The technical words

118 terms you will meet when you build, from AI assistants to containers, in plain words. You don’t need them to start; they’re here for when someone, or an assistant, uses one.

AI and assistants

Agent: An assistant that carries out several steps on its own (reading, deciding, acting) rather than answering one question at a time.
AI assistant: A chat tool you talk to in plain words, which writes, explains and builds things for you.At FirstLine: The approved ones are Claude, Gemini, ChatGPT for Teachers and Codex, always signed in with your FirstLine account.
Artifact: Claude’s word for a page or small tool it makes in the chat. It appears beside the conversation and can be shared as a link.At FirstLine: Shared with FirstLine, it opens for any colleague signed in to Claude.
Canvas: The space beside the chat in Gemini (and in some versions of ChatGPT) where a document, page or small app is built and edited.At FirstLine: Gemini Canvas can share a page straight to Google Classroom.
Claude Code: Anthropic’s coding assistant. It runs in a browser at claude.ai/code, or on a computer.
Codex: OpenAI’s coding assistant, which can work on code kept in GitHub.
Coding assistant: An assistant that works on code: it reads a repository, makes changes and proposes them for review.At FirstLine: claude.ai/code and Codex both work on code kept in FirstLine’s GitHub.
Connector: A link that lets an assistant read another system, such as Google Drive or BigQuery, with your permission.At FirstLine: The BigQuery Student Data Project uses a BigQuery connector that signs in as you.
Context: Everything the assistant can see while it answers: the conversation so far, files you attached, and instructions it was given.
Custom GPT: ChatGPT’s saved, reusable assistant with its own instructions. OpenAI is retiring them in favor of Projects and plugins.
Gem, or skill: Gemini’s saved, reusable instructions for a task. Google is replacing Gems with skills.At FirstLine: On school accounts, Gems stop working no sooner than June 1, 2027.
Gemini Notebook: Google’s research notebook (formerly NotebookLM): you add sources and ask questions about them. Notebooks can be shared with colleagues.
Hallucination: When an assistant states something false as if it were true, such as a made-up number or a link that does not exist. Check facts against the source.
MCP: Model Context Protocol: the common standard many connectors use to plug outside tools and data into an assistant.
Model, or LLM: The AI underneath an assistant. LLM stands for large language model: a program trained on a great deal of text to predict and write language.
Project (Claude or ChatGPT): A workspace inside the assistant with its own instructions and files, so every chat in it starts with the same background.At FirstLine: BigQuery Student Data is a Claude Project shared with all of FirstLine.
Prompt: What you type to an assistant. A clear prompt says who you are, what you want, who it is for and what it should use.At FirstLine: Step 3 on the front page is a fill-in-the-blanks prompt.
Token: Two meanings. For an AI model, a small piece of text it reads or writes; limits and costs are counted in tokens. For sign-in, a temporary digital pass (see Sign-in and access).
Vertex AI: Google Cloud’s service for using AI models, including Gemini, from inside your own software rather than through a chat window.

