The step-by-step guide to setting up your AI-powered workspace. No technical experience required. From download to your first real task in under 30 minutes.
Most people know ChatGPT or Claude as a chatbot — you type a question, you get an answer. Cowork is different. Instead of answering one question at a time, Claude takes on actual work for you: writing documents, building spreadsheets, researching companies, creating presentations, organizing files, analyzing data — all on your computer.
Here's the key difference:
| Regular Claude (Chat) | Claude Cowork |
|---|---|
| You ask, Claude answers | You assign a task, Claude does the work |
| Text responses only | Creates real files: Word docs, Excel, PowerPoint, PDFs, HTML |
| No access to your files | Reads and writes to folders on your computer |
| Forgets everything between sessions | Remembers context through files you set up |
| One-shot interactions | Multi-step projects, runs while you step away |
Here are real examples of tasks people hand to Cowork daily:
Write memos, proposals, reports, blog posts, emails, and LinkedIn content as polished Word docs or PDFs.
Deep-dive a company, person, or topic and get a professional research document with sources.
Create real PowerPoint decks you can open in Microsoft PowerPoint or Google Slides.
Build Excel files with formulas, charts, and analysis. Analyze CSVs. Create dashboards.
Write scripts, fix bugs, automate repetitive tasks, build web tools.
Manage files, rename documents, extract info from PDFs, merge files.
Go to claude.com/download and download the app for your computer (Mac or Windows). Install it like any normal app — drag to Applications on Mac, or run the installer on Windows.
Open the Claude app and sign in with your Anthropic account. If you don't have a paid plan yet, you'll need to upgrade at claude.ai first.
Once you're signed in, look at the top of the app window. You'll see tabs for Chat, Cowork, and Code. Click Cowork to switch. (You can ignore the Code tab — that's for software developers.)
Heads up: The first time you open Cowork, it downloads a ~2GB Linux environment. This only happens once but can take 5–10 minutes on slower connections. Let it finish before moving on.
Cowork will ask you to select a folder on your computer. Don't overthink this — just pick or create a folder like ~/Documents/CoWork or ~/Claude-Workspace. We'll organize it properly in Step 6. Claude can only see files inside this folder (not your whole computer).
When you open Cowork and select your folder, you'll see a text input area — similar to regular Claude chat. But now Claude has superpowers: it can create files, browse the web, run code, and work through multi-step projects.
Type your request in plain English. Be as specific as you can. For example: "Write a one-page executive summary of WuunderFund for potential investors. Make it a Word doc."
If your request is ambiguous, Claude will ask before starting — just like a real colleague would. Answer the questions and Claude begins work.
You'll see a progress tracker showing what Claude is doing: reading files, searching the web, writing code, creating documents. You can watch or switch to another app — Claude keeps working.
When done, Claude will provide a link to the file it created in your workspace folder. Click it to open. The file is already on your computer — no downloading needed.
As of March 2026, Cowork has a first-party Projects feature built right in. A Project is a persistent workspace with its own files, instructions, and memory. Everything you do inside a Project stays scoped to that Project — Claude remembers your context across tasks without you re-explaining anything.
In Claude Desktop, click Projects in the left sidebar. Click New Project.
Give your project a clear name (e.g., "WuunderFund" or "Q2 Marketing"). In the Instructions field, add rules for how Claude should work in this project — tone, formatting, key context. These apply to every task in the project.
Add a local folder from your computer, link a chat project from regular Claude, or paste a URL for Claude to reference. This is the information Claude draws from when working on tasks in this project.
Click into the project and type your first request. Claude now has your instructions and context loaded automatically.
Memory is automatic and always on inside Projects. As you work, Claude periodically extracts useful facts — decisions, preferences, key details — and saves them for future tasks. When you start a new task in the same project, Claude loads those memories automatically.
| Behavior | How It Works |
|---|---|
| Scoping | Memory stays within its project. What Claude learns in Project A doesn't appear in Project B. |
| Persistence | Memories survive across tasks and sessions. Close the app, come back tomorrow — they're still there. |
| Auto-extraction | Claude decides what's worth remembering (roughly every 5,000 tokens of conversation). You can also say "remember this" to force it. |
| Viewing | Click the memory icon in the project view, or open the memory.md file in your project folder directly. |
| Limits | First 200 lines (or 25KB) are loaded at the start of each new task. |
Projects are a great foundation, but they don't give you everything. Here's where the rest of this guide comes in:
Projects don't separate where Claude reads from vs. where it writes. Without a system, files pile up in one folder. Steps 5–6 solve this with a read-only/write-only folder structure.
