Personal AI Infrastructure (PAI)

Open-source personal AI infrastructure for orchestrating your life and work

Features • Quick Start • Documentation • Examples • Community

---

🚀 Recent Updates

Important

🔥 v0.6.0 MAJOR UPGRADE: Repository completely restructured with .claude/ directory!

BREAKING CHANGE - Repository Structure Changed:

• All PAI infrastructure now lives in .claude/ directory
• Repository now properly mirrors your actual ~/.claude/ working system
• Fixes major compatibility issues reported by users
• Action Required: New installations should copy .claude/ to ~/.claude/

See full changelog below →

📅 Click to see all updates

Recent Manual Updates

• ✨ Oct 19: Session-start hook now loads PAI skill - improved Skills system bootstrap
• ✨ Oct 18: Major repo cleanup - fixed missing files, hooks, settings
• ✨ v0.5.0: Skills-based architecture with 92.5% token reduction

Automated Documentation Updates

📅 2025-10-20 - Settings: 1 updated

Updated by pre-commit hook: 1 modified

📅 2025-10-19 - Voice: 2 updated

Updated by pre-commit hook: 3 modified

📅 2025-10-19 - Skills: 1 updated, 5 removed

Updated by pre-commit hook: 1 modified, 5 deleted

📅 2025-10-19 - Hooks: 2 new, 1 removed, Settings: 1 updated

Updated by pre-commit hook: 3 added, 1 modified, 1 deleted

---

Version History

📅 v0.6.0 - Repository Restructure with .claude/ Directory 🔥 MAJOR UPDATE

The Problem: Users reported issues with PAI not working correctly because the repository structure didn't match the actual working system. The real PAI system expects all infrastructure to live in ~/.claude/, but the repo had everything at root level. This caused confusion and compatibility problems.

The Solution: Complete repository restructure to mirror the actual working system:

• Created .claude/ directory at repository root
• Moved ALL PAI infrastructure into .claude/ (agents, commands, documentation, hooks, skills, voice-server, etc.)
• Kept GitHub infrastructure at root (README, LICENSE, .gitignore, .github, etc.)
• Repository now serves as a true reference implementation

What Changed:

Before (v0.5.0):
/PAI/
├── agents/
├── commands/
├── documentation/
├── hooks/
├── skills/
├── voice-server/
├── settings.json
├── .mcp.json
├── setup.sh
└── README.md

After (v0.6.0):
/PAI/
├── .claude/                 # ← NEW: All PAI infrastructure here
│   ├── agents/
│   ├── commands/
│   ├── documentation/
│   ├── hooks/
│   ├── skills/
│   ├── voice-server/
│   ├── settings.json
│   ├── .mcp.json
│   └── setup.sh
├── README.md               # GitHub infrastructure stays at root
├── LICENSE
└── .gitignore

Why This Matters:

1. Proper Emulation: Repository now accurately represents how PAI works in production
2. Easier Setup: Users can see exactly how their ~/.claude/ directory should be structured
3. Less Confusion: Clear separation between GitHub files and PAI infrastructure
4. Better Documentation: Structure itself serves as documentation
5. Reference Implementation: Can be copied/referenced directly for setup

Migration:

• No action required for existing installations
• New users get the correct structure from the start
• All documentation updated to reflect new paths

Rationale: The PAI system is designed to live in ~/.claude/ on your system. By organizing the repository to mirror this structure, we make it immediately clear how PAI should be set up. This is especially important for new users who are trying to understand the system architecture and for contributors who need to know where files belong.

📅 v0.5.0 - Skills-Based PAI Architecture (92.5% Token Reduction)

Major Architectural Improvement:

• Zero hook overhead - Eliminated all context loading from UserPromptSubmit hook
• 92.5% token reduction - From 4000 tokens/interaction to 300 tokens
• Pure skills architecture - Core identity in skill description (always in system prompt)
• On-demand context - Full context loaded only when explicitly needed

What Changed:

• Added YAML frontmatter to skills/PAI/SKILL.md with comprehensive system description
• Core identity + critical security now in skill description (always present)
• Removed MINIMAL.md entirely (no longer needed)
• Hook renamed to update-tab-titles.ts (only handles tab titles, zero context)
• Flat file structure in skills/PAI/ (no /contexts subdirectory)

