Skip to content

Repository files navigation

Second Brain

This is my spin on LLM-Wiki by Andrej Kaparthy and Reysu's vault structure and workflows.

It'll help you ingest content through a CLI LLM Agent, like Claude Code or Gemini CLI! I'm using Gemini CLI as it has some generous free limts, but using local models with this is completely doable!

Installation & Quick Start

  1. Clone this repository (or download it).
  2. Run the bootstrap script:
    ./bootstrap.sh <path-to-your-new-vault>
    If you don't provide a path, it will create the folders in the current directory (not advised).
  3. Open the vault in Obsidian.
  4. Setup Your Person Note: Create 04 People/Your Name.md.
  5. Customize AI Prompts: Open the files in 01 Updates/ (e.g., 📌 Daily Brief.md, 📰 Daily News.md) and edit the prompt instructions within to better align with your specific preferences and workflow.

Philosophy

This vault serves two main purposes:

  • Second Brain: For ephemeral or chronological data like daily logs (02 Daily/), meetings (03 Meetings/), relationships (04 People/), and active projects (05 Projects/).
  • Wiki: For formal knowledge (06 Wiki/). Here, knowledge is distilled into flat, slug-named files, and every claim is backed by strict footnotes, which you can then ask your Agent to query for important information!

Capabilities

This vault structure empowers you to:

  • Capture Everything: Quickly log raw thoughts, links, and documents into the 00 Inbox/.
  • Maintain Context: Keep track of people, meetings, and projects organically as they evolve.
  • Synthesize Knowledge: Transform raw notes into a rigorous, citable knowledge base in the 06 Wiki/.
  • Automate with AI: Leverage built-in AI skills to ingest data, summarize content, and structure meetings automatically.
  • Obsidian Web Clipper (Optional but Recommended): I highly recommend using the Obsidian Web Clipper extension for your browser. It integrates perfectly with the 00 Inbox/, allowing you to seamlessly clip articles, recipes, or documentation directly into your vault for later processing and summarization.

Directory Structure

<vault>/
├── AGENTS.md          # AI Agent context rules
├── CLAUDE.md          # Instructions for Claude Code
├── GEMINI.md          # Instructions for Gemini CLI
├── daily_update.sh    # Daily automation script
├── 00 Inbox/          # Raw inputs to be processed
├── 01 Updates/        # Daily research or topic updates
├── 02 Daily/          # Daily notes (The heartbeat of your session)
├── 03 Meetings/       # Notes and transcripts from calls
├── 04 People/         # Rich snapshots of your relationships
├── 05 Projects/       # Active, multi-step goals
├── 06 Wiki/           # The core knowledge base
│   ├── index.md       # Content catalog
│   ├── log.md         # Operation log
│   ├── overview.md    # Evolving synthesis
│   └── pages/         # Slug-named wiki pages
├── 07 Summaries/      # AI-distilled content
├── _Assets/           # Images, PDFs, etc.
├── _Bases/            # Database views (Bases plugin)
├── _Random/           # Sub-vaults or specific context areas
├── _Templates/        # Reusable markdown templates
└── Archive/           # Completed projects

AI Configuration & Skills

This repository includes GEMINI.md, CLAUDE.md, and AGENTS.md files that instruct LLMs (like Gemini CLI or Claude Code) on how to interact with this specific vault structure.

It also includes specialized skills in .gemini/skills/ and .claude/skills/. When you run bootstrap.sh, these files are copied into your new vault.

Included AI Skills

  • defuddle: Extracts clean markdown content from web pages using Defuddle CLI, removing clutter to save tokens and focus on the main content.
  • json-canvas: Creates and edits JSON Canvas files (.canvas) for visual mind maps and flowcharts.
  • obsidian-bases: Manages .base files to create database-like views, filters, and summaries of notes.
  • obsidian-cli: Interacts with the vault using the Obsidian CLI to read, search, and manage notes and properties from the command line.
  • obsidian-markdown: Native support for creating and editing Obsidian Flavored Markdown, including wikilinks, callouts, properties, and embeds.
  • summarize: Summarizes any external content (videos, articles, books) into rich notes with section breakdowns and wikilinks. Output goes to 07 Summaries/.
  • summarize-call: Transcribes and summarizes call recordings, generating dedicated notes for the meeting and participants.
  • vault-update: Automates daily updates for the vault, including reading and updating daily briefs and news notes.
  • wiki-audit: Fact-checks single wiki pages against their cited sources, ensuring accuracy and surfacing uncited claims.
  • wiki-ingest: Processes and adds new source documents (papers, articles, URLs) into the wiki structure.
  • wiki-lint: Audits the wiki for health issues, such as contradictions, broken cross-references, and coverage gaps.
  • wiki-query: Answers questions based specifically on the content ingested into the personal wiki, rather than general LLM knowledge.
  • wiki-update: Safely revises existing wiki pages when new information contradicts or updates current knowledge.

Automation & Scheduling

You can automate repetitive tasks like the "Daily Brief" using Claude Code or Gemini CLI.

Claude Code (/loop and /schedule)

Claude Code has powerful built-in automation commands:

  • /loop <prompt>: Runs a prompt in a continuous loop until you stop it.
  • /schedule "<cron-expression>" <prompt>: Schedules a prompt to run automatically at specific intervals.

Example for a daily brief:

/schedule "0 9 * * *" "Run the vault-update skill to generate today's Daily Brief"

Cron Jobs (Manual Automation)

Alternatively, you can use the provided daily_update.sh script to run updates via a system-level cron job.

  1. Edit daily_update.sh to point to your vault and specify your LLM agent command.
  2. Add it to your crontab:
    crontab -e
    # Run every day at 8:00 AM
    0 9 * * * /path/to/your/vault/daily_update.sh

References & Inspiration

This setup is heavily inspired by and builds upon the ideas and workflows shared in the following resources:

Contributing

Contributions are welcome! If you have suggestions for new skills, folder structures, or improvements to the AI instructions, please open an issue or submit a pull request.

License

MIT

About

No description, website, or topics provided.

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages