Note: This is a very early v1.0.0 release — consider it an alpha. Expect rough edges and iterative improvements.

PAI (Personal AI Infrastructure) is a methodology and toolkit for building your own AI assistant. The primary PAI releases are built on Claude Code, which is the most powerful AI coding harness available today. But many people want to use local models, open-source models, or a mix of providers — and that's what this scaffold is for.

Pi is a model-agnostic coding agent that works with any LLM provider: Ollama (local), OpenAI, Anthropic, OpenRouter, or any OpenAI-compatible API. This directory contains everything you need to run PAI's core methodology inside Pi.

This scaffold gives you the PAI Algorithm, skills, and core tooling — adapted for Pi's extension system. It's the foundation you build on.

Releases/Pi/
├── config/
│   ├── SYSTEM.md          # System prompt — your AI's identity + PAI methodology
│   ├── AGENTS.md          # Context about you + optional voice config
│   ├── settings.json      # Pi runtime settings
│   └── models.json        # Model provider configuration (examples included)
├── extensions/
│   └── pai-core/
│       └── index.ts       # Core extension: security, PRD tracking, voice, learning
├── skills/                # 9 skill categories with SKILL.md definitions
│   ├── agents/            # Agent composition and team coordination
│   ├── content-analysis/  # Extract wisdom from any content
│   ├── investigation/     # OSINT and research
│   ├── media/             # Diagrams, images, infographics
│   ├── research/          # Multi-mode research methodology
│   ├── scraping/          # Progressive web scraping
│   ├── security/          # Security assessment frameworks
│   ├── telos/             # Life OS — goals, projects, wisdom tracking
│   └── thinking/          # First principles, red team, council debate
└── memory/                # Persistent memory across sessions
    ├── learning/          # Session logs and learning signals
    ├── state/             # Current work state
    └── work/              # PRD files for Algorithm sessions

The core of PAI is the 7-phase Algorithm — a structured methodology for complex work:

1. OBSERVE — Reverse-engineer the request. Generate testable success criteria (ISC).
2. THINK — Identify risks, run a premortem, check prerequisites.
3. PLAN — Design the approach. Choose depth vs breadth.
4. BUILD — Prepare artifacts, invoke capabilities.
5. EXECUTE — Do the work. Mark criteria as they pass.
6. VERIFY — Test every criterion with evidence.
7. LEARN — Reflect on what worked and what didn't.

This methodology works with any model — it's a thinking framework, not a model capability. Even smaller local models benefit from this structured approach.

┌─────────────────────────────────┐
│         Pi Coding Agent         │
│  (model-agnostic runtime)       │
├─────────────────────────────────┤
│       PAI Core Extension        │
│  Security │ PRD │ Voice │ Learn │
├─────────────────────────────────┤
│    SYSTEM.md (Identity + Algo)  │
├──────────┬──────────────────────┤
│  Skills  │  Memory             │
│  (9 cat) │  learning/state/work│
├──────────┴──────────────────────┤
│       Your Model Provider       │
│  Ollama │ OpenAI │ Anthropic    │
│  OpenRouter │ Any compatible    │
└─────────────────────────────────┘

See INSTALL.md for the full setup guide. The short version:

# 1. Install Pi
npm install -g @mariozechner/pi-coding-agent

# 2. Copy this scaffold to Pi's config directory
cp -r config/* ~/.config/PAI-pi/
cp -r extensions/* ~/.config/PAI-pi/extensions/
cp -r skills/* ~/.config/PAI-pi/skills/
cp -r memory/* ~/.config/PAI-pi/memory/

# 3. Configure your model provider in models.json
# Edit ~/.config/PAI-pi/models.json

# 4. Customize your identity in SYSTEM.md
# Replace {{YOUR_NAME}} and {{YOUR_AI_NAME}} placeholders

# 5. Launch
pi

Edit config/SYSTEM.md and replace:

• {{YOUR_NAME}} — Your name
• {{YOUR_AI_NAME}} — What you want to call your AI assistant

Create a new directory under skills/ with a SKILL.md file:

---
name: my-skill
description: What this skill does. USE WHEN trigger words here.
metadata:
  author: you
  version: 1.0.0
---

# My Skill

Instructions for the AI when this skill is activated.

Set environment variables to enable TTS:

export PAI_VOICE_ENABLED=true
export PAI_VOICE_ENDPOINT=http://localhost:8888/notify
export PAI_VOICE_ID=your-voice-id

Works with ElevenLabs, Coqui, or any TTS server that accepts HTTP POST with JSON body.

This scaffold is derived from PAI v4.0.3. The full PAI system on Claude Code has 63+ skills, 21 hooks, and 338 workflows. This Pi version provides the core methodology and 9 foundational skill categories — enough to build a powerful personal AI system that you extend over time.

To upgrade from Pi to full PAI: Install PAI v4.0.3 on Claude Code. The methodology is the same — your skills and memory structure will transfer.

This scaffold is part of the PAI open-source project. Issues and PRs welcome.

Same as the PAI project — see LICENSE.