How to actually
use Claude.

The website at claude.ai. Where you type, upload files, run quick tasks. Best for: research, writing, brainstorming, anything that fits in a conversation.

Claude with your business tools connected. Gmail, Slack, Notion, Calendar, HubSpot, Stripe, Canva, Wix, and more. Lets Claude read from and write to your real systems, not just talk about them.

The most powerful Claude surface. Runs in your computer's terminal. Can read your files, write new files, build websites, ship apps. The advanced track. Best for: anything that becomes a real artifact you keep.

A folder inside Claude.ai that holds persistent context. Upload brand docs, set a system prompt, work across many chats with the same setup. Use when you do the same kind of work repeatedly for the same client or topic.

How one program talks to another. Think of a waiter: you order from a menu, the kitchen makes it, the waiter carries the request there and the food back. When Claude "checks your calendar" or "searches the web," it's calling an API. You'll also hear "the Claude API": that's developers using Claude inside their own products instead of through claude.ai.

A password that proves it's your account when a program calls an API. Looks like a long random string of letters and numbers. Treat it like a bank card number: never paste it in public, never put it in a shared doc, never leave it visible in a screenshot. Anyone who has it can spend on your account.

The login flow where a popup asks "Allow this app to access your account?" It lets a tool use your Gmail or Notion without ever seeing your password, and you can revoke access anytime from your account settings. This is what happens when you connect a Cowork plugin.

A standard plug that lets Claude connect to outside tools. Before MCP, every connection had to be custom-built. MCP is like USB-C: one standard port, and any tool that speaks it (Notion, Gmail, Slack, Canva) can plug into Claude. You never touch MCP directly. Plugins handle it for you.

One ready-made connection in a box. The Gmail plugin, the Notion plugin, the Canva plugin. Install it, log in once, and Claude can use that tool from then on. Under the hood it's MCP. On the surface it's two clicks.

A saved playbook that teaches Claude how to do one specific job your way. "How we write client proposals." "How we research a niche." A prompt is instructions for one task. A skill is instructions Claude keeps and applies automatically every time that kind of task comes up.

Claude using outside abilities mid-task: web search, file reads, code execution. Different from Claude "knowing" something. Tool use is Claude going to look it up.

Claude working in a loop instead of answering once. It makes a plan, uses tools, checks its own results, and keeps going until the job is done or it needs you. Cowork workflows and Claude Code sessions are agents in action.

An API in reverse. Instead of you asking "anything new?", the tool calls you the moment something happens. New payment in Stripe, new form submission, new client signup: the webhook fires instantly and triggers whatever you've connected to it. APIs pull. Webhooks push. Automations run on webhooks.

Tools that chain apps together with "when X happens, do Y" rules. New invoice paid → send thank-you email → log in spreadsheet. They existed before AI and pair well with it: the automation handles the trigger, Claude handles the thinking step in the middle.

The engine behind everything. A large language model is software trained on enormous amounts of text until it can read and write like this. "Claude" is the model family. Haiku, Sonnet, and Opus are versions, like trim levels of the same car: Haiku is fast and light, Opus is the deepest thinker.

How models read. Text gets chopped into pieces: a short word is one token, longer words are two or three. Context limits and API pricing are counted in tokens. Rule of thumb: 1,000 tokens is roughly 750 words.

Everything you send Claude in one message: the request, the context, the examples, the rules. The 4 Layers framework at the top of this program is just the anatomy of a good prompt.

Persistent instructions that shape how Claude behaves across every message. Lives in a Project setting or at the top of a Claude Code session. Think of it as Claude's job description.

A file in your project folder that briefs Claude Code on what the folder is for. Open any folder, drop a CLAUDE.md, and every future Claude session knows the context. Like a sticky note for Claude.

How much Claude can "remember" in one conversation. Big now (hundreds of pages worth). But not infinite. When chats get long, important details drift. Start a new chat for new tasks.

Something Claude builds you can use directly. A file, a webpage, a chart, an app. Different from a chat reply because you can save it, share it, ship it.

When Claude states something false with full confidence. It's not lying, it's pattern-completing plausibly. This is why Module 4 exists: ask for sources, verify the claims that matter, never ship an unchecked number.

A model that handles more than text. Claude reads images, screenshots, PDFs, and charts, not just words. Practical upshot: stop describing things. Screenshot them and upload.

Building software by describing what you want in plain English and letting AI write the code. You judge the result by using it, not by reading the code. It's how Track C works, and it's how this page was built.

The text window where you talk to your computer directly: no buttons, no icons. Type a command, press enter, read the result. Looks intimidating, is actually simple. Claude Code lives here. On Mac it's the Terminal app, on Windows it's PowerShell.

Any tool you use by typing commands instead of clicking. "claude" is a CLI. So is "git". The opposite is a GUI (graphical user interface): the buttons-and-windows version. CLIs feel old-school but they're faster, scriptable, and exactly where AI tools live right now.

Save-game history for files. Every change recorded, every version recoverable, nothing ever truly lost. A "commit" is one save point with a note about what changed. You don't need to master it: Claude Code handles Git for you. You just need to recognize the words.

The website where Git histories live online. A "repo" (repository) is one project's folder there. It's where developers share code, where open-source projects live, and where your projects back up once they get serious.

Frontend is everything you see and click: the page, the buttons, the design. Backend is everything behind it: the database, the logic, the payments. Everything in Track C is frontend. The hand-off rule in Module 15 is about knowing when you've hit backend territory.

A computer that's always on, whose whole job is answering requests from the internet. Every website lives on one. "The server is down" means that computer stopped answering.

Where an app keeps its information so it survives: users, orders, posts, settings. A spreadsheet with superpowers. The moment your app needs to remember things between visits for many people, you need one. That's usually the moment to call a developer.

Your own computer pretending to be a web server. When you "run a site on localhost," only you can see it. It's the dress rehearsal before deploying.

Putting your site or app on the real internet so other people can reach it. Module 14 covers it. Services like Cloudflare Pages and Vercel make it free for simple sites.

A domain is your address on the internet (aexagon.com). DNS is the phone book that points that name at the actual server. When you "connect a domain," you're updating the phone book. Changes take minutes to hours to spread.

Content delivery network. Someone else hosts a file on fast servers worldwide so your page can borrow it. When our pages load Tailwind "via CDN," the styling library comes from their server instead of living in your file. Needs internet on first load.

Pre-built code you stand on instead of starting from zero. A library is a toolbox you pick from (Tailwind for styling). A framework is a structure you build inside (React for apps). When Claude says "I'll use React for this," it's choosing a framework.

Anything that doesn't work the way it should. Not a disaster, a normal part of building. The fix workflow: screenshot what's wrong, describe what you expected, hand both to Claude.

Software whose code is public: free to use, inspect, and modify. Most of the internet runs on it. When a tool is open source, you're not locked in and nobody can take it away.