Claude Glossary

Explanation

A token is one of the small pieces of text a language model processes internally. Usage limits, pricing, and context size are usually measured in tokens rather than words.

Key notes

• Input and output both consume tokens.
• Words, punctuation, and code do not map cleanly to word count.
• Large files or long prompts can use a lot of tokens quickly.

Explanation

The context window is the amount of material Claude can actively consider in a single interaction. It includes your prompt, attached content, relevant chat history, and room for the reply.

Key notes

• It is short-term working context, not permanent memory.
• Bigger context windows help with long documents and complex tasks.
• When the limit is approached, older or less important material may be summarized or dropped.

Explanation

A model is a particular Claude variant with its own balance of speed, cost, and capability. Choosing a different model can change how well it handles coding, reasoning, or lightweight chat tasks.

Key notes

• Smaller models are often faster and cheaper.
• Stronger models tend to do better on difficult reasoning or generation tasks.
• Good prompts still matter no matter which model you choose.

Explanation

Extended thinking means the model spends more effort working through a problem before giving the final answer. It is most useful when the task needs planning, careful logic, or multi-step analysis.

Key notes

• Helpful for debugging, planning, math, and difficult comparisons.
• Usually slower than a lightweight response.
• High-stakes answers should still be verified.

Explanation

A hallucination is an incorrect or invented answer presented as though it were true. This is a known failure mode for language models, especially when information is missing, ambiguous, or highly specific.

Key notes

• Confident tone is not proof of accuracy.
• Ask for sources, evidence, or uncertainty when accuracy matters.
• Hallucinations are more likely when the model must guess.

Explanation

A prompt is the input that tells Claude what you want. Strong prompts give clear goals, useful context, constraints, examples, and the preferred output format.

Key notes

• Be explicit about the task, audience, and desired format.
• Examples usually improve quality and consistency.
• Revising the prompt is often the fastest way to improve the result.

Explanation

The system prompt is a higher-priority instruction layer set by the developer or platform. It can define rules, tone, tool usage, boundaries, and product behavior that the end user does not normally see.

Key notes

• It usually has more authority than a normal user prompt.
• It helps keep behavior consistent across many conversations.
• It is commonly used for safety, formatting, and workflow control.

Explanation

Temperature changes how varied or conservative the model's wording tends to be. Lower values usually make responses more stable, while higher values allow more variation and creative exploration.

Key notes

• Lower settings are often better for structured or factual tasks.
• Higher settings can help with brainstorming and creative writing.
• Changing temperature does not guarantee correctness.

Explanation

Claude.ai is the primary chat interface where most users start. It is the standard web-based experience for asking questions, drafting content, analyzing files, and using Claude in a general-purpose way.

Key notes

• Usually the easiest entry point for new users.
• Best for general chat, writing, and everyday assistance.
• Useful when you want the standard browser or app experience.

Explanation

Claude Desktop is the installed desktop application version of Claude. A desktop app can fit more naturally into daily work and may provide tighter links to local files and system-level workflows.

Key notes

• Good for people who prefer a native app over a browser tab.
• Often better suited to persistent work sessions.
• The source image positions it as the gateway to Cowork and Code.

Explanation

Cowork is presented here as a hands-on desktop agent that can operate with your real files and applications, not only chat text. It is meant for practical multi-step work in your environment.

Key notes

• Useful for workflows that span more than one app or file.
• Permissions and guardrails matter because real systems are involved.
• Best for repeatable, execution-oriented tasks.

Explanation

Claude Code is the developer-focused version of the product, designed for coding workflows inside terminals and editors. It is geared toward tasks like writing, refactoring, debugging, and explaining code.

Key notes

• Best suited to programming and repository work.
• Useful for debugging, generation, and code review support.
• Human review is still important before shipping changes.

Explanation

This refers to a browser-based integration that can use the page you are already viewing as context. Instead of copy-pasting everything into chat, Claude can work with what is on the page directly.

Key notes

• Helpful for web research, summaries, and page-based actions.
• Permissions and page access should be understood clearly.
• Most useful when the task is tied to what you are browsing.

Explanation

This is a spreadsheet integration that brings Claude into Excel-style work. It can help explain formulas, analyze patterns, summarize tables, and assist without moving data back and forth as much.

Key notes

• Useful for cleanup, summaries, and formula assistance.
• Reduces copy-paste friction between chat and spreadsheets.
• Important calculations should always be checked independently.

Explanation

This is a slide-building integration for presentation work. It helps with outlining decks, rewriting bullets, improving clarity, and turning ideas into draft slides more quickly.

Key notes

• Great for first drafts, rewrites, and structure.
• Can speed up presentation creation significantly.
• Design choices and factual content still need review.

Explanation

Claude Marketplace is shown as an ecosystem hub for add-ons and reusable capabilities. It is where users can discover ways to extend Claude beyond its base interface.

Key notes

• Useful for expanding Claude with integrations and packaged workflows.
• Different add-ons may vary in quality, access, and safety.
• Best approached like an app directory: install only what fits the job.

