Magnify what
you're capable of.
PAI is an open-source Life Operating System — a framework on top of Claude Code. A dedicated Digital Assistant that knows who you are, what you're working toward, and moves you from your current state toward your ideal state. Continuously.
The context layer between you and the engines.
Bottom: Claude Code — the engine. Top: you — your life, work, and goals. PAI sits between them as the context layer that knows who you are and translates what you want into what the engine should do.
Claude Code is the engine. PAI is everything else that makes it yours.
Know who you are. Write it down. Pursue it.
TELOS is the context engine that captures the whole of you and translates it into action — your identity, your pursuit, and what you're learning. It reconciles where you are today with where you're trying to go, continuously, file by file. Your Digital Assistant reads TELOS before it does anything else.
The whole game: current state → desired state. Everything else is figuring out how to close the gap.
Your life, always on-screen.
PAI renders a live statusline at the top of every terminal — refreshed every second. Five Life OS dimensions tracked toward your Ideal State. Context loaded, token spend, git state, rating sparklines across 15m / 1h / today / week / month.
The same picture your DA sees when it helps you — visible to you the moment you open a shell.
Continuous learning. Not stateless chat.
A well-designed system with a mediocre model will outperform a brilliant model with poor scaffolding. Every time. PAI's memory is the scaffolding. Three tiers (hot / warm / cold), signals captured from every interaction, corrections that become rules, preferences that become defaults. The system stops asking the same questions because it writes down your answers.
A team of a thousand. Working on your life.
The aspiration: a team of 1,000 or 10,000 people working for you on your personal and business goals. The reality: persistent assistant plus parallel specialists. Your DA is a friend, coach, mentor — not a stateless agent. When depth is needed it dispatches specialists in parallel, each with its own personality and voice, each scoped, each background.
Ambient awareness. Your DA has a voice.
ElevenLabs TTS on every response. Prosody enhancement so speech sounds natural — not the default robot. Named character voices per agent. You don't look at a chat window. You work, and your system tells you when something's done, when something's off, when a decision needs you.
What PAI is made of.
A small set of composable primitives, each you can read, fork, or replace. No black boxes. No vendor lock.
- Claude Code
- TypeScript
- Bun
- Bash
- TELOS (30+ files)
- USER/SYSTEM separation
- upgrade-safe customization
- Skills (63)
- Hooks (21)
- Workflows (338)
- Algorithm v3.6.0
- 3-tier hot/warm/cold
- phase-based learning
- continuous signal capture
- CLI
- Terminal statusline
- Voice (ElevenLabs)
- Notifications (ntfy · Discord)
- policy hooks
- no --dangerously-skip-permissions
- containment zones
- AI-based GUI installer
- platform detection
- identity wizard
- MIT
- open-source
- self-hosted
Clone it. Run the installer. You're running PAI.
The GUI installer detects your system, installs prerequisites (Bun, Git, Claude Code), asks for your identity and preferences, and configures the repo into ~/.claude/. USER/ files are preserved on every upgrade.
Upgrading from an earlier version? The installer auto-detects existing installs, merges settings non-destructively, and never overwrites your USER/ files or custom hooks.
People are building this with us.
PAI is open-source on GitHub. Star it, fork it, open an issue, send a PR. The list below is live — pulled from the repo right now.