Claude Chat

Last verified: 14 April 2026 | Applies to: All plans

In 30 seconds

Claude Chat is the conversation interface. With skills enabled and tools connected, you can build spreadsheets, manage tasks in Asana, pull data from Notion, create slide decks, render interactive artifacts, and use extended thinking for complex analysis. Projects give you persistent context for recurring work.

What it does

Core capabilities:

• Build and edit spreadsheets, documents, slide decks, and PDFs
• Read from and write to Asana, Notion, Linear, Slack via connectors
• Query databases, build visualisations, run analyses
• Create artifacts: interactive components that render directly in chat

Artifacts are visual outputs Claude creates inline: charts, tables, interactive elements, code previews. They render in a panel alongside the conversation, so you can see the result while continuing to refine it.

Extended thinking lets Claude work through complex problems step by step before responding. Useful for financial analysis, strategic planning, and any task where you want Claude to reason carefully rather than respond quickly. Toggle it on when accuracy matters more than speed.

Projects are persistent context spaces for recurring work. Create a project for “Monthly Reporting” or “Q1 Planning” and Claude retains the context between sessions: your team structure, key metrics, and previous outputs.

How to set it up

1. Enable skills: Settings → Capabilities → toggle on relevant skills. Start with document creation skills (Excel, Word, PowerPoint, PDF).
2. Connect tools: Settings → Connectors → browse the directory → authenticate with OAuth.
3. Try a task: Ask Claude to create something you’d normally build manually: a weekly report template, a competitor comparison table, a project status update.

Key features in depth

Projects

Projects are persistent workspaces that carry context across multiple conversations. They solve the biggest frustration operators have with chat: having to re-explain everything each time.

How to create a project:

1. Open claude.ai and click Projects in the sidebar
2. Click New Project and give it a name (e.g., “Q1 Planning”, “Monthly Reporting”, “Client Onboarding”)
3. Add a project description. This is context Claude reads at the start of every conversation within the project
4. Upload project files: documents, spreadsheets, reference materials that Claude should know about
5. Start a conversation inside the project

What persists between conversations:

• The project description (your standing instructions and context)
• Uploaded project files (Claude can reference these in any conversation)

What does not persist:

• Previous conversation content. Each conversation within a project is independent. Starting a new chat does not give Claude access to prior chat threads. If you want Claude to reference a previous output, upload it as a project file or paste it in.
• Tool connections. Connectors still need to be enabled at the account level

Memory vs. project history

The built-in Memory feature (available on all plans including Free since 2 March 2026) synthesises context across conversations over time, including your preferences, team details, and recurring topics. A memory import tool also lets you bring context from competing platforms (ChatGPT, Gemini). But Memory is not the same as Claude having access to your past chat transcripts. Memory captures patterns and key facts, not full conversation history.

When to use Projects vs. regular chat:

• Use Projects for recurring work: weekly reports, ongoing client accounts, quarterly planning, any topic you return to repeatedly
• Use regular chat for one-off questions, quick tasks, and anything you won’t revisit

Project tip

Seed your project description with the context you’d normally type at the start of every conversation: your role, team members, key acronyms, preferred formats, and any standing rules. This alone saves minutes per session.

Extended thinking

Extended thinking makes Claude pause and reason through a problem before responding. Instead of generating an answer immediately, Claude works through the logic step by step, and you can see the thinking process.

Adaptive thinking (Opus 4.6 and Sonnet 4.6): Both Opus 4.6 and Sonnet 4.6 use adaptive thinking. Rather than requiring you to manually toggle thinking on or off, Claude dynamically decides when and how much to think based on task complexity. Simple questions get a fast response; complex analysis triggers deeper reasoning automatically. You can still influence thinking effort through the model selector, but the default behaviour is now adaptive.

Haiku 4.5: Does not have adaptive thinking. For Haiku, the manual extended thinking toggle applies if available.

When extended thinking helps most:

• Complex financial analysis where accuracy matters (variance explanations, reconciliation logic)
• Multi-step reasoning tasks (comparing options across several criteria, evaluating trade-offs)
• Strategic planning that requires weighing multiple factors
• Debugging or troubleshooting where Claude needs to work through possibilities systematically
• Any task where you’ve noticed Claude giving superficial answers and you want depth

What the output looks like: When extended thinking is active, you see two sections in the response: a collapsible “thinking” block showing Claude’s reasoning process, and the final answer below it. The thinking block is useful for verifying Claude’s logic. You can see how it arrived at the conclusion and spot any errors in reasoning.

