How to run your whole life and business from one folder with Claude

Most people use AI like a smarter search box: they open a chat, type a prompt, get an answer, close it, and start again from zero next time. The real power shows up when AI has persistent context about you, your work, and your knowledge—without locking you into yet another cloud app.

This article walks through a powerful approach: running your entire life and business from a single local folder, powered by Claude and a small team of specialized AI agents. It’s part personal knowledge management, part operating system for your work.

Why one folder beats endless AI chats

Standard AI chats (Claude Chat, ChatGPT, Gemini, etc.) have a big limitation: every conversation is basically a fresh start. Even with “memory” features, the model only remembers fragments, and you have no real control over what it knows or how it uses it.

In contrast, a single local folder gives you:

• Persistent context: your notes, projects, CRM, journal, SOPs, and research all live in one place.
• Full control: everything is plain files on your machine—easy to back up, move, or point other models at.
• Tool independence: you can switch from Claude to another LLM later without rebuilding your system from scratch.

Instead of trying to remember you, the AI simply reads from and writes to your folder. The folder becomes the memory. Claude becomes the brain that works with it.

The core idea: a personal knowledge assistant (PKA)

At the heart of this setup is a concept called a Personal Knowledge Assistant (PKA). It’s not a single bot, but a small team of AI agents that all operate inside one folder on your computer.

This folder holds everything that matters in your life and business:

• CRM (people and organizations)
• Projects and deliverables
• Journal and meeting notes
• Research and reference material
• Images, invoices, contracts, and more

All of it is stored as simple files and folders—Markdown documents, PDFs, images, etc. The PKA reads this information, updates it, and connects the dots between everything you do.

Inside the folder: structure that actually scales

The system starts from a scaffold—a prebuilt folder structure you can download and adapt. It’s designed to be personal first (not a shared team drive), but it scales to multiple businesses.

Key top-level folders include:

• PKM (Personal Knowledge Management)
Holds your core knowledge and life structure, including:

– CRM: people, organizations, and related files (contracts, invoices, etc.)
– Journal: daily entries, reflections, and logs (even backfilled from other tools)
– Images and media: screenshots, reference images, and visual assets going back years

• Deliverables
Where AI agents place finished work for you to review: social graphics, drafts, assets, code snippets, documents, and more.

• Team
A folder where each AI agent “lives” with its own instructions, avatar, and role description.

• Team knowledge
Shared guidelines, SOPs, and workstreams that all agents use. This is the single source of truth for how work should be done.

Because everything is just files, you can open and edit them with any editor you like—VS Code, Obsidian, a plain text editor, or anything else.

Meet the AI team living in your folder

Instead of one overloaded “super assistant” trying to do everything, this system uses multiple specialized agents. Each one has its own agent.md file with clear instructions and responsibilities.

Some examples:

• Larry – the orchestrator and your single point of contact (SPOC). You only talk to Larry. He routes work to the right specialists.
• Pax – the researcher. Uses tools like Perplexity and Brave Search via APIs to gather and cross-check information.
• Pen – the journaler. Turns your raw notes, screenshots, and audio into structured journal entries and links them to people, projects, and topics.
• Pixel – the thumbnail and social image creator. Follows your design system and brand guidelines.
• Sage – the writer. Drafts posts, emails, and content in your voice using your existing material as reference.
• Vera – quality assurance. Checks outputs for structure, quality, and alignment with your rules before they reach you.
• Nolan – HR. Helps “hire” new agents by defining roles, skills, and instructions when your needs grow.
• Mack – integrations and automations. Connects to external tools and APIs (e.g., Supabase, Notion, ClickUp).
• Vita – health and personal data analyst, interpreting long-term health data in context.

Each agent is lightweight: just a text file with instructions and links to the SOPs and guidelines it should follow. That’s what makes the whole system LLM-agnostic—you can swap out the model, but keep the same agents and knowledge.

How Larry orchestrates work like a real AI team

When you ask for something, you never talk to the individual agents. You always talk to Larry. His job is to understand what you want and dispatch tasks to the right specialists.

Here’s the basic workflow:

1. You make a request
For example: “Create a square social image with an infographic of last week’s most relevant AI productivity releases, plus a LinkedIn post in my voice.”

2. Larry checks the team index
He reads a central agent_index.md file that lists all agents, their roles, and which SOPs and guidelines they use.

3. He spins up the right agents in parallel
For the example above, Larry might route to:
– Pax for research
– Charter or Pixel for the infographic
– Sage for the LinkedIn post

4. Agents use only the relevant knowledge
Instead of loading your entire knowledge base every time, each agent only pulls the SOPs, guidelines, and files needed for that task. That keeps token usage efficient and outputs consistent.

5. Vera checks quality
Before anything hits your inbox, Vera runs a quality and safety check against your rules (structure, tone, privacy, brand, etc.).

6. Larry delivers the result
The final assets land in the Deliverables folder (for example, Deliverables/Inbox/...) for you to review and approve.

This mirrors Anthropic’s own dynamic workflows (planner, implementers, verifiers, fixers), but implemented entirely with local files and a simple folder structure.

Why not just use Claude Cowork or regular chat?

Claude Cowork and similar features are a big step forward because they let AI work against a folder. But there are some trade-offs:

• In Cowork, it’s still essentially one agent “switching hats” rather than a true parallel team.
• You’re tied to a specific interface and vendor implementation.
• You have less control over how agents are defined and how they share knowledge.