The instructions field is great for project-specific rules, but there's no dedicated place for "who you are" context that spans all projects. Step 7 creates your About Me file.
Auto-memory is convenient but imprecise — Claude decides what to save. If you want to guarantee specific facts persist with your exact wording, you need user-triggered memory. Step 8 covers this with CLAUDE.md and MEMORY.md.
Projects are stored locally. No sharing, no sync across devices. This is a known limitation — plan accordingly.
CoWork OS is a free plugin (created by Paul J Lipsky) that handles your entire workspace setup through a guided conversation. It creates your folders, About Me, CLAUDE.md, MEMORY.md, and project subfolders — all from a single /begin command. Steps 5–7 explain what each of these files does if you want to understand the details later.
Download the plugin file: cowork-os-plugin.zip (14 KB). Unzip it anywhere on your computer — you'll get a folder called cowork-os-plugin.
Open Cowork → click Customize (top-right) → Browse plugins → search for "CoWork OS". If you find it in the marketplace, click Install. Otherwise, use the "Install from folder" option and point it to the unzipped cowork-os-plugin folder.
/begin — That's ItOpen a new Cowork session and type /begin in the chat. CoWork OS takes over from here. You just answer questions.
/beginClaude will walk you through a short interview. Here's exactly what to expect:
| Claude asks you… | What it builds from your answer |
|---|---|
| What's your name and role? | Your About Me file — Claude's identity reference for who you are |
| What company/org are you at? What do you do day to day? | Context section in About Me so Claude understands your world |
| What tools do you use? (Google Docs, Slack, Excel, etc.) | Tools section — Claude knows what formats to create |
| How do you like to work? (Direct? Detailed? Short answers?) | Personal Preferences and behavior rules in CLAUDE.md |
| Any projects you're working on right now? | Project folders with isolated memory for each one |
When it's done (about 5 minutes), your workspace will have:
| What you say | What happens |
|---|---|
| "Remember this: [fact]" | Claude writes it to MEMORY.md. Persists across sessions. Say "forget that" to remove it. |
| "Create a folder for [project]" | Claude creates a project subfolder with its own CLAUDE.md and MEMORY.md — completely isolated from other projects. |
| "New project: [name]" | Same as above — sets up PROJECTS/[name]/ and CLAUDE OUTPUTS/[name]/ automatically. |
SKILL.md files work, how commands are defined, and how memory isolation is handled across subfolders.
A free plugin handles the entire setup for you. Install it, type one command, answer a few questions — and your workspace is built in 5 minutes. Folders, identity file, instructions, memory system, all of it.
Go to Step 5: Install CoWork OS →Prefer to build everything manually? Steps 6–8 walk you through each piece by hand.
Think of your workspace folder as Claude's desk. The better organized it is, the better Claude performs.
Here's the proven folder structure used by power users:
A simple text file that tells Claude who you are, what you do, and how you like to work. Claude reads this at the start of every session. This is what makes Claude feel like YOUR assistant instead of a generic AI. We'll set this up in Step 7.
Drop in any document structure you want Claude to follow. A report format, an email style, a slide deck structure. Claude will study the pattern and apply it — without copying the content. Start empty; add templates as you discover patterns you reuse.
This is where you put context Claude needs to do good work: briefs, research, decks, past work. Organize by project. Claude reads from here but never writes to it — your originals stay untouched.
Every file Claude creates goes here, organized by project (mirroring your PROJECTS folder). This keeps your workspace clean and makes it easy to find what Claude built.
When you start a new project, just tell Claude: "New project: [Project Name]". Claude will automatically create:
1. A PROJECTS/[Project Name]/ folder for your reference materials (briefs, docs, research)
2. A CLAUDE OUTPUTS/[Project Name]/ folder where Claude delivers all work for that project
3. A MEMORY.md inside the project folder — scoped to that project only
This means every project has its own memory. Decisions, deadlines, and context for Project A don't bleed into Project B. Claude knows which project it's working on and remembers the right things.