Code and GitHub

.gitignore: A list of files Git should never save, such as passwords and keys.
Branch: A separate line of changes, so work in progress doesn’t touch the live version until it is ready.
Branch rule, or ruleset: A setting that protects main, for example by requiring a pull request and a passing check before anything merges.
Check, or CI: Automatic tests that run on every pull request; a red check blocks the merge. CI stands for continuous integration.At FirstLine: The starter kit’s check compiles the code and runs its tests.
CLAUDE.md and AGENTS.md: Instruction files at the top of a repository that coding assistants read before working: Claude reads CLAUDE.md, Codex reads AGENTS.md.
Clone: A full copy of a repository on another computer. Builders working in the browser don’t need one.
Code: The written instructions a computer follows to run a tool. Assistants can write it for you.
Commit: One saved set of changes, with a short note saying what changed.
Conflict: When two changes edit the same lines and Git can’t tell which to keep, so a person or assistant has to choose.
Diff: The exact lines that changed between two versions, shown side by side, often as +added and −removed.At FirstLine: Reading the diff is how you review what an assistant changed.
Git: The system that records every change to a set of files, who made it and when, so any earlier version can be brought back.
GitHub: The website where Git repositories are kept and shared, and where changes are reviewed.At FirstLine: FirstLine’s code lives in the firstline-schools organization on GitHub.
GitHub Actions, or workflow: GitHub’s way of running steps automatically, such as tests on a pull request or a deploy after a merge.
GitHub Pages: GitHub’s free hosting for static pages from a repository. On FirstLine’s plan the page is always public.At FirstLine: This guide is hosted on GitHub Pages.
Issue: A note filed on a repository to report a problem or ask for something.
Main, or the default branch: The branch that holds the real, current version. At FirstLine, merging into main is what makes a change go live.
Merge: Bringing a branch’s changes into another branch. Merging into main publishes the change.
Organization: A shared GitHub account that owns many repositories, so code belongs to FirstLine rather than to one person.At FirstLine: firstline-schools. Joining it takes a seat on FirstLine’s plan.
Public or private repository: Who can read the code. Public: anyone on the internet. Private: only people given access.At FirstLine: This guide’s repository is public; tool repositories are private.
Pull request, or PR: A request to merge one branch into another. It shows the diff, runs the checks and records the review.
README: The file that explains a repository: what it is, who owns it, how to use it.
Repository, or repo: One project’s folder of files on GitHub, with its whole history of changes.At FirstLine: Each tool gets its own repository, such as builders-guide or health-office.
Seat: One paid user on a plan, such as FirstLine’s GitHub plan.
Squash merge: A merge that combines all of a pull request’s commits into one, keeping main’s history tidy.
Template repository: A repository made to be copied as the starting point for new ones.At FirstLine: The starter kit, fls-starter-kit, is a template.
Test: A small piece of code that confirms the tool still does what it promised, run automatically on every change.

Running a tool: Google Cloud

API: A way for one program to ask another for data or actions. API stands for application programming interface.
Artifact Registry: Google Cloud’s storage for container images.
Cloud Build: The Google Cloud service that builds container images from code.
Cloud Run: The Google Cloud service that runs a container as a web app and handles its web address.At FirstLine: Every starter-kit tool runs on Cloud Run.
Cold start: The short wait the first time a scaled-to-zero tool is opened, while a copy starts up.
Container: A sealed package holding a tool’s code and everything it needs to run, so it runs the same way on any computer.
Deploy: Putting a new version of a tool live. At FirstLine a merge to main deploys automatically.
Dockerfile: The recipe that says how to build a tool’s container image.At FirstLine: In the starter kit it lists every file the tool needs, one by one.
Endpoint, or route: One address on a web app that does one thing, such as /api/health.
Environment variable: A setting handed to a tool when it starts, so the same code can run in different places.
Front end and back end: The front end is what people see in the browser: the page. The back end is the part on the server that checks who you are, reads data and saves entries.
Google Cloud, or GCP: Google’s service for running software and storing data on Google’s computers. GCP stands for Google Cloud Platform.
Health check: An automatic test that a tool is up and answering.At FirstLine: Every starter-kit tool is checked each morning.
HTML, CSS and JavaScript: The three languages of a web page: HTML for content, CSS for look, JavaScript for behavior.
Image: The saved, ready-to-run version of a container, built from the code.
Logs: The running record a tool writes as it works: what happened, when, and any errors.
Project (Google Cloud): A container in Google Cloud that holds a set of services, data and permissions, and its bill.At FirstLine: firstline-tools holds the builder tools; fls-data-warehouse holds FirstLine’s data.
Python and Flask: Python is a programming language; Flask is a small Python framework for web apps.At FirstLine: The starter kit is a Flask app.
Region: Which of Google’s data centers runs a service.At FirstLine: FirstLine uses us-central1.
Revision: Each version of a Cloud Run service; rolling back means sending traffic to an earlier revision.
Rollback: Putting the previous version back after a bad change.
Scale to zero: Running no copies of a tool while nobody is using it, so it costs almost nothing when idle.
Scheduled job: Something set to run on a timer, such as a nightly data refresh (Cloud Scheduler on Google Cloud, a trigger in Apps Script).
Secret, or Secret Manager: A password or key the tool needs, kept in Google’s locked store rather than in the code.
Server: A computer that runs a tool all the time, waiting for people to use it.
Service: One running tool on Cloud Run, with its own address.
Static site and web app: A static site shows the same thing to everyone. A web app has a back end, so it can know who you are and remember things.

