The words you'll run into, in plain English. No prior knowledge assumed — that's the point.
- Agent
- An AI that doesn't just answer questions but takes actions toward a goal — editing files, running commands, checking its own work, and looping until the task is done. Coding agents like Claude Code and Codex can build whole features across many files from one instruction.
- API
- Application Programming Interface — the menu of requests one piece of software will accept from another. When your app fetches weather data or sends an email, it's calling someone's API. Your own backend exposes an API that your app's pages call.
- Backend
- The server-side half of an app: the code and database running on machines you control, where secrets can be kept and rules enforced. Its counterpart is the frontend — everything running in the visitor's browser, all of which is visible to them.
- Build
- The step that converts your project's source code into the optimized files that actually get served to visitors. 'The build failed' means this conversion hit an error — the message it prints says where, and deployment platforms run a build on every push.
- CLI / terminal
- Command Line Interface — controlling software by typing commands into a text window (the terminal) instead of clicking buttons. Coding agents like Claude Code live here. Looks hostile, is learnable in an afternoon, and mostly you'll paste commands the AI gives you.
- Context window
- The amount of text an AI model can 'see' at once — your conversation, the code it's reading, its own recent work. When a chat gets long or a project gets big, older material falls out of view, which is why long debugging threads go in circles and why starting a fresh chat often fixes a stuck AI.
- DNS
- The internet's contact list — the system that translates a name like vibestar.ai into the address of an actual server. Connecting a custom domain to your app means adding a couple of DNS records at your registrar.
- Environment variable
- A named setting (like an API key or database URL) supplied to your app from outside the code — set in a dashboard for production, a local file for development. This is where secrets live so they never get committed to git.
- Framework
- A pre-built skeleton for apps that handles the universal plumbing — pages, routing, data loading — so code only has to describe what's unique about yours. Next.js, SvelteKit, and Rails are frameworks; most AI tools pick one for you, and that's fine.
- GitHub
- The default place git repositories live online: backup, sharing, and the connection point that hosting platforms watch to auto-deploy your app. (Owned by Microsoft; alternatives like GitLab exist.)
- Hook
- A way to run your own code automatically when something happens. Coding agents offer hooks ('run the tests after every edit'), git offers them ('check the code before every commit'), and React confusingly uses the same word for functions like useState that plug into its rendering machinery.
- Localhost
- Your own computer, as a web address (usually http://localhost:3000). It's where your app runs during development — visible only to you. If you send someone a localhost link, they see nothing: deploy first.
- MCP
- Model Context Protocol — an open standard that lets AI tools plug into outside services (databases, browsers, project trackers) in a uniform way. Practically: installing an MCP server gives your coding agent new abilities, like directly querying your Supabase database.
- Migration
- A saved script that changes your database's structure ('add a playlists table'), kept in order alongside your code. Migrations mean your database's evolution is written down, repeatable, and in git — instead of a pile of undocumented dashboard clicks.
- Model
- The underlying AI brain a tool talks to — Claude, GPT, Gemini, and friends, each in several sizes. Tools swap models frequently; bigger models reason better and cost more. 'Which model is this using?' is a fair question to ask any AI tool.
- Token
- The unit AI models read and write text in — roughly three-quarters of a word. Model pricing, context windows, and usage limits are all measured in tokens, which is why long conversations and big files cost more and eventually overflow.
Missing a term you had to look up elsewhere? Tell us in any project discussion — the glossary grows from real confusion.