Architecture:

• Tier 1 (Always On): Skill description in system prompt (~300 tokens) - identity, critical security, architecture explanation
• Tier 2 (On Demand): SKILL.md body loaded when PAI skill invoked (~4000 tokens) - contacts, preferences, voice IDs, detailed security
• Hook: Only updates tab titles (0 tokens context overhead)

Benefits:

• Cleanest possible architecture - fully embraces Claude Code skills system
• Context always relevant - skill description always present, full context on-demand
• Easy to customize - clear YAML frontmatter with [CUSTOMIZE:] markers
• Scales efficiently - adding content doesn't multiply token costs

Files:

• skills/PAI/SKILL.md - Full context with YAML frontmatter
• skills/PAI/contacts.md - Contact templates
• skills/PAI/preferences.md - Stack preferences templates
• skills/PAI/response-format.md - Response format templates
• skills/PAI/security-detailed.md - Security procedures
• skills/PAI/voice-ids.md - Voice system configuration (optional)
• hooks/update-tab-titles.ts - Tab title updates only

📅 v0.4.0 - Repository Restructure 🔥 BREAKING CHANGE

⚠️ Breaking Changes:

• PAI_DIR environment variable: Change from /path/to/PAI/PAI_DIRECTORY to /path/to/PAI
• Repository renamed: PAI → Personal_AI_Infrastructure

What Changed:

• Moved all PAI_DIRECTORY/ contents to repository root (agents/, skills/, commands/, etc.)
• Repository renamed for clarity and better SEO
• All functional directories now immediately visible on GitHub
• GitHub automatically redirects old URLs to new

Migration:

1. Update PAI_DIR: export PAI_DIR="/path/to/PAI" (remove /PAI_DIRECTORY)
2. Reload shell: source ~/.zshrc
3. Pull latest: git pull
4. Update remote: git remote set-url origin git@github.com:danielmiessler/Personal_AI_Infrastructure.git

📅 v0.3.2 - Fabric Skill with Intelligent Pattern Selection

Fabric skill now intelligently selects the right pattern from 242+ options based on user intent. Complete Fabric repository bundled locally. Categories: Security (15), Summarization (20), Extraction (30+), Analysis (35+), Creation (50+), Improvement (10), Rating (8).

📅 v0.3.1 - Research Skills & API Key Infrastructure

Multi-source research with parallel agent execution. New skills: alex-hormozi-pitch, research. New agents: perplexity-researcher, claude-researcher, gemini-researcher. Added .env.example with API key documentation.

📅 v0.3.0 - Skills System Migration

Migrated to Anthropic's Skills architecture. Modular skill packages with progressive disclosure. Context system → Skills system. See documentation for details.

📅 v0.2.4 - README Cleanup

Collapsed updates section, reduced visual clutter, optimized space.

📅 v0.2.3 - Visibility & Portability

.claude → PAI_DIRECTORY, vendor agnostic, dynamic paths with ${PAI_DIR}, full portability.

📅 v0.2.2 - Voice System

Migrated to macOS native Premium voices (zero cost, offline, private).

📅 v0.2.0 - v0.1.0 - Initial Releases

Public release with voice server, PAI_HOME support, comprehensive documentation, MCP detection, hooks system.

📅 Previous Updates

September 20, 2025

• 🗣️ Added /voice-server with ElevenLabs integration
• 🔧 Fixed hardcoded path issues
• 🪝 Working on missing hooks

September 12, 2025

• 🧠 Dynamic resource loading system
• ⚡ Submit-user-hook for context loading
• 🗺️ Dynamic routing via load-dynamic-requirements

---

🎯 What is PAI?

Core Mission: Augment humans with AI capabilities so they can survive and thrive in a world full of AI.

🌍 The Problem

It doesn't matter how powerful AI becomes if it's not accessible to everyone

Right now, AI is trapped behind corporate APIs, expensive subscriptions, and complex interfaces that only developers can use. Meanwhile, billions of people who could benefit from AI augmentation—artists, teachers, small business owners, researchers, parents—are left behind.

