Building a Personal AI Operating System with Claude Code
Simon Scrapesgo watch the original →
the gist
A guide to structuring a local folder-based 'Operating System' for Claude Desktop, using markdown files for persistent memory, brand voice, and modular 'workstations' to reduce prompt engineering overhead.
The Architecture of a Personal AI OS
Instead of treating Claude as a stateless chat interface, this method treats the Claude Desktop app as an agentic system by connecting it to a local folder structure. By maintaining a hierarchy of markdown files, the user provides the model with persistent context, brand guidelines, and specific operational rules that persist across sessions. The core of this system relies on a root claude.md file, which acts as the system's instruction manual, and a memory.md file, which serves as a dynamic store for project decisions and user preferences.
Establishing Brand and Operational Context
To move beyond generic AI outputs, the system uses a modular approach to brand identity. By creating a brand_context folder, the user stores specific markdown files for Voice Profile, Ideal Customer Profile (ICP), and Visual Identity (design tokens). These files are not just static text; they are referenced by Claude to ensure that every asset—whether written or visual—aligns with the user's established style. The process involves using 'Skills'—pre-defined process documents—that guide Claude through a step-by-step interview to generate these files, ensuring the AI 'learns' the user's specific constraints and preferences.
Scaling with Workstations
As the system grows, the user can implement 'Workstations' to prevent context pollution. By nesting subfolders (e.g., Finance, Operations, Marketing) within the root directory, each workstation can house its own claude.md and memory.md. This allows for specialized rules that apply only to specific tasks. The global claude.md handles universal instructions, while workstation-specific files handle domain-specific logic, ensuring the AI only pulls relevant context for the task at hand.
Managing Memory and Iteration
The system is designed to be self-updating. By instructing Claude to write to memory.md whenever a significant decision is made, the user creates a feedback loop where the AI 'remembers' past project states. This reduces the need for repetitive prompting. The workflow encourages running the system through the 'Code' view in Claude Desktop for better markdown rendering and file management, while using the 'Co-work' view for interactive task execution.