How I Use Claude Projects as My Personal AI Operating System

Every time you open a new Claude conversation, you start from zero. You re-explain who you are, how you like things structured, what context matters. It works, but it's friction you shouldn't have to deal with every single time.

Claude Projects solves this, and once I set mine up properly, it genuinely changed how I use AI day to day. Here's exactly how I did it and why it works.

The Problem: Stateless AI Is Exhausting

The default AI chat experience is stateless. The model doesn't know your name, your role, your preferences, or your working style unless you tell it. Every session. Every time.

For occasional use that's fine. But if you're using Claude for serious work, you're burning time re-establishing context that should already be there.

What you actually want is an agent that already knows the important stuff before you type a single word.

What Claude Projects Actually Does

Projects give you a persistent workspace. Inside a Project, you get:

  • A system prompt (called "Project Instructions") that's always active, across every conversation in that project
  • Uploaded files that the model can reference
  • Access to connectors like Google Drive, Gmail, and Calendar baked right in

The system prompt is the key part. Whatever you put there becomes the permanent baseline context your agent works from. Every conversation inside that project inherits it automatically.

The Setup: Two Documents, One Smart Agent

My setup revolves around two files I upload into every Project I care about.

The User Document is about me. My background, timezone, preferences, the kind of work I do, how I like responses structured. Think of it as a briefing doc the agent reads before every session.

The Soul Document is about how the agent should behave. Its operating principles, its tone, what it should and shouldn't do, how it should handle edge cases. This is where personality and judgment live.

Together, they give the agent both context (who am I talking to) and character (how should I show up).

Once you have those two files, you reference them directly in the Project Instructions:

You are Nash's personal assistant.
Before every session, reference the following files that have been uploaded to this project:
- User.mdContains personal context about Nash: background, preferences, timezone, and ongoing projects.
- SOUL.mdDefines your core personality, behavior principles, and how you should respond.
Always apply both. User.md tells you who you're helping. SOUL.md tells you how to help.

That's it. Now every conversation in this project starts with the agent already oriented. It knows who you are and how to behave before you've typed a word.

Transferring Memory Between Agents

Here's where it gets useful at scale.

Say you want a dedicated writing agent, a separate research agent, and another one just for code reviews. Normally you'd have to configure each one from scratch.

With this setup, you just drop both docs into each new Project and update the instructions for that agent's specific focus. The core identity travels with you. You're not starting over, you're extending.

It also means if Claude's Projects feature ever changes, or you want to spin up a new workspace, you're not locked out of your own context. The docs live with you, not inside the tool.

A Note on Persistent Memory

There's a broader ecosystem of tools built specifically around giving AI agents dynamic, cross-session memory. Tools like Mem0, Letta, and Zep extract facts from your conversations over time and retrieve them automatically in future sessions. They go further than what Claude Projects does natively, where each new conversation starts fresh with only your static instructions and uploaded files carried forward.

If you need that level of memory sophistication, those tools are worth exploring. But they come with a tradeoff: you're adding infrastructure, configuration, and another system to maintain.

For my use case, static persistent context through Project Instructions and uploaded docs is enough to get 90% of the value with none of the setup overhead. And crucially, it works natively inside the tools I'm already using.

Why I Stayed in the Anthropic Ecosystem

Honestly, a lot of this comes down to Claude Code.

I spend a significant chunk of my dev time in Claude Code already. Having my Projects, my connectors, and my coding environment all under one roof removes a category of friction I don't want to think about. The context compounds over time. The more you build out your setup, the more useful it gets.

The connectors for Calendar, Gmail, Google Drive, and more are one-click setups inside the Claude interface. No dedicated machine. No gateway to configure. No security hardening before it's safe to run. You connect your accounts and it works.

Switching to a multi-tool stack might give you marginal gains in specific areas, but you pay for it in integration overhead. For people already living in the Anthropic ecosystem, the path of least resistance is to stay there.

What This Looks Like Day to Day

On any given day I'm using my Project agent to:

  • Draft and review emails with full awareness of my communication style
  • Brainstorm product ideas where it already knows my background and what I've been building
  • Plan around my calendar without re-explaining my preferences each time
  • Review code with context about the project it's part of

The shift is subtle but real. It stops feeling like a tool you open and starts feeling like a collaborator that's already up to speed.

Getting Started: Build Your Own

You don't need my exact files. You need your own version of them. Here's what to cover:

Your User Doc should include:

  • Your name and what to call you
  • Your timezone and location context
  • Your current projects and roles
  • Your preferences (communication style, response format, things that annoy you)
  • Relevant background the agent should know

Your Soul Doc should include:

  • Tone and personality principles
  • What the agent should always do
  • What the agent should never do
  • How it should handle uncertainty or edge cases
  • Any hard rules around external actions (sending emails, posting, etc.)

Create a new Claude Project, upload both files, paste your Project Instructions referencing them, and you're live.

This Is Still Early

Projects and connectors are still evolving. The native integrations will get better, the memory layer will get smarter, and the workflows you can build will get more interesting.

But the core idea, giving your AI persistent identity and context, already works well enough to be worth setting up today. The compounding effect is real. The earlier you build it, the more useful it gets.

If you set something up, share it. I'm @neevash on X.