PAI exists to solve this. This project's goal is to give the most powerful AI capabilities, in the form of a complete platform, to everyone on Earth. Not just the tech elite. Not just the wealthy. Everyone.

graph TB
    You[👤 You] --> PAI[🧠 PAI/Kai]
    
    PAI --> Work[💼 Work]
    PAI --> Life[🏠 Life]
    
    Work --> Blog[📝 write-blog]
    Work --> News[📊 get-newsletter-stats]
    Work --> Site[🌐 get-site-analytics]
    Work --> Consult[💼 create-consulting-document]
    
    Life --> Finance[💰 answer-finance-question]
    Life --> Memory[🔍 get-life-log]
    Life --> Alert[🔔 send-text-alert]
    Life --> Health[🏥 track-health-metrics]
    
    Blog --> Output[✨ Augmented Output]
    News --> Output
    Site --> Output
    Consult --> Output
    Finance --> Output
    Memory --> Output
    Alert --> Output
    Health --> Output
    
    style You fill:#24283b,stroke:#7aa2f7,stroke-width:2px,color:#c0caf5
    style PAI fill:#1a1b26,stroke:#bb9af7,stroke-width:3px,color:#c0caf5
    style Work fill:#24283b,stroke:#9ece6a,stroke-width:2px,color:#c0caf5
    style Life fill:#24283b,stroke:#ff9e64,stroke-width:2px,color:#c0caf5
    style Output fill:#1a1b26,stroke:#f7768e,stroke-width:3px,color:#c0caf5

Loading

PAI is an open-source, full personal AI platform that's completely agnostic to who you are and what you're trying to accomplish.

Think of PAI as your personal AI that actually knows you—all your projects, your style, your data—and can handle anything you throw at it. Whether you're an artist transitioning to independence, a founder building a company, or someone managing complex health needs, PAI becomes whatever you need it to be.

PAI is part of the journey toward Human 3.0—where humans are augmented by AI to amplify their capabilities.

📖 Learn how PAI fits into the Human 3.0 vision

🧠 Life Management Finance Intelligence answer-finance-question - Analyze bank statements get-finances-state - Track spending patterns Parse PDFs for utility bills and expenses Memory & Conversations get-life-log - Search recorded meetings Query Limitless.ai pendant recordings Find what was discussed at any event Health Tracking Monitor wellness metrics Track medical records Analyze sleep and fitness data

💼 Professional Content Creation write-blog - Generate posts in your style publish-blog - Deploy to production enhance-content - Improve existing work Business Analytics get-newsletter-stats - Beehiiv metrics get-site-analytics - Website performance create-consulting-document - Client deliverables Development design-review - Code analysis browser-tools-setup - Automation update-daemon - API management

🎯 Personal Growth Knowledge Capture extract-knowledge - Learn from content capture-learning - Document insights web-research - Deep investigations Communication send-email - Automated outreach send-text-alert - SMS notifications send-discord-alert - Team updates Productivity add-linear-ticket - Task management create-mermaid-visualization - Idea mapping query-context - Access any stored knowledge

🌟 Why This System is Different

This system is designed from the very beginning to be available to anybody and to grow and scale with you throughout your life.

• 🔓 Open Framework: No vendor lock-in, complete transparency, you own everything
• 🌍 Universal: Works for anyone, anywhere, in any profession or life situation
• 🤖 Platform Independent: Core structure works with Claude, GPT, Gemini, or any AI platform
• 🧠 Persistent Memory: Your AI remembers every context, project, and conversation
• 🎯 Task Agnostic: From writing a blog to tracking medical data to running a business
• 📁 Plain Text: All configuration in human-readable files you can edit and understand
• 🔌 Extensible: Add your own commands, agents, and integrations

📖 Read the full technical deep-dive on PAI

---

⚡ Architecture

🏗️ The PAI System Architecture

graph TD
    You[👤 You] --> PAI[🧠 PAI]

    PAI --> Skills[📚 Skills]

    Skills --> Research[🔍 Research]
    Skills --> Development[💻 Development]
    Skills --> Content[📝 Content]
    Skills --> Life[🏠 Life Management]
    Skills --> Custom[🎯 Your Custom Skills]

