Words about your computer
- Editor
- A program that shows you the contents of files and lets you change them. This course uses VS Code as the common baseline. (Microsoft Word is also an editor, but it's designed for prose, not for code or plain text.)
- Terminal
- A program that lets you type commands one at a time and see the response. On Mac it's literally called “Terminal.” On Windows the modern app is called “Windows Terminal.” Inside VS Code, View → Terminal opens an integrated terminal pane.
- Shell
- The program inside a terminal that interprets your commands. On Mac the default shell is zsh; on Linux it's usually bash; on Windows it's PowerShell. They all do the same job; commands differ slightly between them.
- Command line
- Generic name for the whole typing-based interface. “Type that into the command line” means “type it into your terminal.”
- PowerShell
- Microsoft's shell, the default on Windows. Used inside Windows Terminal. Some commands (like cd, ls) work the same as on Mac/Linux; others (like Get-Content instead of cat) are different.
- Console (developer)
- A web page a cloud provider gives you for managing your account, generating API keys, viewing usage, and setting billing caps. Different from your terminal/console even though both are sometimes called “the console.” When in doubt, check whether you're being asked to open a browser or a terminal.
- Path
- A string of folder names that points to a file. Mac/Linux uses forward slashes (folder/subfolder/file.md); Windows uses backslashes (folder\subfolder\file.md). Both refer to the same thing.
- PATH (environment variable)
- A list of folders your shell searches when you type a command. When an installer asks “Add to PATH?”, say yes — that's how typing code or claude from any folder works.
- Working folder / working directory
- The folder you're currently “in” inside a terminal. pwd prints it; cd changes it.
Words about the AI stack
- Local LLM runner
- A class of tool that downloads model weights to your disk and gives you a way to send prompts to the model on your own machine. Ollama and LM Studio are the two this course uses.
- Quantization
- Compressing a model's numerical weights to lower precision (e.g., 4-bit instead of 16-bit) so the model fits in less RAM and runs faster. Costs roughly 3–5% of the model's quality and saves roughly 3× the memory. q4_K_M is the standard “small but good” version this course uses.
- Tokens per second (tok/s)
- How fast a model generates text once it's started. For comparison: human reading speed is about 4–6 tok/s; comfortable AI chat feels like 20–40 tok/s; under 5 tok/s is too slow for real iterative work.
- Latency to first token
- The pause between hitting Enter and seeing the first character of a response. Usually 1–2 seconds on a good local setup, 3–8 seconds on CPU-only inference.
- Localhost
- An address (127.0.0.1) that means “this same computer.” When a program on your machine listens at localhost, other programs on your machine can talk to it without going through the internet. Local LLM runners expose models at a localhost address.
- Model name format
- Local models are named family:size-purpose-quantization, like llama3.1:8b-instruct-q4_K_M. The model name is exact — typing llama3.1:8b without the rest will fail.
Hardware vocabulary
- CPU
- The main processor in your laptop. Every laptop has one. Runs everything. Not specialized for AI math.
- GPU (graphics processing unit)
- A chip designed for the parallel math AI models love. Originally for video games; now the default hardware for AI work.
- Discrete GPU
- A separate, dedicated GPU chip. Common on gaming laptops (NVIDIA RTX, AMD Radeon RX). Uncommon on thin-and-light or budget laptops.
- Apple Silicon
- Apple's M1/M2/M3/M4 chips, found in Macs from late 2020 onward. CPU and GPU share the same RAM, which makes them surprisingly capable for local AI.
- CPU-only inference
- Running the model on your CPU because no compatible GPU is available. Works for small models; just slower.
- RAM (random-access memory)
- Your computer's working memory. AI models load into RAM while running. A quantized 8 B model needs about 5–7 GB.
Cloud account and API vocabulary (optional advanced — only if you took the API path in Lesson 2.4)
- Anthropic Pro
- The paid Anthropic subscription this course is built around. Around $20/month, billed directly by Anthropic. Unlocks the Claude desktop app and its three tabs (Chat, Cowork, Code). Month-to-month; you can pause it between modules.
- API
- Application Programming Interface. A way for one program to talk to another program. Cloud AI providers expose APIs that your code can call. Most students don't need to use the API directly — the Pro subscription and the desktop app cover the bulk of the course.
- API key
- A secret string that identifies your account when your code calls the provider's API. Treat it like a password with your money attached. Only relevant for students on the optional advanced path.
- Subscription billing
- A flat monthly fee that covers a provider's consumer chat product and its agentic features for unlimited or very-high use. Predictable cost; no per-request charge. Anthropic's tiers as of this writing are Free, Pro, and Max. On Pro, the same subscription unlocks Claude.ai chat, the Claude desktop app, and all three of its tabs (Chat, Cowork, Code).
- Per-token billing
- Pay-as-you-go pricing where you are charged for each request based on the number of tokens (input + output) it consumed. Used for direct API access, custom code, MCP servers, and scheduled tasks. Total cost varies with usage; you set a monthly cap to bound it.
- .env file
- A plain-text file holding environment variables — named settings your code reads at runtime. The leading dot is a Unix convention for “config file.” One KEY=value pair per line, no spaces around the =, no quotes.
- Environment variable
- A named setting that programs running on your computer can read. Different from a setting written into your code, because the value isn't visible in the code itself.
