Portal Sign In
Back to All Cheatsheet Libraries cheatsheets

Anthropic Claude

The claude.ai consumer app — Projects, Artifacts, memory, the current model lineup and plan differences. For the terminal coding agent, see Anthropic Claude Code.

This page is claude.ai. For the CLI, see Claude Code

Anthropic makes two genuinely different products under the same name. This page covers the consumer app — the web, desktop and mobile chat interface at claude.ai. For the terminal coding agent, see Anthropic Claude Code — the two share a model family but almost nothing else.

Surface Notes
Web (claude.ai)The full feature set lands here first. Everything else follows.
Desktop appmacOS and Windows. Adds desktop extensions and a persistent window, otherwise mirrors the web app.
Mobile appiOS and Android. Voice input, camera capture, and the same Projects/Artifacts as web.
Chrome extensionClaude in Chrome — reads and acts on the page you're viewing.
Claude ConsoleDeveloper surface at console.anthropic.com — API keys, usage, and the Workbench playground. Separate billing from claude.ai plans.

Projects

A self-contained workspace with its own knowledge base

A Project bundles its own chat history, uploaded documents, and custom instructions — separate from your general conversation history. Upload the documents once, and every chat in the project can reference them.

What differs by plan
  • Free accounts get up to 5 projects. Paid plans (Pro, Max, Team, Enterprise) get unlimited.
  • Enhanced RAG is paid-only. When a project's knowledge approaches the context limit, RAG mode expands effective capacity up to 10x. Free accounts don't get this — a large knowledge base degrades instead.
  • Sharing is Team/Enterprise only. Two permission levels: Can view (read-only, can still chat) and Can edit (instructions, knowledge, and members).
  • Custom instructions apply project-wide — set a tone or role once rather than repeating it every chat.

Artifacts

A separate window for anything substantial and self-contained

Claude opens an Artifact automatically for content over roughly 15 lines that you're likely to want to edit, iterate on, or reuse — code, a full app, Markdown, an SVG, a diagram, a React component, a single-page site. No special command needed; it triggers on the content, not a keyword.

Editing, versions, and publishing
  • Ask for a change in chat and Claude edits the Artifact in place, keeping version history you can step back through.
  • For Markdown documents specifically: highlight text and choose "Edit with Claude" for inline editing.
  • Editing an Artifact does not change what Claude remembers about the original — you can branch by editing an earlier message instead of the newest version.
  • Artifacts are private until you click Publish. Publishing adds it to your public Artifacts section, makes it embeddable, and makes it discoverable by others.
  • AI-powered Artifacts — ones with Claude intelligence embedded — can be shared for free; usage counts against the viewer's own plan limits, not the creator's.

The current generation

Model names and capabilities change often — this reflects Anthropic's own current-generation listing. Verify against the live models page before building anything that depends on a specific one staying available.

Model Positioning
Claude Fable 5Next-generation intelligence, built for long-running agents.
Claude Mythos 5Shares Fable 5's capabilities without the safety classifiers. Limited release, through Project Glasswing — not generally available.
Claude Opus 5Complex agentic coding and enterprise work.
Claude Sonnet 5Frontier intelligence at scale — coding, agents, and enterprise workflows. The default workhorse for most tasks.
Claude Haiku 4.5Fastest model with near-frontier intelligence. Right choice when latency or cost matters more than the last few points of capability.
Two ways to build with Claude
  • Messages API — direct model prompting. Full control over the agent loop; the right choice when you're building your own orchestration.
  • Claude Managed Agents — a pre-built, configurable agent harness that runs in Anthropic's own managed infrastructure. Suited to long-running and asynchronous tasks rather than a tight request/response loop.
  • Both are Console/API concepts — neither is something you pick from claude.ai's own chat interface, which always routes through whichever model you select from the model switcher.

Memory saves topics as you go, not a transcript

Claude doesn't store a word-for-word record of your chats. As you talk, it extracts and saves individual topics — your role, recurring projects, communication preferences — and pulls on them in later conversations. It's a running set of facts, not a searchable archive of every message.