    Research --> Agents1[Perplexity/Claude/Gemini Researchers]
    Development --> Agents2[Engineer/Architect/Designer]
    Content --> Agents3[Writer/Artist]
    Life --> Tools[Finance/Health/Memory Tools]

    Agents1 --> MCPs[🔌 MCP Servers]
    Agents2 --> MCPs
    Agents3 --> MCPs
    Tools --> MCPs

    MCPs --> Output[✨ Augmented Output]

    style You fill:#1f2937,stroke:#10b981,stroke-width:2px,color:#e5e7eb
    style Skills fill:#111827,stroke:#8b5cf6,stroke-width:3px,color:#e5e7eb
    style Output fill:#111827,stroke:#ef4444,stroke-width:2px,color:#e5e7eb

Loading

Skills are the core organizational unit. Everything in PAI is built around modular skills that activate based on your intent.

📚 Skills System

~/.claude/skills/ ├── prompting/ # Prompt engineering standards ├── create-skill/ # Skill creation framework ├── ffuf/ # Web fuzzing for pentesting (by @rez0) ├── alex-hormozi-pitch/ # $100M Offers pitch framework ├── research/ # Multi-source research (requires API keys) ├── fabric/ # Intelligent Fabric pattern selection (242+ patterns) ├── web-scraping/ # Web data extraction ├── chrome-devtools/ # Browser automation ├── youtube-extraction/ # YouTube transcript extraction ├── ref-documentation/ # Technical docs search ├── webapp-testing/ # Playwright testing patterns └── [your-custom-skills]/

Features: ✅ Modular capability packages 📄 Progressive disclosure (SKILL.md → CLAUDE.md) ⚡ Intent-based activation 📂 Self-contained with templates 🔌 Inherits global context

🎯 How Skills Work

Skills activate automatically based on your intent:

• Say: "Research the latest AI developments" → research skill loads → Launches parallel Perplexity/Claude/Gemini researchers

• Say: "Build a task tracker app" → development skill loads → Invokes architect and engineer agents

• Say: "Create a threat model" → fabric skill loads → Selects create_threat_model pattern from 242+ options

Each skill contains:

• 📄 Intent triggers ("USE WHEN...")
• 🤖 Specialized agents (if needed)
• 🔌 MCP integrations (if needed)
• ⚡ Commands and tools (if needed)
• 📖 Documentation and examples

Tip

You don't manage agents or commands directly. Just tell PAI what you want to do, and the right skill activates with all necessary resources.

💡 Example Skills (Included)

Skill	What It Does	Example Usage
🔍 research	Multi-source research with parallel agents	"Research quantum computing trends"
🧵 fabric	242+ AI patterns (threat modeling, summarization, extraction)	"Create a threat model for our API"
💻 development	Full-stack development with architect and engineer agents	"Build a meditation timer app"
🎨 design	UX/UI design with shadcn/ui and Figma integration	"Design a dashboard for analytics"
🔒 ffuf	Web fuzzing for penetration testing	"Test this API for vulnerabilities"
📊 alex-hormozi-pitch	Create irresistible offers using $100M Offers framework	"Create a pitch for my SaaS product"
🌐 web-scraping	Extract data from websites (BrightData + Apify)	"Scrape product listings from this site"
📖 ref-documentation	Search technical docs (React, Next.js, 100+ frameworks)	"How do I use React hooks?"
▶️ youtube-extraction	Extract transcripts and content from YouTube videos	"Summarize this YouTube video"
🎭 webapp-testing	Browser automation and visual testing	"Test the login flow"

Skills use MCP servers for integrations: Chrome DevTools • Apify • BrightData • Stripe • Anthropic Content • Daemon (your data) • And more...

---

🚀 Quick Start

Automated Setup:

./.claude/setup.sh  # Interactive setup script (recommended)

Or read: Getting Started Guide | Quick Reference

Note

PAI was originally built with Claude Code, but the architecture supports any AI platform (GPT, Gemini, etc.) Requires Bun JavaScript runtime for the current implementation.

