# Liz OS — Technical Specification & Case Study

> Workstation operations and context orchestration engine built on Google Antigravity SDK. Solves ADHD session amnesia and multi-project context thrashing through a Centralized Cortex (`.liz/`), the Open Knowledge Format (OKF v0.1), and deterministic subagent tool coordination.

- **Status:** Internal Production // Workstation Runtime
- **HTML Case Study:** https://asimansari.com/projects/brain-liz.html
- **Canonical Domain:** https://liz.asimansari.com

---

## 1. Architectural Motivation

Managing multiple complex projects (model distillation, backend systems, university academics, client deliverables) across terminal sessions leads to severe context fragmentation. Traditional chat interfaces suffer from session amnesia upon closure.

Liz solves this by decoupling persistent domain intelligence from ephemeral chat threads.

---

## 2. Centralized Cortex & Open Knowledge Format (OKF v0.1)

Instead of cluttering individual code repositories with agent scratchpads, Liz operates from a centralized root cortex:

```
.liz/
├── MEMORY.md                 # 3-day sliding window of battlefield progress & active goals
├── AGENTS.md                 # System instructions, subagent role definitions, and voice rules
├── projects/                 # Per-project specifications, verified architectures, and state
├── voice/                    # Local TTS synthesis pipeline (Kokoro-82M on CPU)
└── plugins/cortex/           # Native Model Context Protocol (MCP) server
    ├── server.py             # Export, report generation, and task spec tools
    └── skills/               # Sandboxed Python skill execution modules
```

---

## 3. Subagent Orchestration & Native Tooling

- **Google Antigravity SDK Core:** Manages reactive agent lifecycles, background execution, and inter-agent communication.
- **Dedicated Subagent Hierarchy:** Dispatches specialized subagents (`ai_tutor` for Feynman LangGraph evaluations, `dotnet_tutor` for C# architecture, and research subagents).
- **Communication Invariant:** Enforces high-density, ADHD-optimized communication standards: zero layout fluff, flat single-level lists, and in-flow bold anchors for 3-second visual scannability.