Claude AI User Guide

What Claude.ai Is

Claude.ai is Anthropic's web and mobile interface for Claude. It runs in your browser, on iOS and Android, and via a desktop app. No coding required -- just sign in and start a conversation.

✨ Key Strengths

📜 Long Context: 200K tokens

Upload entire codebases, long reports, or dozens of documents in a single conversation. Claude reads and reasons across all of it simultaneously -- no chunking required.

✍ Nuanced Writing and Analysis

Claude is widely considered one of the strongest AI models for prose quality, tone matching, persuasive writing, and nuanced document analysis with careful attention to detail.

Core Features

Feature	What it does
Projects	Organize conversations by topic with persistent context, custom instructions, and uploaded files. Claude reads your project instructions at the start of every conversation.
Memory	Added March 2, 2026 (all tiers including Free). Claude remembers key facts about you across conversations. Scope per-project to keep memories separate.
Extended Thinking	Off / Light / Full modes. Claude reasons step-by-step before answering. Best for math, complex decisions, and multi-step debugging. Skip it for simple tasks.
Research mode	Multi-step web research with cited reports (Pro and above). Claude searches, reads sources, and compiles findings into a structured output.
File uploads	Up to 20 files per conversation, 30 MB each. Supports PDFs, Word docs, spreadsheets, code files, images, and more.
Image analysis	Upload photos, screenshots, diagrams, or charts. Claude describes, analyzes, and answers questions about visual content. (Image generation not available on claude.ai.)
Artifacts	Claude renders live previews of code, HTML, and documents alongside the chat. Iterate on the artifact in real time.
Web search	Claude can search the web to ground answers in current information. Included on Free and above.

How to Use Claude in the Browser

Go to claude.ai and sign in (or create a free account)
Start a new conversation or open an existing Project
Type your request in plain language -- no special syntax required
Upload files or images using the paperclip icon
Enable Extended Thinking (brain icon) for complex reasoning tasks
Use Research mode (Pro+) to have Claude search the web and compile a cited report
Iterate in the same conversation -- Claude keeps full context of everything said
Save useful conversations into Projects to build persistent context over time

Connectors and Integrations

Slack: Connect your Slack workspace so Claude can reference channel context
Google Workspace: Connect Drive, Gmail, and Docs (Pro and above)
Microsoft 365: Word, Excel via add-in, and Outlook (Pro and above)
Remote MCP connectors: Connect custom tools and data sources using Model Context Protocol servers
Desktop extensions: Access Claude from any desktop app via the desktop client

Top 8 Claude.ai How-Tos

Practical prompts and workflows for common use cases.

Analyze a long document

Upload a PDF or paste text. Claude reads the whole thing and answers your questions with 200K token context.

"Summarize the key findings of this report, highlight any risks, and list the top 5 action items."

Draft and refine writing

Give Claude a role, audience, and goal. Iterate in the same conversation to sharpen tone and content.

"You are a senior technical writer. Draft a 300-word executive summary for a non-technical audience explaining [topic]. Avoid jargon."

Use Extended Thinking for hard problems

Enable Extended Thinking (Full mode) before asking Claude to reason through a complex decision or analysis.

"Think through the tradeoffs of [Option A vs Option B]. Consider cost, risk, timeline, and long-term impact. Recommend one option with reasoning."

Set up a Project with instructions

Create a Project, add your background context once, and every new conversation in that project starts with that context already loaded.

In Project instructions: "I am a product manager at a B2B SaaS company. We use agile sprints. Always format responses in bullet points with a clear recommendation at the end."

Research a topic with cited sources

Use Research mode (Pro+) to have Claude search the web, gather sources, and compile a structured report.

"Research the current state of [topic]. Include recent developments from the past 6 months, key players, and cite all sources."

Analyze uploaded images or charts

Upload a screenshot, chart, diagram, or photo and ask Claude specific questions about what it shows.

"Look at this chart. What trend is most notable? Are there any anomalies I should investigate?"

Compare multiple documents

Upload several files in one conversation and ask Claude to compare, contrast, or synthesize across all of them.

"I have uploaded three vendor proposals. Compare them on price, timeline, support terms, and technical approach. Recommend the best option."