Manual Installation

Step 1: Install Prerequisites

# Install Bun (JavaScript runtime)
brew install oven-sh/bun/bun 

# Get Claude Code (AI interface)
# Visit: https://claude.ai/code

Step 2: Clone PAI Repository

# Clone the PAI repository
git clone https://github.com/danielmiessler/Personal_AI_Infrastructure.git
cd Personal_AI_Infrastructure

# All PAI infrastructure is now in .claude/ directory
ls -la .claude/  # See agents/, skills/, commands/, etc.

Step 3: Copy .claude/ to Your Home Directory

# Copy the .claude directory to your home directory
cp -r .claude ~/.claude

# OR create a symbolic link if you want to keep it in the repo
# ln -s ${PAI_DIR}/.claude ~/.claude

Step 4: Configure PAI Directory Variable ⚠️ IMPORTANT

Important

You MUST configure the PAI_DIR variable to point to your PAI installation directory. This makes the PAI system completely agnostic to your installation location.

# Add to your shell config (~/.zshrc or ~/.bashrc)
# Replace /path/to/PAI with YOUR actual PAI installation path
export PAI_DIR="/path/to/PAI"  # Point to the PAI repository root
export PAI_HOME="$HOME"  # Your home directory

# Example (adjust to YOUR installation path):
# export PAI_DIR="$HOME/Projects/PAI"
# export PAI_DIR="$HOME/Documents/PAI"
# export PAI_DIR="/Users/yourname/PAI"

# Reload your shell configuration
source ~/.zshrc  # or source ~/.bashrc

Step 5: Configure Environment & API Keys

# Copy environment template
cp ~/.claude/.env.example ~/.claude/.env

# Configure your API keys
vim ~/.claude/.env

Important

Some skills and agents require API keys to function:

Required for Research Agents:

• PERPLEXITY_API_KEY - For perplexity-researcher agent (Get key)
• GOOGLE_API_KEY - For gemini-researcher agent (Get key)
• Claude WebSearch is built-in (no key needed for claude-researcher)

Optional for AI Generation:

• REPLICATE_API_TOKEN - For AI image/video generation (Get key)
• OPENAI_API_KEY - For GPT integration and Sora billing (Get key)

See .env.example for complete list and setup instructions. NEVER commit your .env file to version control!

Step 6: Launch PAI

# Optional: Set up voice notifications (macOS only)
# Download Premium/Enhanced voices from System Settings → Voice (Live Speech)
# Then start the voice server:
cd ~/.claude/voice-server && bun server.ts &
# See ~/.claude/documentation/VOICE-SETUP-GUIDE.md for detailed setup instructions

# Open Claude Code and start using PAI!
# Your personal AI infrastructure is ready 🚀

⚙️ Environment Variables

# ============ REQUIRED CONFIGURATION ============
PAI_DIR="/path/to/PAI"                  # PAI repository root (system agnostic)
PAI_HOME="$HOME"                        # Your home directory

# ============ RESEARCH AGENTS (Skills-Specific) ============
PERPLEXITY_API_KEY="your_key"          # For perplexity-researcher agent
GOOGLE_API_KEY="your_key"              # For gemini-researcher agent
# Claude WebSearch built-in - no key needed for claude-researcher

# ============ AI GENERATION (Optional) ============
REPLICATE_API_TOKEN="your_token"       # For Flux, Sora, etc.
OPENAI_API_KEY="your_key"              # For GPT integration, Sora billing

# ============ SYSTEM CONFIGURATION ============
PORT="8888"                             # Voice server port (default: 8888)

# ============ DIGITAL ASSISTANT CUSTOMIZATION ============
DA="YourAssistantName"                  # Your AI assistant's name (default: "Assistant")
DA_COLOR="purple"                       # Display color (purple, blue, green, cyan, etc.)

Tip

Why PAI_DIR is Important: The PAI_DIR variable makes the entire PAI system portable and installation-agnostic. All commands, skills, and hooks reference ${PAI_DIR} instead of hardcoded paths. This means you can install PAI anywhere on your system, and as long as PAI_DIR points to your PAI repository root, everything will work seamlessly.

---

📚 Documentation

📖 Guide	🎯 Purpose	⏱️ Time
Quick Start	Get up and running	5 min
Architecture	Understand the system	10 min
SECURITY.md	Security guidelines	5 min
Voice Server	Enable voice interaction	10 min
Commands Directory	Browse all commands	15 min

---

🎬 Real-World Examples

[!IMPORTANT] These are examples from my personal PAI setup. Your system will have completely different commands based on YOUR unique needs, workflows, and goals. The beauty of PAI is that it becomes whatever YOU need it to be.