Feature Plan access Notes
MemoryOn by default: Free, Pro, Max. Off by default: Team, Enterprise (admin can enable org-wide).Settings > Memory. Saves individual topics as they come up in conversation.
Chat searchPaid plans only — Pro, Max, Team, Enterprise.Ask "what did we discuss about X" and Claude searches past chats via RAG, shown as a visible tool call.
Per-project memoryAll plans with Projects.Each project keeps its own memory and summaries, isolated from your general chats and other projects.
Incognito chatsAll plans.Ghost icon when starting a chat. Nothing from it feeds memory or becomes searchable.
Controlling what gets remembered
  • Settings > Memory — toggle "Generate memory from chats" off entirely, or manage what's already stored.
  • Turning it off gives two choices: Pause memory (keeps existing entries, stops new ones) or Reset memory (permanently deletes everything — irreversible).
  • Say "remember this" in any chat to save something directly, or tell Claude what to forget — the change applies from the next conversation.
  • View, edit, or delete individual stored topics from the Memory settings page.
  • Sensitive topics are excluded by default — health and political beliefs won't be saved unless you separately enable "Include sensitive topics in memory."
  • Government IDs, financial account numbers, and immigration status are never saved, regardless of any setting.

Prices and limits move fast — check before you rely on a number here

This is a snapshot. Verify current pricing and limits at claude.com/pricing before making a purchasing decision.

Plan Price What's exclusive to it
Free$0Web/mobile chat, memory, up to 5 projects. No Claude Code included. Reduced weekly usage limits.
Pro~$17–20/moAdds Claude Code, Cowork, Design and Science. Unlimited projects, enhanced RAG, higher usage limits.
Max 5x / 20xFrom ~$100/mo5x or 20x Pro's usage limits. Priority access during peak load, early access to new features.
Team~$20–125/seat/moCentral billing, admin controls, SSO, enterprise search. Standard and Premium (5x usage) seat tiers.
Enterprise~$20/seat + usageSCIM, audit logs, a HIPAA-ready option, custom data retention, role-based access, IP allowlisting.
The plan trap worth knowing
  • Claude Code is not included on the Free plan. If you've been told to "just use Claude Code" and only have a free claude.ai account, you need Pro or above first.
  • Console/API billing is entirely separate from claude.ai subscriptions. A Pro plan doesn't give you API credits, and API usage doesn't count against your claude.ai chat limits.
  • Team's Premium seat and the Max tiers both multiply usage limits rather than unlocking new features — read carefully before assuming a price jump buys capability rather than just headroom.

Gotchas

Things that catch people
  • Claude.ai and the Console are different accounts in practice — different billing, different usage pools. Signing up for one doesn't give you the other.
  • Memory doesn't cross projects. A fact Claude learned in one project won't surface in a different project or your general chats — that isolation is deliberate, not a bug.
  • Free accounts cap at 5 projects. Hitting the limit means archiving or deleting one before creating another.
  • An Artifact's version history is not the same as chat memory. Rolling back an Artifact to an earlier version doesn't undo what Claude has since learned about your intent.
  • Incognito chats leave zero trace — no memory, no search, and nothing recoverable afterward. Don't use it for something you might need to find again later.
  • Claude Mythos 5 is not generally available. It ships without Fable 5's safety classifiers and is limited-release through Project Glasswing — most accounts will never see it in the model switcher.
  • The model switcher in claude.ai shows only models your plan permits — Free accounts see a reduced set at reduced limits.

Tips

Start a Project before a long thread

If you'll return to a topic across many sessions, a Project with the source documents uploaded beats re-pasting context into a fresh chat every time.

Say "remember this" explicitly

Don't rely on automatic extraction for something important. Stating it directly is faster to verify and guaranteed to stick.

Use chat search before starting over

On a paid plan, ask what you discussed on a topic before re-explaining it — it's often faster than you'd expect, and keeps context consistent across sessions.

Incognito for anything sensitive or one-off

A quick question you don't want shaping future memory — drafting something for someone else, a one-time lookup — belongs in an incognito chat.

Publish an Artifact instead of screenshotting it

A published Artifact is a real, embeddable link. Far more useful to hand someone than a screenshot of code they'll have to retype.

Need the CLI instead?

None of this — Projects, Artifacts, memory — exists in Claude Code. See Anthropic Claude Code for the terminal agent and its 88 slash commands.

Resources

claude.ai The app itself. claude.ai Anthropic Help Center Official support articles — Projects, Artifacts, memory, billing. support.claude.com Models overview The authoritative, current model list and capabilities. platform.claude.com Pricing Current plan prices and limits — verify before deciding. claude.com Claude Console API keys, usage, and the Workbench playground. Separate billing from claude.ai. console.anthropic.com Anthropic Claude Code The terminal coding agent — a different product sharing only the model family. www.ty1er.com

Compiled from Anthropic's official documentation and Help Center, August 2026. Model names, plan pricing, and usage limits change frequently — verify against the live pages before relying on a specific figure. This page covers claude.ai, the consumer app; for the CLI coding agent, see Anthropic Claude Code.