Use this pattern for everything Claude creates: Project_ContentType_v1.ext
| Example | What It Is |
|---|---|
Acme_Proposal_v1.docx | First draft of a proposal for Acme |
Q2-Launch_Deck_v2.pptx | Second revision of the Q2 launch deck |
Competitor_Research_v1.docx | Competitive research report |
Monthly_Report_v3.xlsx | Third revision of a monthly report |
When you ask Claude to revise something, it automatically increments the version number (v1 → v2 → v3). This gives you a built-in history of how deliverables evolved — no git required.
The About Me file is a simple text file that lives in your ABOUT ME folder. Here's what goes in it:
Here's what a real About Me file looks like (based on an actual power user's setup):
If you've been using ChatGPT, you've already built up months of context — your role, preferences, style corrections, project details. Instead of starting from scratch, you can extract all of that and use it to pre-fill your About Me, CLAUDE.md, and MEMORY.md files.
noreply@openai.com with a download link (expires in 24 hours). Check spam if you don't see it.
| Claude setup file | What goes in it | Where to find it in ChatGPT |
|---|---|---|
| About Me | Who you are, your role, how you work | Every time you told ChatGPT "I'm a [role] at [company]" or corrected its tone |
| CLAUDE.md | Workspace rules, naming conventions, behavior | Instructions you repeated — "always format it this way," "don't use bullet points," etc. |
| MEMORY.md | Decisions, deadlines, project context | Facts you kept re-explaining each session — deadlines, stakeholders, strategies |
| Personal Preferences | Communication style, personality | Times you said "be more direct," "stop over-explaining," "match my tone" |
| TEMPLATES/ | Reusable document structures | Prompts you copy-pasted repeatedly — report formats, email templates, frameworks |
Unzip the export file (you'll get chat.html). Drop it into your Cowork workspace folder and tell Claude:
"I just exported my ChatGPT history. It's in chat.html. Read through it and extract: (1) everything it knew about me — role, company, preferences, (2) instructions I repeated often, (3) recurring projects or topics, (4) key decisions and facts. Use what you find to draft my About Me file and seed my MEMORY.md."
What took you months to teach ChatGPT, Claude gets in one conversation.
This file lives in the root of your workspace. It tells Claude the rules of your workspace: what folders exist, how to name files, where to save outputs, and any specific behaviors you want. Claude reads this automatically at the start of every session.
This is Claude's external brain. It's a simple text file where Claude stores things you ask it to remember between sessions. Unlike regular chat (which forgets everything), MEMORY.md persists.
How Memory Works:
When you create a new project (see Step 4), each project gets its own MEMORY.md. This is powerful because:
Cross-project stuff: your role, recurring contacts, preferences, tool accounts, things that are always true regardless of project.
Project-scoped notes: this project's deadlines, decisions, status updates, stakeholder feedback, version history. Stays isolated from other projects.
These two files serve different purposes. Here's when to update each:
| MEMORY.md | CLAUDE.md | |
|---|---|---|
| What goes here | Things that change: status, decisions, deadlines, contacts, what happened last session | Things that don't change: rules, folder structure, naming conventions, workflow instructions |
| How to update | Say "remember this" or "make a note" during any session — Claude writes it automatically | Edit the file directly when your workflow or rules change (rare) |
| How often | Every session or two | Maybe once a month |
| Who updates it | Claude (when you ask) | You (manually) or Claude (when you explicitly ask "update CLAUDE.md") |
Before connectors and plugins, there's one capability worth setting up immediately: Claude in Chrome. It's a browser extension that lets Claude see and interact with your browser — fill out forms, navigate websites, take screenshots, pull data from pages, and automate web tasks.
When Claude can see your browser, it can research things live, reference what you're looking at, and do web-based tasks without you copy-pasting content back and forth. It's especially powerful combined with Cowork — you can say "look at this page and write a competitive analysis" and Claude reads the live page, then creates a Word doc.
To install: Search for "Claude" in the Chrome Web Store or look for the Claude in Chrome option in your Claude Desktop settings.
Connectors give Claude direct access to your tools — no copy-pasting, no exporting. Claude reads and writes natively. Adding one takes about 30 seconds: click "Connect" and sign in. There are 50+ connectors available across every category. Here are the most popular:
| Connector | What Claude Can Do |
|---|---|
| Google Drive | Read, search, and reference your Google Docs, Sheets, Slides |
| Gmail | Read emails, draft and send responses, search your inbox |
| Google Calendar | Read your schedule, find conflicts, prep for upcoming meetings |
| Slack | Read channels, search messages, catch up on what you missed |
| Notion | Read and reference your Notion workspace, databases, and wikis |
| Jira | Read tickets, track issues, understand sprint context |
| HubSpot | Access CRM contacts, deals, and company records |
| Salesforce | Pull account, opportunity, and contact data from your CRM |
| Canva | Create and edit designs directly |
| GitHub | Work with repos, issues, PRs, and code |
| Airtable | Read and query your Airtable bases and records |
Full directory at claude.com/connectors — categories include communication, data, design, finance, productivity, sales, and more. All connectors are free.