💼 Professional 📈 Business & Analytics 📊 Newsletter Performance # Track newsletter metrics in real-time get-newsletter-stats # Returns subscriber growth, open rates, top content 🚀 Startup Dashboard # Monitor all key metrics get-site-analytics # Traffic and conversions get-finances-state # Revenue tracking add-linear-ticket # Development velocity 🔒 Development & Security 🔍 Code Security # Security automation check-for-sensitive --scan all design-review --security-focus browser-tools-setup --headers 🔌 API Infrastructure # Personal API management update-daemon --endpoints create-cloudflare-mcp-server update-tools-context --integrations startup-context-load ✍️ Content Creation 📝 Blog Pipeline # End-to-end content creation web-research "emerging AI trends" extract-knowledge --key-points write-blog --style personal --length 1500 enhance-content --seo --readability publish-blog --schedule tomorrow 🎨 Visual Content # Create stunning visualizations create-d3-visualization --data metrics.json create-custom-image "futuristic AI assistant" create-mermaid-visualization --type flowchart

🏠 Personal 💰 Financial Intelligence 📉 Expense Analysis # Instant financial insights answer-finance-question "monthly utilities average?" answer-finance-question "subscription spending?" process-finances-data --categorize --visualize 📈 Financial Planning # Comprehensive financial overview get-finances-state prepare-finances-dashboard push-finances-state --notify 🧪 Memory & Research 🔍 Meeting Recall # Never forget a conversation get-life-log "investor meeting last week" get-life-log "product roadmap decisions" query-context "action items from standup" 📚 Knowledge Building # Build your second brain capture-learning --source conference extract-knowledge --summarize learn-from-iterations --improve query-context "machine learning notes" 🎯 Personal Productivity 툿 Morning Automation # Start your day right get-life-log "yesterday's highlights" query-context "today's priorities" send-text-alert "heading to office" reset-visitor-peak # Reset daily counters 📧 Smart Communication # Intelligent outreach send-email --template followup send-discord-alert --channel team add-links --resource documentation update-public-content

🎨 Creative & Other Fields 🎭 Artist Workflow 🎨 Creative Portfolio # Manage artistic work track-project-progress "oil painting series" generate-artist-statement --style contemporary update-portfolio-website --new-pieces schedule-gallery-submission --deadline friday 📷 Photography Business # End-to-end photography workflow process-shoot-metadata --wedding-2024-09 generate-client-gallery --password-protected send-invoice --client johnson-wedding backup-raw-files --cloud-storage 🎵 Musician Setup 🎤 Music Production # Creative music workflow track-song-ideas --voice-memo-to-text generate-chord-progressions --key-of-c schedule-studio-time --next-available export-stems --for-mixing 🌿 Health & Wellness Coach 🏋️ Client Management # Wellness practice automation track-client-progress --weekly-checkin generate-meal-plans --dietary-restrictions send-motivation-message --personalized schedule-follow-up --in-two-weeks 📚 Academic Researcher 🔬 Research Pipeline # Academic workflow search-literature --keywords "machine learning ethics" organize-citations --zotero-import generate-literature-review --outline track-experiment-results --statistical-analysis 🏠 Real Estate Agent 🏡 Property Management # Real estate automation analyze-market-trends --neighborhood-data generate-property-reports --comparative-analysis schedule-showings --calendar-optimization send-client-updates --weekly-digest

---

🤝 Community

Help Us Give AI to the World

⭐ Star this repo to stay updated with the latest features!

---

🙏 Acknowledgments

Special thanks to contributors who have enhanced PAI with their expertise:

• Joseph Thacker (@rez0) - FFUF skill with comprehensive web fuzzing guidance and penetration testing expertise

📄 License

PAI is MIT licensed. See LICENSE for details.

---

The Path to Human 3.0

"Humans are what matter. AI is only as useful as it is to people. A system like this is need to level the field with AI and help us get to Human 3.0."

Created by Daniel Miessler

📧 Newsletter • 📝 Blog • 💼 LinkedIn • 🎬 YouTube