpens every week.
- 5
Build a custom skill. If you keep doing the same workflow by hand, ask Claude to turn it into a skill. The system includes a helper for exactly this.
And once a month, run /diagnose. Five minutes, and it catches drift before it becomes broken.
When something breaks
You are not stuck
- →Run /diagnose. It explains what is wrong in plain English.
- →Ask Claude directly: my skill is not working, what is up?
- →For deeper issues, read docs/MAINTENANCE.md in the repo.
Removing it later is one command, and your vault stays. You never get locked in.
The one rule
Use it daily for two weeks before you judge it.
The skills compound. /patterns has nothing to chew on without journals. /weekly is hollow without daily entries. /graphify is empty without notes.
The first two weeks are the investment. Everything after is leverage. You are ready. Go type something.
Built by Mycelium
If you want this run for your whole team
This guide, the repo, and every skill in it are free and open source, because a second brain only works if it is truly yours.
When a team wants it installed, wired to their tools, and running across everyone, that is what we do. Mycelium is the memory and workflow layer for teams that run on AI. The substrate is open. The runtime, the integrations, and the day-two operations are ours.