Explanation

An artifact is a concrete output Claude creates as a reusable deliverable rather than just chat text. That could be a document, code file, chart, small app, or another object you want to review and refine.

Key notes

• Good when the result should feel like a file or deliverable.
• Useful for code, docs, charts, and interactive mockups.
• Keeps the working output separate from the conversation itself.

Explanation

A project groups all the material for one body of work into a single workspace. It helps Claude keep the right files, history, and instructions together over time.

Key notes

• Best for ongoing work with recurring context.
• Helps reduce repetition from one session to the next.
• Makes it easier to separate one initiative from another.

Explanation

Memory refers to persistent facts or preferences that Claude can carry across separate conversations, depending on product behavior and settings. The goal is to make future interactions feel more informed and personalized.

Key notes

• Most helpful for stable preferences, recurring needs, or personal defaults.
• Should not replace checking important one-off details.
• Users should be able to review or manage what is remembered.

Explanation

Deep research is a more involved research flow where Claude gathers information from multiple sources and synthesizes it into a structured answer or report. It is broader and slower than a single quick lookup.

Key notes

• Useful for comparisons, background briefs, and literature-style summaries.
• Best when the result should cite and synthesize multiple sources.
• Source quality still matters, especially for high-stakes topics.

Explanation

Web search is the ability to fetch current online information instead of relying only on training data. It is especially useful for recent events, changing facts, and direct verification.

Key notes

• Helpful when freshness matters.
• Trust depends on the credibility of the sources used.
• A quick search is lighter than a full research workflow.

Explanation

Style is the way Claude adapts to a preferred voice, tone, structure, and phrasing pattern. That can mean sounding more formal, concise, friendly, brand-aligned, or similar to examples you provide.

Key notes

• Works best when you give reference examples.
• Style affects presentation, not whether the content is true.
• Useful for keeping writing consistent across a team or project.

Explanation

User preferences are saved defaults that shape how Claude responds before you even write the next prompt. They help reduce repetitive setup by applying your preferred tone, structure, and response style automatically.

Key notes

• Great for default formatting rules and tone choices.
• Helps make responses feel more consistent over time.
• Worth revisiting when your needs or workflows change.

Explanation

Compaction is a compression step for long conversations. Instead of keeping every earlier message verbatim, the system summarizes older material so the chat can continue while preserving the most important context.

Key notes

• Useful for long-running threads that would otherwise exceed context limits.
• Minor details may be lost during summarization.
• Restate mission-critical constraints if they must never be missed.

Explanation

A connector is an integration bridge that lets Claude access data or actions in another application. Connectors are what make assistants more useful in real workflows instead of staying isolated in a chat box.

Key notes

• Common targets include email, docs, chat tools, and knowledge bases.
• Permissions determine what Claude can read or do.
• Good connectors are essential for practical agents.

Explanation

MCP, or Model Context Protocol, is described as the interoperability standard behind connectors. In plain language, it is the rule set that helps models receive structured context and tool access in a consistent way.

Key notes

• It reduces the need for one-off custom integrations.
• It helps standardize how tools expose context to models.
• Useful wherever agents need reliable access to external systems.

Explanation

A skill is a reusable instruction package for doing a task the same way every time. Think of it as a workflow recipe that captures steps, expectations, and best practices in a form Claude can follow.

Key notes

• Best for repeated jobs that should be handled consistently.
• Can include steps, examples, templates, and checks.
• Well-written skills improve output quality and predictability.

Explanation

The source image shows this as an unlabeled item, so it is preserved here as “Bundled package.” It appears to mean a higher-level bundle that groups several reusable pieces together for a complete role-based workflow.

Key notes

• Combines multiple capabilities into one role-oriented starter kit.
• Can package skills, integrations, and shortcuts together.
• The original infographic did not show a visible subject label for this line.

Explanation

A slash command is a short trigger phrase that launches a predefined workflow or instruction set. It makes repeat tasks faster because users do not have to rewrite the same prompt every time.

Key notes

• Excellent for common repeatable actions.
• Helps standardize team workflows and naming.
• Commands work best when they are easy to remember.

Explanation

A scheduled task is automation that runs on a defined cadence without needing a new manual prompt every time. Examples could include summaries, reports, checks, or reminders.

Key notes

• Good for recurring work with a clear rhythm.
• Timing, error handling, and delivery destinations matter.
• The simpler and more deterministic the job, the better.

Explanation

Dispatch is described here as a remote-control layer for Cowork, letting you start or manage work from a phone. The idea is that you can trigger or oversee desktop-agent tasks while away from the main machine.

Key notes

• Useful for lightweight control when mobile.
• Good status visibility becomes especially important here.
• Sensitive actions should still preserve approval safeguards.

Explanation

In the source image, computer use is framed as persistent background context that Cowork reads at the start of each session. In practical terms, it sounds like a standing operating brief for the agent so you do not repeat the same setup instructions.

Key notes

• Helps reduce repeated setup and explanation.
• Useful for stable rules, assumptions, and environment context.
• Those background instructions should be reviewed and kept current.