Data

Authorized, or scoped, view: A view that returns only the rows the person asking may see.At FirstLine: BigQuery Student Data uses these, so each person sees only their own students.
BigQuery: Google Cloud’s database for large amounts of data and analysis: counts, trends, comparisons.
Cloud SQL, or Postgres: A traditional database of related tables, run by Google Cloud. Postgres is the database software.
CSV: A plain text file of rows and columns that almost any tool can open.
Data warehouse: One central database where an organization’s official data is gathered from its source systems.At FirstLine: FirstLine’s is fls-data-warehouse, in BigQuery.
Database: An organized store of information that a tool can search and update.
Dataset: A named group of tables in BigQuery.At FirstLine: Each starter-kit tool gets its own dataset.
FERPA: The U.S. law that protects students’ education records.
Firestore: Google Cloud’s database for live app records that people add and edit often.At FirstLine: Being tested at FirstLine as a second option for a tool’s own data.
PII: Personally identifiable information: names, IDs, birthdates, anything that points to a particular person.
Reserved word: A word the database keeps for its own use. Naming a column with one can break every save.At FirstLine: A column named at once broke a FirstLine tool.
Row and column: A row is one record, such as one student or one response; a column is one kind of fact about it, such as a date.
Schema: The list of a table’s columns and what kind of value each holds: text, number, date.
Source of truth: The one place the real answer lives, which every other copy should match.
SQL and query: SQL is the language for asking a database for data; a query is one such question.
Sync, or load: Copying data from one system into another, often on a schedule.
Table: Data in rows and columns, like one tab of a spreadsheet.
View: A saved query that looks like a table but is calculated fresh each time.

Sign-in and access

API key: A secret code that lets a program use a service. It goes in Secret Manager, never in code.
Authentication, or auth: Proving who you are, usually by signing in.
Authorization: Deciding what you may see or do once the tool knows who you are.
Execute as: An Apps Script web app setting: run as the person who built it (every viewer sees what the builder can see) or as each viewer.
IAM, roles and permissions: IAM (identity and access management) is Google Cloud’s list of who may do what. A role is a named bundle of permissions.
IAP: Identity-Aware Proxy: Google checks the FirstLine sign-in before anyone reaches a tool.At FirstLine: Starter-kit tools sit behind IAP by default.
Least privilege: Giving each person or program only the access it needs, and nothing more.
OAuth: The standard behind “Sign in with Google”: you approve access in Google’s own window, and the tool never sees your password.
Service account: An account for a program rather than a person, with its own narrow permissions.At FirstLine: Each starter-kit tool runs as its own service account.
Sharing setting: The choice of who can open a file, page or link: only you, named people, everyone at FirstLine, or anyone.
SSO: Single sign-on: one account opens many tools.At FirstLine: Your FirstLine Google account is FirstLine’s sign-on.
Two-factor authentication, or 2FA: A second check after your password, such as a code on your phone.At FirstLine: Turn it on for your GitHub account.
Workload Identity: A way for GitHub to deploy to Google Cloud without any stored key: Google trusts GitHub’s proof of which repository is asking.

Google Workspace tools

Apps Script: Google’s built-in code for Workspace, opened from a Sheet with Extensions → Apps Script. It can send email, run on a timer, and turn a Sheet into a page.
Google Forms and Sheets: Forms collect answers; each answer lands as a row in a linked Sheet.
Google Sites: Google’s no-code website builder.
Google Workspace: Google’s apps for work and school: Gmail, Drive, Docs, Sheets, Forms, Sites and more, signed in with your FirstLine account.
Looker Studio: Google’s free tool for charts and reports over a Sheet or database.
Trigger: An Apps Script setting that runs code on a schedule or when something happens, such as a form being submitted. It runs as the person who made it.
Web app (Apps Script): An Apps Script published as a page with its own address ending in /exec.
Workspace Studio: Google’s no-code tool for automating routine tasks across Gmail, Sheets and Drive with Gemini.

How we work

Acceptance criteria: Checks a reviewer can watch happen to confirm the tool works.
Adversarial pass: Asking a second AI to try to break a tool’s rules, such as seeing another school’s data.
Mock, or fake data: Invented data used for testing. It must not contain real students or staff.
Must Never list: The things a tool must never do, written down so they can be checked.
Safe to Ship: AI for Equity’s two-part review before a tool is relied on: one guide for the builder and one for the supervisor.
Spec, or plan: A short written description of what a tool does, who uses it and what each person may see, written before building.
Version: One saved state of a tool. Every merge makes a new one.

← Back to the Builders Guide