Glossary
Glossary
Short definitions for the words a first AI-built app throws at you. Each one is a working definition for this site, not a vendor spec.
Answer
These terms show up in the guides and on the tool pages. The definitions are plain English so a beginner can follow a prompt, a deploy screen, or a price table without already knowing the jargon.
API
A way for one program to ask another program to do something. Your app calls an API with a request, and the other service sends back data or an error. A weather site, a login service, and a model all expose APIs.
API key
A secret string that tells a service which account is calling its API. Treat it like a password. It belongs in an environment variable on the server, not in the code you paste into a public chat or commit to a public repo.
Environment variable
A named value the app reads from the machine it runs on, instead of from the source file. People write them as KEY=value in a local file such as .env, and they set the same names in the host's dashboard when they deploy.
Deploy
Putting the app on a server other people can open. A local preview on your laptop is not a deploy. A URL that still works after you close the laptop is.
Repo
Short for repository. The folder of files for a project, plus the history of how those files changed. On GitHub that folder lives at a URL you can clone.
Git
The tool that records snapshots of a repo. You use it locally. GitHub is a website that stores those snapshots so other people, and deploy hosts, can fetch them.
Commit
One saved snapshot in Git, with a message that says what changed. A commit is how you get back to a version that worked after an agent rewrites a file.
Branch
A line of commits you can change without touching the main line. You try a feature on a branch, then merge it when you want those commits on the main line.
Pull request
A request, usually on GitHub, to merge one branch into another. It is also the place people leave comments on the diff before the merge.
Agent
A coding assistant that can take more than one step: read files, edit them, and run commands, instead of only answering in chat. The product pages on this site say what each agent is allowed to touch.
MCP
Model Context Protocol. An open standard for connecting an AI app to an outside system, such as a file store or an issue tracker. A server you install can see whatever that server's docs say it can see.
Token
A small chunk of text a model counts. Vendors bill and cap usage in tokens, not in words. A long prompt and a long reply both spend tokens. The exact count is the vendor's, and this site does not estimate one for you.
Context window
How much text the model can hold in one turn: your prompt, the files it read, and the reply so far. When the window fills, older text drops out or the request fails. A bigger window is a product limit, not a sign the answer is right.
IDE
Integrated development environment. An editor with a file tree, a terminal, and usually a way to run the app. Cursor and Devin Desktop are IDEs. A browser builder is not.
CLI
Command-line interface. You type commands in a terminal instead of clicking. Gemini CLI, GitHub Copilot CLI, and Git itself are CLIs.
Localhost
Your own computer, as a server only you can open. A preview at http://localhost:3000 is running on your machine. It is not on the internet until you deploy it.
Framework
A library that decides the shape of the app: how pages are files, how data loads, how you deploy. Next.js is the framework this site is built with. Picking one early saves the agent from inventing a structure.
Frontend
The part of the app that runs in the browser: layout, buttons, and what the visitor sees. It should not hold secrets.
Backend
The part that runs on a server: saving data, checking a login, calling an API with a secret. Visitors do not read this code in the browser.
Database
Where the app stores rows or documents that should still be there after a refresh. A file on your laptop is not a database other visitors can share.
Authentication
Checking who someone is, usually with an email and password, a magic link, or a sign-in provider. Authorization, a different word, is what that person is allowed to do after you know who they are.
Rate limit
A cap on how often you can call a service, per minute or per day. Free plans hit these first. The error usually says 429 or "rate limit" and waiting, or a paid plan, is what the vendor's docs describe.
Prompt
The instructions you give the model. A useful prompt names the file, the change, and what "done" looks like. A vague prompt still produces code, just not the code you meant.
Hallucination
A confident answer that is not true. Agents invent package names, API fields, and prices. Check the claim against the vendor's docs or against the file on disk before you ship it.