Get honest feedback on your work

Claude defaults to being helpful and positive. Explicitly ask for critical feedback to get real critique.

"Review this draft. Be critical and direct. Point out every weakness in the argument, any unclear sections, and anything that could be cut. Do not soften your feedback."

Fast Tips for Claude.ai

Use Projects

Set your background context once in Project instructions. Never paste it into every new chat again.

Assign a role

Start with "You are a [role]..." Claude performs significantly better with a clear persona and goal.

Iterate, don't restart

Refine in the same conversation. Claude retains full context and adjusts on follow-up instructions.

Extended Thinking: pick the right mode

Use Light mode for everyday tasks. Reserve Full mode for hard multi-step reasoning. Skip it entirely for simple requests.

Scope your memory

Enable "Memory: Scope per Project" so project memories don't bleed into unrelated chats.

Turn off unused tools

Web search, code execution, and connectors add token overhead even when not invoked. Disable what you don't need.

What Claude Code Is

Claude Code is an agentic coding tool that runs in your terminal, IDE, web browser, or desktop app. Unlike a chat-based assistant, it autonomously reads and edits files, runs tests, manages Git, and works across your entire codebase.

Claude Code

Agentic coding tool · Terminal, IDE, Web, Desktop

How Claude Code works

Install via terminal: curl -fsSL https://claude.ai/install.sh | bash (macOS/Linux/WSL)
Navigate to your project: cd your-project, then run claude
Claude maps the codebase via agentic search, understanding structure and dependencies
Describe your task in plain language -- Claude plans, writes code, runs tests, and verifies
Claude asks permission before editing files or running commands
For CI/CD: runs in GitHub Actions or GitLab CI/CD pipelines

Supported platforms

Platform	Details
Terminal (macOS, Linux, WSL)	curl install or Homebrew cask
Terminal (Windows PowerShell)	irm https://claude.ai/install.ps1 \| iex
VS Code, Cursor, Windsurf	Extensions Marketplace
JetBrains IDEs	JetBrains Marketplace plugin
Web	claude.ai/code
Desktop App	macOS (Intel + Apple Silicon), Windows x64, Windows ARM64

Key capabilities

Capability	Description
Multi-file edits	Coordinated changes across the entire repository simultaneously
Bug fixing	Paste an error or describe a symptom; Claude traces root cause and fixes it
Test writing	Write tests for untested code, run them, and fix failures
Git workflows	Stages changes, writes commit messages, creates branches, opens pull requests
Web search	Searches documentation and Stack Overflow mid-task to fill knowledge gaps
Background agents	Run multiple full sessions in parallel, monitored from one screen
CI automation	Automated code review and issue triage via GitHub Actions / GitLab CI/CD
Slack integration	Mention @Claude in Slack with a bug report; receive a pull request

CLAUDE.md Setup

The CLAUDE.md file is a plain-text instructions file at the root of your project. Claude reads it at the start of every session, giving it persistent context about your codebase.

What to include in CLAUDE.md

Build commands (how to run tests, start the dev server, lint)
Project conventions (naming patterns, file structure, code style)
Known debugging patterns (common errors and their fixes)
What Claude should never do in this codebase (e.g., "never delete migration files")
Key architecture decisions and dependencies
Environment setup notes

A well-written CLAUDE.md dramatically reduces the number of clarifying questions Claude asks and improves the quality of first-attempt solutions.

Context Management Best Practices

Claude Code's precision degrades as the context window fills. Monitor fill level and act before it gets too high.

Context Fill	Behavior	Action
0-50%	Full precision	Work freely
50-70%	Attention warranted	Monitor
70-90%	Precision loss begins	Run /compact
90%+	Hallucinations increase	Run /clear (mandatory)

Rate limits (as of May 6, 2026)

Claude Code uses rolling 5-hour windows, same as Claude.ai
On May 6, 2026, Claude Code 5-hour limits doubled for all Pro, Max, and Teams plans
Peak-hour throttling removed for Pro and Max; limits are now consistent 24/7
Free plan does not include Claude Code

When to Use Claude Code vs. Claude.ai

Use Claude Code when...	Use Claude.ai when...
You need to edit multiple files across a codebase	You want a chat-based coding explanation or review
You want Claude to run tests and verify changes	You need writing, analysis, or document work
You need Git operations (branches, commits, PRs)	You are uploading a file for one-off analysis
You are building a CI/CD automation pipeline	You need research, image analysis, or web search
You want parallel background agent sessions	You are using Projects for ongoing non-code work
Your task requires reading your actual filesystem	The task fits in a single paste or file upload

Claude Code How-Tos

Fix a bug across multiple files

Paste the error message and let Claude trace the root cause across the entire codebase.