To add a connector: Go to Cowork → Customize → Connectors → browse or search → click "Connect" and authorize access.
Plugins teach Claude how to do domain-specific work at a professional level. Each one bundles skills, templates, and workflows built for a specific role. All free, all built by Anthropic.
Task management, memory system, daily workflow. The foundation plugin — pairs well with everything else.
Account research, call prep, outreach drafting, pipeline review, competitive battlecards, forecasting.
Content creation, campaign planning, SEO audits, brand voice review, email sequences, performance reports.
Journal entries, reconciliation, variance analysis, financial statements, SOX compliance, close management.
Contract review, NDA triage, compliance tracking, risk assessment, DPA review, meeting briefings.
PRDs and specs, roadmaps, stakeholder updates, competitive analysis, user research synthesis, metrics tracking.
SQL queries, data exploration, visualization, interactive dashboards, statistical analysis, data validation.
Process documentation, risk assessment, vendor management, compliance tracking, change management, resource planning.
Design critique, accessibility audits, UX writing, design system management, developer handoff specs, user research.
Search across all your connected tools at once — email, Slack, docs, wikis. Daily and weekly digests.
Not sure where to start? Here's what to install based on your role:
| If you're a… | Install these first |
|---|---|
| Founder / CEO | Productivity + Sales + Claude in Chrome + Gmail + Google Calendar |
| Marketer | Marketing + Productivity + Claude in Chrome + Google Drive + Slack |
| Salesperson | Sales + Productivity + Claude in Chrome + HubSpot or Salesforce + Gmail |
| Product Manager | Product Management + Data + Jira + Slack + Notion |
| Finance / Accounting | Finance + Data + Google Drive |
| Legal | Legal + Productivity + Google Drive + Gmail |
| Designer | Design + Claude in Chrome + Canva |
| Ops / Chief of Staff | Operations + Enterprise Search + Productivity + Slack + Google Calendar |
| Data / Analytics | Data + Airtable + Google Drive |
| Just exploring | Productivity + Claude in Chrome + Google Drive — add more as you need them |
Under the hood, every plugin is made up of skills — small instruction files that teach Claude how to do one thing really well. A skill might teach Claude how to write a contract review, build an Excel dashboard, or research a company using a specific framework.
You don't need to know this to use Cowork — plugins handle it for you. But understanding skills matters because you can eventually create your own. If you find yourself giving Claude the same instructions repeatedly ("always format reports this way" or "research companies using this framework"), that's a skill waiting to be built. We cover this in the Level 2 Advanced Guide.
"Research [Company Name]. Give me a Word doc covering what they do, their business model, key people, recent news, and competitive positioning."
Why it's great first: Zero setup needed. Claude uses web search and creates a real .docx file.
"Write a one-page executive summary of my company for potential investors. Here's what we do: [brief description]. Make it a Word doc."
"Create a 10-slide pitch deck for [your company/project]. Include: problem, solution, market size, business model, team, and ask. Make it a .pptx file."
Drop a CSV or Excel file into your workspace folder, then ask: "Analyze the data in [filename]. Tell me the key trends and create a summary dashboard."
"I'm preparing for a meeting with [Company/Person]. Research them, create a briefing doc, and draft 3 talking points I should bring up. Save everything to my project folder."
The biggest mindset shift for Cowork: stop trying to write the perfect prompt and start giving Claude the right context. That's what the folder structure, About Me file, CLAUDE.md, and MEMORY.md are doing — they're building a context layer that makes every conversation better. A mediocre prompt with great context beats a perfect prompt with zero context every time. This is why the setup in Steps 5-7 matters so much.
This is the single easiest way to get better results. Instead of just stating what you want, close your prompt with "ask me any questions to give better context before you start." Claude will then interview you — asking about audience, tone, scope, format, and details you didn't think to mention. The 30 seconds you spend answering saves you from getting something wrong on the first pass.
