Glossary

These definitions describe how the course uses each term. Product-specific behaviour still comes from current official documentation.

AI and working context

Agent An AI system that can use tools, observe results, and take several steps toward a goal. Tool access does not grant permission to use every available action.

Alignment Training and system design intended to shape model behaviour toward instructions, safety requirements, and human preferences.

API (application programming interface) A structured way for one program to request data or an action from another.

API key A credential commonly used to identify and authenticate an API caller or project. Whether it is secret, publishable, restricted, or privileged depends on the provider's documentation.

Connector An integration that lets a tool access another system, such as a repository or file store, within granted permissions.

Context The working material available to a model for a request. It can include instructions, messages, files, retrieved passages, selected code, and tool results.

Context window The model-specific limit on how much input and output can be handled in one interaction or sequence, as defined by the provider.

Embedding A numerical representation used to compare meaning or similarity. Retrieval systems often use embeddings to find relevant passages.

Evaluation (eval) A repeatable set of cases and checks used to measure whether a model or workflow meets its requirements.

Fine-tuning Additional training on examples to shape recurring model behaviour for a defined task. It is not a live document store.

Inference Running a trained model to produce an output from current input.

Language model A model trained to generate or score sequences such as text. A fluent answer is not automatically a sourced fact.

Model The trained system that transforms an input into an output. The surrounding product adds instructions, tools, context, permissions, and storage.

Model tier A provider's grouping or choice that trades capability, speed, and cost. Names and availability change.

Prompt Instructions and context supplied to a model for a task.

RAG (retrieval-augmented generation) A pattern that finds relevant source material and adds it to the model's request-time context.

Reasoning effort A setting exposed by some models or products that controls how much work the model spends before answering. It is distinct from model choice.

Token A unit processed by a model. Tokens are not the same as words, and providers use them to describe limits and usage.

Weights Numerical parameters adjusted during training and used by a model to produce outputs. They are not a browsable table of the original training documents.

Files, code, and local tools

Branch A named line of Git history used to develop or review changes separately.

Build The process that turns project source into output that can run or be deployed.

CLI (command-line interface) A program controlled by commands typed in a terminal.

Commit A Git snapshot of the staged project state with an identifier and message.

Dependency External code or a package that the project declares it needs.

Diff A line-by-line representation of changes between two file states.

Directory Another name for a folder.

Editor A program for reading and changing files.

Environment variable A named value supplied by the environment where a program runs. It is configuration, not automatically a secret.

File path The sequence of folders and filename that locates a file.

Git A distributed version-control system that records project history locally and can exchange it with remotes.

GitHub A service that hosts Git repositories and collaboration features. It is not Git itself.

IDE (integrated development environment) An editor with integrated project tools such as search, terminal, debugging, version control, and agent panels.

Lockfile A package-manager file that records the resolved dependency graph for reproducible installs.

Markdown Plain text with lightweight markers for headings, lists, links, images, code, and other structure.

Migration A versioned change to a database schema or its data. Applying one can be consequential or destructive.

Node.js A JavaScript runtime commonly used by web build tools and servers.

Package manager A tool such as npm or Bun that installs declared packages and runs project scripts.

Project root The top folder of one project, usually containing its README, manifest, and version-control metadata.

Repository A project folder tracked by Git, including its history and configuration.

Runtime The environment that executes a program.

Secret A value that grants access or authority and must not be exposed, such as a private API key or database credential.

Terminal A text interface for running commands in a particular working directory.

Web, hosting, and data

Astro The web framework taught for the landing-page project.

Authentication Checking who a user is.

Authorization Checking what an authenticated user is allowed to access or change.

Callback URL A route to which an external authentication or integration flow returns the browser.

Canonical URL The preferred public URL for a page when more than one URL could display it.

Client Code running in the user's browser or device. Public client code must not contain server secrets.

Database A system for storing and querying structured data with rules about access and consistency.

Deployment A built version of the project made available in an environment such as preview or production.

DNS (Domain Name System) Records that map a domain name to the service responsible for it.

Domain A human-readable internet name, such as example.com.

Environment A distinct configuration and runtime target, commonly local, preview, or production.

Framework Code and conventions that organize how an application or site is built.

Next.js The React framework used by the course's application starter.

OAuth A protocol used to authorize access and support sign-in through another provider without giving the application the user's provider password.

Preview A non-production deployment used to inspect a proposed change.

Production The environment serving the real audience or users.

PRD (product requirements document) A reviewed definition of the user, problem, flows, constraints, acceptance criteria, and non-goals for the app.

Pull request A GitHub proposal to compare, discuss, test, and merge changes from one branch into another.

RLS (row-level security) Database policies that control which rows a database role or user may access. A privileged server connection can have different enforcement behaviour, so the actual query path still requires review.

Server Code running in a controlled runtime rather than the public browser. Server-side location alone does not make a query authorized.

Session State that lets an application recognize an authenticated user across requests.

Supabase The service used by the app route for Postgres and authentication.

Vercel The deployment host taught for the landing-page and app routes.

Vertical slice One complete user workflow through interface, server logic, data, and verification.

If a course page uses an unexplained term, open an issue or propose a glossary addition. Do not guess from the name.