"I'm getting this error: [paste error]. Find the root cause and fix it. Run the tests to confirm the fix works."

Add a feature with tests

Describe the feature and ask Claude to implement it, write tests, and confirm they pass.

"Add a [feature description]. Follow the existing patterns in this codebase. Write unit tests and run them before finishing."

Open a pull request

Claude stages changes, writes a commit message, and opens a PR with a description.

"Create a PR for these changes. Write a clear commit message and a PR description that explains what changed and why."

What Claude Enterprise Adds

Claude Enterprise includes everything in Teams, plus a set of security, compliance, and scale features designed for organizations with compliance obligations, large user bases, or sensitive data requirements.

🔒 Enterprise-Exclusive Features

📜 500K Token Context Window

Enterprise supports a 500K token context on supported models (vs. 200K on all other plans). For very large project knowledge bases, Claude automatically switches to RAG-powered retrieval that enables effectively millions of tokens of accessible knowledge.

📋 Zero Training Guarantee

Enterprise contracts include a contractual zero-training guarantee -- Anthropic does not train models on Enterprise prompts or completions. Other paid tiers have this as a default policy, but Enterprise makes it a binding contractual commitment.

Identity and Access Management

Feature	Details
SAML/OIDC SSO	Full single sign-on integration with your identity provider (Okta, Azure AD, Google, etc.)
SCIM provisioning	Automates user provisioning and deprovisioning. When an employee leaves, their access is revoked automatically via your identity provider.
Role-based access	Fine-grained permissioning per user or role. Assigns a primary workspace owner with granular control.
IP allowlisting	Restrict Claude access to corporate network IP ranges. Employees on personal devices or outside VPN cannot access Claude.ai.
MCP server allowlist	Admins publish approved MCP servers and deploy them to developer machines. Unauthorized servers cannot be added by users.

Compliance and Audit

Feature	Details
Audit logs	Full trace of system activities for compliance and security monitoring. Covers authentication, admin actions, and conversation activity.
Compliance API	Real-time programmatic access to conversation content (prompts and completions) for compliance teams. Supports selective deletion. Note: as of May 2026, agentic tool calls are excluded from this coverage.
HIPAA-ready	Available on sales-assisted contracts. Covers Business Associate Agreement (BAA) requirements for healthcare organizations.
OpenTelemetry (OTEL)	Route audit telemetry from Claude for Excel and other agents to your corporate SIEM or observability platforms.

Data and Security Controls

Feature	Details
Custom data retention	Set retention policies that match your organization's legal and compliance requirements. Override Anthropic's default retention period.
Zero training on prompts	Contractual guarantee that Anthropic does not use Enterprise prompts or completions to train models. Not just a default policy -- a binding agreement.
Admin spend controls	Set per-user and per-workspace spending limits to prevent runaway usage or billing surprises.
Claude Security (beta)	Additional security tooling available on Enterprise contracts.

Organization-Scale Features

Feature	Details
Custom org-level system prompts	Deploy organization-wide custom instructions so every user's Claude starts with the same consistent context and guardrails.
Org-wide Skills deployment	Package repeatable workflows (like /review-pr or /deploy-staging) as Skills and share them across the entire organization.
Native integrations	GitHub, databases, CRM systems, project management tools, and development environments.
Bedrock / Vertex AI routing	Enterprise users can route Claude Code against models in existing Amazon Bedrock or Google Cloud Vertex AI instances.

