Second Brain

A persistent knowledge vault that gives AI agents full context across conversations.

Second Brain turns Claude Code into a long-term memory system. It continuously ingests your meetings, emails, and Slack messages, processes them into structured intelligence files, and connects everything through a graph of wiki-links. The result: Claude always knows what you discussed, decided, and need to do next.

---

How It Works

You work normally                    Three agents run in the background
  Meetings (Granola)  ──┐
  Emails (Gmail)      ──┼──▶  Intelligence Sync (hourly)
  Slack messages      ──┘         │
                                  ▼
                           Processed intelligence files
                           (tagged, linked, cross-referenced)
                                  │
                                  ▼
                           Daily Wrap-up (5 PM)
                           (synthesized digest by project)
                                  │
                                  ▼
                           Weekly Sync (5:25 PM)
                           (open items → weekly plan)

The vault lives as a folder of markdown files you can browse in Obsidian. Claude reads the vault at the start of every conversation, giving it full context about who you are, what you're working on, and what happened recently.

Vault Structure

your-vault/
├── CLAUDE.md              ← Navigation and operating rules for Claude
├── context/               ← Who you are, your company, strategy, preferences
├── daily/                 ← Synthesized daily digests and weekly plans
├── intelligence/          ← Processed meetings, emails, Slack threads
│   └── _raw/              ← Full raw transcripts
├── projects/              ← Active projects with status, decisions, open questions
├── resources/             ← Templates, frameworks, agent skill files
└── teams/                 ← One file per person you work with

Everything is connected through [[wiki-links]]. When Claude needs to answer "Tell me about Project X," it doesn't just grep -- it follows the link graph from the project file through related intelligence, team profiles, and strategic context.

---

Prerequisites

1. Claude Code installed and working
2. Obsidian installed (for browsing and editing vault files)
3. Obsidian Tasks plugin -- required for the weekly planning queries (## This Week -- Open by Day, ## Forwarded) to render. Install from Obsidian → Settings → Community plugins → Browse. A theme with alternate checkbox states (e.g., Minimal) is recommended so [>], [!], [/] render distinctly.
4. QMD (optional) -- local semantic search over the vault. Install with npm install -g @tobilu/qmd, register the vault (qmd collection add <vault-path> --name <collection> --mask "**/*.md"), then qmd embed. Adds a third search mode alongside flat grep and graph traversal -- useful for conceptual / cross-cutting queries and for duplicate-coverage checks during intelligence sync. See vault/resources/qmd-reindex-skill.md for an optional hourly re-indexing task.
5. MCP servers connected in Claude Code:
   • Gmail -- for email intelligence
   • Granola -- for meeting transcripts
   • Slack -- for Slack messages and threads
   • Google Calendar (optional) -- for schedule-aware context

---

Quick Start

1. Clone the repo

git clone https://github.com/YOUR_USERNAME/second-brain.git
cd second-brain

2. Copy the vault template to your preferred location

cp -r vault/ ~/path/to/your/vault

Pick a location that syncs across devices (Google Drive, iCloud, Dropbox) so you can browse in Obsidian from anywhere.

3. Run the interactive setup

Open Claude Code in the vault directory and run the setup skill:

cd ~/path/to/your/vault
claude

Then tell Claude:

Read and follow the instructions in resources/setup-skill.md to set up my Second Brain.

The setup skill will:

• Ask you about your role, team, projects, and stakeholders
• Generate your personalized CLAUDE.md
• Create context files, team profiles, and project files
• Set up the three automated agents (intelligence sync, daily wrap-up, weekly sync)
• Do an initial backfill of your recent meetings and emails
• Connect the wiki-link graph

The whole process takes about 15-20 minutes.

4. Open in Obsidian

Open your vault folder in Obsidian. Enable the Graph View to see how your files are connected.

5. Verify the agents

The setup will have created three scheduled tasks in Claude Code:

• intelligence-sync -- runs every hour, pulls new meetings/emails/Slack
• daily-wrapup -- runs at 5 PM weekdays, synthesizes the day
• weekly-sync -- runs twice daily (9:25 AM + 5:25 PM, every day). On Sundays, also pre-creates next week's weekly planning file so it's ready Monday morning.

Check they're running: in Claude Code, run /tasks to see scheduled tasks.

---

Daily Usage

Once set up, the vault is mostly self-maintaining:

• Monday morning: Open the weekly file (daily/YYYY-Www.md) -- it was pre-created Sunday night with carry-forward items from last week, grouped by topic. Distribute items into day plans in the manual section.
• During the week: Just work normally. Attend meetings, send emails, chat on Slack. The intelligence sync picks everything up and drops new open items into the auto section of the weekly file.
• End of day: Read the daily digest (daily/YYYY-MM-DD.md) for a synthesized overview. In the weekly file, use the ## This Week -- Open by Day query at the top as a live dashboard, and mark items:
  • [x] done
  • [>] forwarded (pushed to a later day or next week -- stays visible in the ## Forwarded pile)
  • [-] cancelled
  • [!] blocked
• Anytime: Ask Claude questions with full context: "What did we decide about X?", "Prepare me for my meeting with Y", "What's the status of Project Z?"

Asking Questions

Claude uses two search strategies depending on your question:

• Flat search (grep) -- for narrow lookups: "What did Alice say about pricing on Thursday?"
• Graph traversal -- for broad context: "Tell me everything about the customer acceleration project" or "Prepare me for a meeting with Bob"

Graph traversal follows wiki-links across files, reading first-degree and second-degree connections to build a complete picture.

---

Customization

Adding a new intelligence source

Edit resources/intelligence-sync-skill.md to add new data sources. The processing format and file naming conventions are documented in CLAUDE.md.

Changing agent schedules

In Claude Code, use the scheduled tasks tools to update cron expressions:

• Every 30 minutes: */30 * * * *
• Every 2 hours during business hours: 0 9,11,13,15,17 * * 1-5

Adding context files

Create new files in context/ for any slow-changing background information Claude should know: company strategy, competitive landscape, org structure, etc. The setup skill creates initial ones, but you can add more anytime.

Writing preferences

Add communication rules to context/writing-preferences.md. Examples:

• "Never use em dashes"
• "Always use bullet points over paragraphs"
• "Match the tone of the recipient"

---

Architecture

See docs/architecture.md for a deep dive into:

• The three-agent processing pipeline
• File naming and frontmatter conventions
• Graph traversal procedure
• Propagation rules (how intelligence updates flow to projects, teams, and context)
• Weekly sync merge logic (preserving user triage)

---

Requirements

• Claude Code with Claude Opus or Sonnet
• MCP servers for your data sources
• ~5-10 minutes/day for weekly planning (optional but recommended)

License

MIT -- see LICENSE.