Glossary

You do not need to speak the language of AI to use Claude well on a live deal, but a handful of terms come up often enough that it helps to have them straight. This page explains, in plain English, the words you will meet across this wiki and inside Claude itself. No jargon is left to guess at. If a term in another page is unclear, it is almost certainly defined here. The entries are grouped loosely by theme, but you can read them in any order.

How Claude is organized

Project : A dedicated workspace inside Claude for a particular piece of work, holding its own files, instructions, and conversations in one place. You use one for a live deal or for a function you repeat across deals, so context stays clean. See What Projects Are.

Custom Instructions : Standing guidance you set once for a Project, telling Claude how to behave every time, such as who it is assisting, the tone to use, and the standards to apply. You write them once instead of repeating yourself in every prompt. See Writing Effective Instructions.

Project Knowledge : The documents and reference material you load into a Project so Claude can draw on them across every conversation in that space. For a deal, this is where the data-room material and your reference notes live. See The Two-Layer System.

Custom Style : A saved way of writing that you can apply to Claude’s output, so it consistently matches a tone or format you prefer (for example, the measured register of an investment committee paper) without re-explaining it each time. See Writing Effective Instructions.

What Claude is and how it thinks

Model : The underlying engine that produces Claude’s answers. Your firm’s plan gives you access to capable models in the Claude family; Opus is the most capable, suited to complex reasoning and careful analytical work. You select it in the interface; you do not need to understand how it works internally.

Context : Everything Claude can “see” for a given task: your prompt, the files you have attached, the Project knowledge, and the conversation so far. The quality of the context largely determines the quality of the answer. See Why Context Is Everything.

Context Window : The practical limit on how much Claude can hold in mind at once, across all your documents and the running conversation. It is large, but not infinite. Very large or numerous documents can fill it, which is one reason to keep each Project focused on the work in hand.

Token : The small unit of text that models read and write in, roughly a word or part of a word. You rarely need to think about tokens directly; the idea matters only because the context window is measured in them, which is why “how much can it hold” has a ceiling.

Extended Thinking / Effort : A setting that lets Claude reason for longer before answering, trading a little speed for more careful, thorough analysis. Use higher effort for complex, high-stakes tasks and lower effort for quick, simple ones. See Thinking & Effort.

Hallucination : When Claude produces something that sounds confident and plausible but is not actually true or supported by the source. This is the central reason to verify output: fluent does not mean correct. See Verify Everything.

Working with Claude

Prompt : The instruction or question you give Claude. A good prompt is a clear brief: who Claude is helping, what you want, how you want it back, and what to watch for. See Anatomy of a Good Prompt.

Artifact : A self-contained piece of work that Claude produces in its own panel beside the conversation, such as a drafted document or a structured table, which you can read, refine, and reuse without it getting lost in the chat. See Artifacts.

Analysis Tool : A capability that lets Claude work through data and calculations more reliably than writing them out in prose, useful when you need it to handle figures carefully rather than describe them. You still verify the result. See The Analysis Tool.

Connector : A link that lets Claude reach an approved external source, such as a document store, so you do not have to upload everything by hand. Connectors are switched on or off centrally by your IT team, not by you. See Connectors.

Web Search : A capability that lets Claude look up current, public information from the open web when a task needs it, rather than relying only on what it already knows. Treat what it finds as you would any web source: useful, but to be checked. See Web Search.

Running Claude in your tools

Terminal : The text-based window where you type commands to your computer. Claude Code runs in any terminal, and VS Code’s built-in terminal is the most convenient option. See Before you Start.

Command Line (CLI) : A way of controlling a program by typing text commands instead of clicking buttons. See Claude Code vs claude.ai.

VS Code : A free code editor from Microsoft, and the recommended environment for Claude Code because it puts your files, editor, and terminal in one window. See Before you Start.

Repository : A folder holding a project and its full history of changes, often shortened to “repo”. See Saving Work & Going Live.

Commit : A saved snapshot of your project at a point in time, with a short note describing what changed. See Saving Work & Going Live.

Deploy : To publish your work so it becomes a live, shareable link others can open. See Saving Work & Going Live.

CLAUDE.md : A plain-text file in your project that tells Claude Code the standing rules and context for that project. See CLAUDE.md Rulebook.

Your account and plan

Enterprise / Team plan : The plans your firm provides, and the appropriate place for professional work. Their defining property is that your prompts and uploaded files are not used to train the model, unlike consumer (Free and Pro) plans. See Your Account & Confidential Data.

A few terms used loosely elsewhere

Conversation (or chat) : A single running thread with Claude. It remembers what has been said within it, which is why refining a draft inside one conversation works better than starting a new one each time.

Inference : A conclusion Claude reasons to, rather than one stated directly in your source. Inferences can be valuable, but they are yours to accept or reject, not facts to repeat. Asking Claude to label its inferences is a good habit. See Anatomy of a Good Prompt.

Verification : The discipline of checking Claude’s output against the original source before relying on it: every figure, quote, and claim. It is the professional habit that makes everything else safe to use. See Verify Everything.

Where to go next

If a term sent you here from another page, head back and carry on. If you are browsing, the natural next stops are the Prompt Library for ready-made prompts, a task-specific playbook such as CIM Analysis or Due Diligence, and the Cheat Sheet for the whole approach on one page.