An opinionated folder structure that gives any LLM instant context on startup. Your second brain — structured for AI, owned by you.
A personal vault that makes any AI assistant immediately useful — not after weeks of training, but from the first session.
One command. A few questions. You get:
- An AI agent that knows who you are, how you work, and what your priorities are
- A planning system that adapts to your week (Monday kickoff, daily focus, Friday review, quarterly rocks)
- A relationship manager that tracks the whole person — not just deals
- Background scripts that keep your vault alive between sessions
- Everything in plain markdown. No databases, no lock-in, no dependencies beyond bash.
This is the context layer. It's not an agent framework — it complements Claude, GPT, Cowork, Claude Code, or any LLM you use. Drop it in, and your AI gets you.
Download the latest release (zip), unzip, and run the installer for your platform:
| Platform | What to do |
|---|---|
| macOS | Double-click Install Brain.command |
| Windows | Double-click Install Windows Brain.bat |
| Linux | Run bash install.sh |
Answer a few questions — your vault is ready in five minutes.
macOS: If blocked, go to System Settings → Privacy & Security → click "Open Anyway"
Windows: Requires Git for Windows (the installer will guide you if it's missing)
Or clone and run directly:
git clone https://github.com/Bermanmt/My-Portable-Brain
cd My-Portable-Brain
bash start.shUpgrading: Run brain update to get the latest features without losing your personal data.
~/Brain/
├── 00-Inbox/ ← everything lands here first
│ ├── meetings/ ← meeting notes and transcripts
│ ├── emails/ ← email captures
│ └── captures/ ← quick captures, screenshots, links
├── 01-Projects/ ← active work with a finish line
├── 02-Areas/ ← ongoing responsibilities
├── 03-Resources/ ← knowledge base — evergreen reference
├── 04-Archive/ ← completed or abandoned work
├── 05-Meta/ ← how the vault works
├── 06-Agent/ ← AI agent runtime
│ ├── workspace/
│ │ ├── AGENTS.md ← operating instructions + planning protocol
│ │ ├── SOUL.md ← agent personality and tone
│ │ ├── USER.md ← your profile
│ │ ├── memory.md ← long-term context
│ │ ├── vault-health.md ← auto-generated hygiene score
│ │ └── pending-actions.md ← auto-detected patterns and open loops
│ ├── subagents/
│ │ ├── crm-manager/ ← relationship intelligence
│ │ ├── calendar-agent/ ← schedule context for planning
│ │ └── inbox-processor/ ← routes items to the right places
│ └── cron/jobs/ ← background scripts
├── 07-Systems/ ← CRM, goals, planning cascade
│ ├── CRM/contacts/ ← relationship files
│ └── goals/ ← daily → weekly → quarterly → yearly
└── 08-CoreSystem/ ← your roles, principles, values
Plain markdown files. Works forever.
Your agent adapts its greeting based on the day:
| Day | What happens |
|---|---|
| Monday | Last week's results, suggested Big 3 with reasoning, project pulse with staleness, calendar, people follow-ups |
| Tue–Thu | Big 3 progress, today's schedule, anything needing attention |
| Friday | Week review snapshot, then a planning conversation for next week |
| Quarter end | Big Rocks status table, area check, then Q+1 planning |
| Quarter start | Previous Q final score, new rocks, suggested W1 priorities |
| Ad-hoc | Say "let's do a weekly review" any day — agent loads the right context |
The agent suggests priorities based on carries, deadlines, and project staleness — you approve or override.
Your main agent (the orchestrator) doesn't process raw data. Specialist agents handle their domains and maintain briefing files that the orchestrator reads on startup.
| Agent | What it does |
|---|---|
| CRM Manager | Tracks relationships (not just clients). Auto-updates from daily notes. Surfaces follow-ups, birthdays, open loops. |
| Calendar Agent | Schedule context for planning. Enriches meetings with vault context (contacts, projects). Detects time crunches. |
| Inbox Processor | Routes items to projects, areas, CRM. Flags action items. Relationship-aware. |
Each specialist has:
AGENT.md— instructions and access rulesbriefing.md— short summary, rewritten each run (~300-500 tokens)processing-log.md— append-only audit trail
Auto-update policy: silently update interaction logs, preferences, project links, dates. Only flag for user approval: new contacts, contradicting info, archiving.
Cron jobs keep the vault alive between sessions — no LLM needed:
| Script | What it does | Schedule |
|---|---|---|
crm-scan.sh |
Detects contact mentions in daily notes, updates last-contact dates, refreshes CRM briefing | Nightly |
vault-health.sh |
Scores vault hygiene (stale projects, inbox health, daily streak, dormant contacts) | Weekly |
pattern-check.sh |
Surfaces recurring carries, stale project actions, CRM open loops, unchecked items | Nightly |
daily-briefing.sh |
Creates morning note with dashboard stats | Morning |
rebuild-context.sh |
Regenerates CONTEXT-PACK.md from vault sources | Morning |
This is the "dumb detection" layer (bash, free, runs overnight). The "smart processing" layer (context extraction, preference updates) happens when the LLM reads the briefings next session.
Not a sales CRM — a relationship manager. The same person can be a friend, client, and collaborator:
- Multi-type relationships (friend + client + collaborator on one contact)
- Personal context (birthday, kid's school, gift ideas) alongside professional context (deal stage, project links)
- Auto-updated from daily notes and meetings
- Interaction log with reverse-chronological entries
- Open loops (commitments to people) surfaced in morning briefings
Tier 1 — Lean: Minimal structure. Every folder has a LEARN.md explaining what goes there. Best for getting started.
Tier 2 — Full: Complete vault with all systems pre-built: CRM, goals cascade, scheduled automation, all specialist agents. Best if you want everything on day one.
Both tiers include the agent, planning system, and self-improvement loop.
| Requirement | Notes |
|---|---|
| bash 3.2+ | Comes with macOS and Linux |
| git | Optional — for version-controlling your vault |
| Claude / any LLM | The vault is model-agnostic. Works with any AI that reads markdown. |
| Obsidian | Optional but recommended for vault UI |
No Python. No pip. No databases.
- Interactive onboarding wizard with two tiers
- Planning system: day-type detection, structured greetings, Big 3 cascade
- Specialist agent architecture (CRM, calendar, inbox)
- CRM as relationship manager with auto-updates
- Background scripts (crm-scan, vault-health, pattern-check)
- Vault health scoring and pattern detection
- Quarter kickoff and ad-hoc planning support
- Calendar MCP integration (real calendar data instead of manual input)
- Email agent + email MCP
- Meeting agent + transcript MCP
- Notification/nudge system (desktop alerts for planning, follow-ups, birthdays)
- Linux cron support (currently macOS launchd)
- Model-agnostic agent runner (run specialists on any LLM provider)
- Brain modules (users pick "Relationships", "Schedule", etc. during onboarding)
- House Manager and Financial Manager agents
- Calendar write mode (time blocking, meeting auditing)
- Tested on macOS, Linux, Windows WSL
Issues and discussions welcome. See CONTRIBUTING.md.
This is early — if something doesn't work or could be better, please open an issue.
MIT — do whatever you want with it.
Stop organizing. Start thinking.