Personal AI Infrastructure (PAI) is an open-source framework by Daniel Miessler that transforms a general-purpose language model into a Digital Assistant (DA): a persistent, configurable companion that knows your goals, context, working patterns, and projects. Rather than treating the AI as a chatbot or a tool, PAI builds a continuous relationship between practitioner and AI across sessions.
Core Architecture
PAI organizes personal AI infrastructure into several interconnected layers:
TELOS: The Life OS
TELOS (from the Greek for “end” or “purpose”) is the heart of PAI. It is a structured collection of documents encoding the practitioner’s identity and context:
- MISSION.md: Core purpose and why it matters
- GOALS.md: Active goals across life domains
- BELIEFS.md: Core worldview and principles
- PROJECTS.md: Active projects with their full context
- MEMORY/: Cross-session memory (user profile, feedback patterns, project context, external references)
TELOS transforms a generic AI into a context-aware companion. Instead of re-explaining who you are and what you’re building every session, the DA already knows. The quality of TELOS directly determines the quality of assistance.
The Algorithm
PAI’s structured approach to every non-trivial task follows six phases:
- OBSERVE: Gather prima materia, dissolve assumptions, read context without preconception
- THINK: Classify effort, identify what “done” looks like, surface complications
- PLAN: Elevate understanding into precise specification and verifiable success criteria
- BUILD/EXECUTE: Alchemist and DA work in conjunction
- VERIFY: Test and validate across conditions that matter
- LEARN: Crystallize volatile insights into persistent memory
Skills and Commands
Skills are domain-specific knowledge packs that teach the DA about particular projects, stacks, or workflows. A Holochain skill encodes HDK patterns and zome architecture. A digital garden skill encodes content conventions and publishing workflows. Skills eliminate constant re-explanation: the practitioner switches context; the DA adapts.
Commands are trigger phrases that invoke specific workflows. Together with skills, they make the DA fluent in specialized domains.
Hooks
Automated behaviors triggered by events in the development environment (tool calls, session start, session end). Hooks implement persistent behavioral rules without relying on in-context instructions that disappear when context windows rotate.
CLAUDE.md: The Grimoire
The project-specific instruction file that governs the DA’s behavior within a given codebase or context. In the alchemical vocabulary of Mystical Oriented Programming (MOP), the CLAUDE.md is the Grimoire: the living charter of principles, constraints, and operational rules that shape every output.
The quality of the Grimoire determines the character of the assistance. Instructions that encode values (“errors must be honest at the type level”) produce code with integrity. Instructions that only specify syntax (“use TypeScript”) produce competent but characterless output.
The Alchemical Connection
PAI maps precisely onto the alchemical framework of MOP. The practitioner becomes the Alchemist (holding Sulphur: purpose, will, ethical direction). The DA becomes the Homunculus (holding Mercury: structural attention, pattern execution, presence without clinging). TELOS and CLAUDE.md together become the Grimoire: the instruction set that governs the Homunculus and encodes the quality of the Work.
This is not a metaphor. The structural parallels are exact: the Alchemist originates the why, the Homunculus executes the how with precision that may exceed the Alchemist’s own manual capacity, and the Grimoire governs the boundary between them.
See The Digital Homunculus: AI Assistants as Alchemical Companions for a full philosophical treatment of this relationship.
SoushAI: A Living Implementation
SoushAI is Soushi’s personal instance of PAI, built on Claude Code and instantiated through a customized configuration that includes the AlterNef TELOS, the PAI Algorithm, and a growing library of domain-specific skills (Holochain development, digital garden management, ValueFlows economics, and more).
See SoushAI: A Living Homunculus for a detailed self-analysis of this specific instantiation.
The Collegium Vision
PAI is designed to scale from individual use to collective infrastructure:
- PAI (Personal AI Infrastructure): One Alchemist, one Homunculus, one laboratory. The DA knows your TELOS, your projects, your patterns.
- CAI (Commons AI Infrastructure): Multiple practitioners share skill packs and learned patterns as commons resources. Each DA benefits from collective wisdom without any practitioner losing control of their own laboratory.
- Network of Laboratories: At full maturity, each node in a cooperative network could run its own DA with shared skills, local TELOS, and commons-contributed patterns, with intelligence emerging from cooperation rather than from central authority.