- JSON
- A labeled-text format programs use to exchange data — {"key": "value", "list": [1, 2, 3]}. The same {key: value} shape you saw in Module 1's traces.
- Token (cloud pricing)
- A piece of text — roughly three-quarters of a word. Cloud APIs charge by tokens in (your prompt) and tokens out (the response), usually quoted as dollars per million tokens.
- Pay-as-you-go
- A billing model where you pay only for what you use, charged monthly against the card on file (rather than a fixed monthly subscription).
- Cost ceiling / spend cap
- A monthly maximum you set in the provider's developer console. Once exceeded, API access stops until next billing cycle or you raise it. Always set one before you write production code.
- Haiku / Sonnet / Opus
- Anthropic's model size tiers, named after poetry forms. Haiku is fastest and cheapest; Sonnet is the everyday workhorse (5–10× the per-token price of Haiku); Opus is the most capable and most expensive.
- Usage dashboard
- The provider's web page showing month-to-date spend, broken down by day and by model. Bookmark it; check weekly.
Source-control vocabulary (introduced lightly here, deepened in Module 3)
- Git
- A tool that tracks changes to files over time. The most common version-control system in software development.
- GitHub
- A website that hosts git projects. Sometimes free, sometimes paid; sometimes public, sometimes private. Public repositories are scraped by bots looking for leaked API keys.
- Source control / version control
- Generic name for systems like git. Tracks who changed what, when, and lets you go back.
- Commit
- A saved snapshot of your project's state in git. “Commit a key” means accidentally including it in a snapshot you push online — which is how keys leak.
- .gitignore
- A file in your project that tells git which files to never track. .env should always be in .gitignore.
- Repository (repo)
- A project tracked by git. A “starter repo” is a small example project the course gives you to practice on.
- Clone
- Make a local copy of a repository from GitHub onto your computer. git clone <url> does it; downloading the ZIP from the GitHub page also works.
The Claude desktop app and its tabs
- Claude desktop app
- The primary tool for the rest of this course. A single app you install on macOS or Windows. One Anthropic Pro subscription unlocks all three of its tabs — Chat, Cowork, and Code — with no separate installs.
- Chat tab
- The conversation tab. Like the regular Claude.ai chat — no file access, no agent autonomy. Use it for quick questions and brainstorming.
- Cowork tab
- The autonomous tab. You hand the agent a task and it works on it independently in a cloud environment, sometimes for minutes, sometimes on a schedule. Most of this course (Modules 4 through 8 and the capstone) lives here.
- Code tab
- The interactive coding tab. The agent reads and writes files in a folder you've granted it access to, and proposes each change with a real-time diff for you to accept or reject. Module 3 lives here.
- Coding agent
- A model + a loop + tools (Module 1 vocabulary), aimed at directing your computer to read and modify code. Different from a chat window because it acts on real files. The Code tab is one. The Claude Code CLI is another. They share the same engine.
- Claude Code
- The Anthropic coding-agent engine. Available two ways: as the Code tab inside the desktop app (the default for this course), or as a separate CLI install you run from a terminal (optional advanced — for students who'd rather work in a shell). Same engine, two interfaces.
- Diff
- The highlighted before-and-after of a change. When an agent edits a file, the diff is what you review before accepting.
- Directing surface
- The combination of editor + terminal + coding agent. Together they let you see what changed (editor), act on your computer (terminal), and orchestrate the work (agent).
Install-tool vocabulary
- Python
- A programming language. The most common one for AI work and the one this course's example projects use. You'll run Python scripts and read short Python files; you won't write Python from scratch — that's what the agent is for.
- pip
- Python's package installer. pip install <package> installs a Python library you can import; python3 -m pip install <package> is a slightly more reliable way to invoke it. Roughly Python's equivalent of npm.
- Node.js
- A runtime that runs JavaScript outside a browser. Many command-line tools (including Claude Code) are written as Node packages.
- npm
- Node's package installer. npm install -g <package> installs a tool globally so you can run it from any folder.
- LTS (Long-Term Support)
- A stable, maintenance-friendly release of a tool. The opposite of “bleeding-edge.” Choose LTS when you're installing tools you'll use every day.
- Global install (-g)
- A package installed globally is available from any folder, not just the one you ran npm install in. Used for command-line tools that you want to invoke as a single command.
- OS keychain
- Your computer's built-in password vault. Mac: Keychain Access. Windows: Credentials Manager. Linux: depends on distro. A safe alternative to .env files for storing API keys, especially if you don't trust yourself to keep .env out of git.
Posture vocabulary (from Lesson 2.5)
- Posture
- A short, written commitment about how you work — specific enough to act on, general enough to survive new tools. Different from a “rule” because it's about habits, not just a single do/don't.
- Cost posture
- Four lines: monthly cost ceiling, how you'll know you're approaching it, what you'll do when you hit it, and your high-volume rule.
- Privacy posture
- Four lines: default placement rule, always-local categories, fine-in-cloud categories, when-unsure rule.
- Sneaky-cost drift
- Slow, invisible increase in recurring spend from low-cost tools running frequently. Caught by weekly dashboard checks.
- Sneaky-privacy drift
- Slow, invisible widening of the data surface going to cloud — usually from a new tool that “helpfully” uploads context. Caught by reading what every new tool uploads, before you install it.