Using the Claude CLI in a terminal gives you the raw capabilities of Claude with maximum flexibility. You can:

• Run multiple sessions in parallel inside VS Code or another terminal app.
• Point Claude at different folders (your PKA, a clean test folder, a client folder, etc.).
• Swap in other models (Gemini, DeepSeek, etc.) against the same folder and instructions.

The folder is the system. The model is just the interchangeable brain you plug into it.

Setting it up with Claude in the terminal

Working in the terminal might sound intimidating, but the setup is straightforward:

1. Install the Claude CLI
Follow the official instructions (usually a single command you paste into your terminal). Log in with your Claude account (Pro or Max both work).

2. Open a terminal in your PKA folder
On macOS, for example, right-click the folder and choose “New Terminal at Folder”.

3. Launch Claude
Type claude and hit Enter. If your folder is set up with the scaffold, Claude will introduce itself as Larry, your team orchestrator, not just a generic assistant.

4. Start working in natural language
From there, you can ask for anything: research, content, images, SOP updates, CRM entries, or project planning. Larry routes it all.

Many people prefer to run this inside VS Code, where you can see your folder tree, open files, and run multiple terminals (multiple Larry sessions) side by side.

How the system connects to external tools

Even though the core memory lives in your local folder, the system can still talk to external tools and data sources via APIs. That’s where agents like Mack come in.

Typical connections include:

• Supabase – as a backend database (for example, for a membership site or app).
• Perplexity and Brave Search – for higher-quality web research than default web browsing.
• Notion, ClickUp, or other SaaS tools – to bridge existing workflows into your PKA.
• Social media schedulers – like Metricool, via an agent that drafts and schedules posts.

API keys and secrets are stored in a local .env file inside the Team Knowledge area. A security-focused agent (like Vex) and guardrails ensure only specific agents can access specific keys.

Standard operating procedures as AI fuel

The real leverage in this system comes from Standard Operating Procedures (SOPs) and workstreams. These are simple Markdown files that describe:

• How a process works step by step (e.g., “Generate a styled image” or “Publish a video”).
• Which agent owns each step.
• Which guidelines and constraints apply (brand, tone, privacy, etc.).
• Version history and reasons for changes.

Each SOP has a unique ID and lives in Team Knowledge/SOPs. Agents reference them by ID, so multiple agents can share the same SOP without duplicating information.

When something goes wrong—say an image doesn’t match your brand— you don’t just complain to the AI. You tell Larry to update the relevant SOP or guideline. Over time, the system gets sharper without you rewriting prompts every day.

From scattered notes to a connected life OS

Most people have their life spread across tools: journaling in one app, CRM in another, tasks in a third, files in cloud drives, and AI chats on top. The result is friction and lost context.

In this one-folder approach, everything is connected:

• A new person you meet gets added to the CRM and linked to your journal entry about them.
• Audio notes are transcribed and turned into structured entries with links to projects and people.
• Health data from years back is analyzed in context of your current lifestyle and goals.
• Your own book, courses, and content become the reference library for future content—so AI writes in your true voice, not generic “AI slop”.

This is also where this approach overlaps with broader ideas about AI productivity. If you’re curious about how this fits into different “levels” of AI usage, you may find this breakdown of the 3 levels of AI usage helpful context.

Real-world impact: one-person teams that feel like five

In practice, this setup lets one person do the work of a small team:

• Running a membership site and app
• Producing YouTube videos and social content
• Handling support tickets
• Building and maintaining products
• Managing multiple businesses

Previously, this might have required several human roles: designer, writer, researcher, QA, automation engineer, and more. With a well-structured PKA, those roles become AI agents that work alongside you, 24/7, always aligned with your rules and knowledge.

This is very similar in spirit to building a one-person business around Claude, as explored in this guide to launching a one-person e-commerce business with Claude. The difference here is that the system is fully local and tool-agnostic, so you’re not locked into any single SaaS platform.

Start small: your first version of the folder

You don’t need to replicate a 24,000-file system on day one. A practical way to start:

1. Download a scaffold that mirrors this structure (PKM, Deliverables, Team, Team Knowledge).
2. Keep the default core agents: Larry (orchestrator), Nolan (HR), Pax (research), Pen (journal).
3. Point Claude CLI at the folder and confirm it introduces itself as your orchestrator.
4. Feed it your existing knowledge: past journal entries, key documents, important contacts, and a few SOPs for recurring tasks.
5. Let the system grow with your needs: when you hit a limitation, ask Larry and Nolan to define and “hire” a new agent for that kind of work.

Over time, your folder becomes a living, evolving operating system for your life and business—one that any capable AI model can plug into.

Final thoughts

AI becomes truly transformative when it stops being a disposable chat window and starts acting as a persistent, structured partner in your work. A single, well-designed local folder—backed by Claude and a handful of specialized agents—can replace a mess of disconnected tools and contextless prompts.

If you’re serious about AI productivity, consider shifting your focus from “better prompts” to “better systems”. Start with one folder, one orchestrator, and a small team of agents. Let your workflows, not your tools, drive how you build it—and you’ll have a setup that survives model changes, app shutdowns, and the next wave of AI hype.