Cost implications: Extended thinking uses more of your plan’s usage allocation. A response with extended thinking typically consumes 2-5x the usage of a standard response. On Pro plans with limited usage, be selective about when you enable it. On Max plans, the higher allocation makes this less of a concern.

When to leave it off

For straightforward tasks (drafting emails, creating simple documents, answering factual questions), extended thinking adds latency without improving quality. With adaptive thinking on both Opus 4.6 and Sonnet 4.6, Claude generally handles thinking effort on its own. For Haiku 4.5, you may want to manually enable extended thinking for complex tasks.

Artifacts

Artifacts are rich outputs Claude generates directly in the chat interface, rendered in a side panel so you can view the result while continuing the conversation.

What counts as an artifact:

• Code: scripts, functions, and programs with syntax highlighting and a copy button
• Documents: formatted text, reports, and structured content
• Spreadsheets: tables with data that you can review and download
• Visualisations: charts, graphs, and interactive data displays rendered as live components
• Interactive components: HTML/CSS/JavaScript applications that run directly in the panel (calculators, dashboards, interactive tools)
• SVG graphics: diagrams, flowcharts, and visual elements

How to use them:

• Ask Claude to “create” or “build” something visual: Claude automatically generates an artifact when appropriate
• Click the artifact to expand it in the side panel
• Ask for modifications: “make the chart a bar chart instead” or “add a column for variance percentage”: and Claude updates the artifact in place
• Download artifacts as files when you need them outside of chat

Limitations:

• Artifacts render in a sandboxed environment. They cannot make network requests or access external data: everything must be self-contained.
• Complex interactive artifacts may not work perfectly on the first attempt. Iterate with feedback.
• Artifacts are tied to the conversation. If you need to reuse one, download it or copy the code.

Interactive visualisations

Since ~13 March 2026, Claude can create interactive HTML/SVG charts, diagrams, and data visualisations directly inline in chat. Available on all plans including Free. These are different from Artifacts: they render inline in the conversation (not in the side panel) and are temporary. Useful for quick data exploration, visual summaries, and any time you want Claude to show you a chart without downloading a file.

Voice mode

Claude supports two-way voice conversations on claude.ai (web) and the Claude mobile apps (iOS and Android): available on all plans including Free. You speak, Claude listens, and Claude speaks back. This is useful for brainstorming, working through ideas hands-free, or getting quick answers while away from your keyboard.

On Claude Desktop (macOS/Windows), voice is limited to dictation only: press Caps Lock to speak and Claude converts your speech to text, but Claude does not speak back. Full two-way voice is not available on desktop at this time.

File uploads

You can upload files directly into any chat conversation. Claude reads and analyses the content: useful for reviewing contracts, processing data, and working with documents.

Practical limits:

• Up to 30 MB per file
• Up to 20 files per chat message
• Supported formats include PDF, DOCX, CSV, XLSX, images (PNG, JPG, GIF, WebP), and more
• For larger datasets or ongoing reference material, upload files to a Project instead of individual chats

Incognito chats

Incognito chats are available on all plans. When you start an incognito conversation, it is not saved to your chat history and not written to Claude’s Memory. This is relevant for operators handling sensitive data: employee reviews, confidential financial figures, or anything you don’t want persisting in your account history.

Starred conversations and search

You can star important conversations to bookmark them for quick access later. On paid plans, chat search lets you find past conversations by keyword: useful when you need to revisit a workflow or output from a previous session.

How operators actually use it

Build me a variance analysis comparing this month's actuals to budget. Use the numbers from the attached CSV. Format it as an Excel file with conditional formatting: green for favourable, red for unfavourable.

Claude builds the spreadsheet, applies formatting, and delivers it as a downloadable .xlsx file: all in the chat window.

Draft a board update covering Q1 performance. Pull the key metrics from our Notion workspace and format it as a slide deck.

With the Notion connector enabled, Claude reads your data and produces the deck.

What to watch out for

• Context window limits: Even on paid plans, very long conversations may lose earlier details. Start new conversations for new topics rather than running one endlessly.
• Artifact rendering: Some complex visualisations may not render perfectly on first attempt. Iterate: Claude improves with feedback.
• Connector permissions: Each connector requires separate authentication. Claude can only access tools you’ve explicitly connected.

Related

• Skills: the knowledge layer that makes chat smarter
• Connectors: linking Claude to your tools
• Cowork: when you need Claude to act autonomously
• First 30 Minutes: getting started walkthrough

---

Something wrong or outdated? Let us know →

Get weekly workflows. Subscribe to the newsletter.