When you're asking Claude to do something that requires deep reasoning — strategy, analysis, debugging, research synthesis — include "think hard" in your request. This activates Claude's extended thinking mode, where it spends more time reasoning through the problem before responding. The difference is significant on complex tasks. You can also say "think step by step" or "take your time with this."
Claude comes in three tiers. You can select your model in the Cowork interface:
Opus — The most capable. Best for complex analysis, deep research, multi-step projects, and anything that needs careful reasoning. Slower but smarter. Use this when quality matters more than speed.
Sonnet — The sweet spot. Fast, capable, handles 90% of tasks well. Great default for everyday work: documents, research, presentations, code.
Haiku — The fastest. Best for quick tasks: reformatting text, simple questions, file organization, rapid iteration. Use when speed matters most.
Rule of thumb: start with Sonnet. Switch to Opus when the task is high-stakes or complex. Switch to Haiku when you need something quick and simple.
Go to Claude Desktop → Settings → Personal Preferences. Paste a block that describes your name, role, working style, and communication preferences. Unlike the About Me file (which only works in one folder), Personal Preferences apply to every Claude session — Chat and Cowork. This is where you tell Claude your personality expectations.
Each project can have its own CLAUDE.md and MEMORY.md. This means Claude can have different instructions and memory for different projects. A marketing folder can have Claude act as a content strategist; a finance folder can have Claude act as an analyst.
Whenever something important comes up — a decision, a deadline change, a new contact — say "remember this: [fact]." Claude will write it to MEMORY.md. Next session, it'll already know.
Cowork can run tasks on a schedule. Example: "Every Monday at 9am, compile a summary of my unread emails and create a priorities list." Set this up once, get it delivered automatically.
Drop files directly into the chat or put them in your workspace folder. Claude can read PDFs, Word docs, Excel sheets, CSVs, images, and more. The more context Claude has, the better the output.
If you skipped this in Step 9, go back and install it. Claude in Chrome lets Claude see and interact with your browser — research, form filling, screenshots, web automation. It's one of the highest-impact additions you can make.
Always tell Claude what format you want: "Make it a Word doc," "Create a .pptx," "Give me an Excel file," "Build an HTML page." Claude can create nearly any file type — but only if you ask.
When starting fresh work, just say "new project: [Name]." Claude creates the folder structure, project-specific memory, and output directory automatically. Everything for that project stays together from day one. See Step 4 for the full workflow.
Anthropic's own research on large language models shows that Claude develops internal representations that mirror collaborative dynamics. In practice: collaborative, respectful framing produces better outputs than demanding or terse commands. Instead of "Write me a report," try "I need a report on X for my board — here's the context. Can you draft something and I'll refine it with you?" The difference isn't politeness for its own sake — it's that collaborative framing gives Claude more context about your intent, audience, and quality bar. Claude works best when you treat it like a sharp colleague, not a vending machine.
| Resource | URL |
|---|---|
| Download Claude Desktop | claude.com/download |
| Getting Started Guide | Cowork Help Center |
| Safety Guidelines | Using Cowork Safely |
| Plugins Guide | Using Plugins |
| Scheduling Tasks | Recurring Tasks |
| Connectors | Extending Capabilities |
| Skills Documentation | Building Skills |
| Cowork Product Page | claude.com/product/cowork |
/begin — this handles Steps 6–8 automaticallyNo. Cowork is designed for non-technical users. You describe what you want in plain English. Claude handles the technical execution.
No. Claude can only see files in the specific folder you selected. It cannot access your entire computer, your browser history, passwords, or anything outside that folder.
Files stay on your computer. Claude processes them locally in a sandboxed environment. However, text from your conversations and files may be sent to Anthropic's servers for processing. Review Anthropic's privacy policy for details.
The task stops. Files created up to that point are saved, but the task won't resume automatically. You'd need to ask Claude to continue where it left off.
Claude Pro is $20/month and includes Cowork. Max ($100/month) gives you more usage. Team and Enterprise plans also include Cowork with additional features.
Cowork requires the desktop app (Mac or Windows). However, Pro and Max users can message Claude from their phone while the desktop handles execution. Your output files sync via iCloud/OneDrive if you set up your workspace in a synced folder.
You've got the workspace, the context files, and the tools connected. The best way to learn Cowork is to use it. Pick a real task from Step 10 and go. When you're ready to level up, the guides below show you what's next.