Pricing

Self-serve: $20/seat/month + API usage (annual billing, minimum 20 seats)
Sales-assisted: Contact sales (minimum 50 seats; custom terms, multi-year discounts)
Payment options include invoice, net terms, and ACH -- not just credit card

Who Enterprise Is For

▶

Healthcare organizations

Need HIPAA BAA, zero training guarantee, audit logs, and data retention controls for PHI-adjacent workflows.

▶

Financial services and legal

Require Compliance API for content monitoring, IP allowlisting, SCIM for rapid offboarding, and custom retention policies.

▶

Large engineering teams

Need 500K context for large codebases, org-wide CLAUDE.md Skills, Bedrock/Vertex AI routing, and Claude Code at scale.

▶

Organizations with 150+ users

Teams plan caps at 150 seats. Enterprise has no seat ceiling and includes negotiated pricing for large deployments.

Plan Comparison

All plans include web, iOS, Android, and desktop access. Usage operates on rolling 5-hour windows, not daily resets. Limits vary by message length, model, file attachments, and active tools.

Plan	Price	Usage vs. Free	Models	Claude Code	Key extras
Free	$0	Baseline (~15-50 msgs / 5-hr window)	Sonnet 4.6, Haiku 4.5	No	Memory (Mar 2026), Projects (light), web search, file uploads, Extended Thinking
Pro	$20/mo ($17/mo annual)	5x Free	All: Opus, Sonnet, Haiku	Yes (limited quota)	Research mode, chat search, unlimited Projects, priority access, M365 and Workspace integrations
Max 5x	$100/mo	25x Free (~225 msgs / 5-hr window)	All models	Full access	Highest priority (faster than Pro), doubled Code limits (May 2026), Guest Passes
Max 20x	$200/mo	100x Free (~900 msgs / 5-hr window)	All models	Full access, highest limits	Zero-latency priority, sustained agentic workflows all day
Teams Standard	$20/seat/mo (annual) or $25/seat/mo (monthly)	~5x+ Free	All models	No	SSO (SAML/OIDC), central billing, admin controls, M365, Slack, no training on data
Teams Premium	$100/seat/mo (annual) or $125/seat/mo (monthly)	~25x+ Free	All models	Yes (5x Standard)	All Standard features, Claude Code, Claude Cowork, engineering team scale. Min 5 seats, max 150.
Enterprise	$20/seat + API usage (annual, 20+ seats self-serve; 50+ sales)	High (custom)	All models	Yes	500K token context, SCIM, audit logs, Compliance API, HIPAA-ready, IP allowlisting, custom data retention, zero training guarantee

How Rolling 5-Hour Windows Work

The window opens when you send your first message in a session and resets exactly 5 hours later -- not at midnight or a fixed daily time.
Pro and Max plans also have a weekly cap that resets 7 days after the session starts (not on calendar days).
Exact message counts are not published. One long document upload, a heavy agentic session, or Extended Thinking on Full mode can exhaust the window in minutes.
A separate Sonnet-specific weekly limit also exists on Max plans.
There is no rollover of unused quota from one window to the next.

What happens when limits are hit

Claude shows a message telling you when the window resets. You must wait.
Options: wait for reset, upgrade to a higher plan, or purchase usage credits.
Context (length) limit: start a new conversation or use Projects to continue with a clean context while retaining file and instruction context.

Priority Access by Plan

Plan	Priority level
Free	No priority; may experience latency during peak US business hours
Pro	Ahead of free users; peak throttling removed as of May 6, 2026
Max 5x	Highest priority; faster than Pro
Max 20x	Zero-latency priority; minimal wait even at peak times
Teams	Standard priority; consistent access for team members
Enterprise	Custom SLA-backed service levels available

Context window by plan

Free, Pro, Max, Teams: 200K tokens (roughly 150,000 words or 500 pages)
Enterprise: 500K tokens on supported models; RAG-based retrieval for even larger project knowledge bases

Model training policy

Free: Conversations may be used for model training (opt-out not available)
Pro, Max, Teams: No training on user data by default (as of August 28, 2025)
Enterprise: Zero training guarantee contractually, not just by default

When to Upgrade

Stay on Free if...

Casual use, fewer than 5 sessions per week, no coding agent needed, and you can work around occasional pauses.

Choose Pro if...

Daily productive use, you need Claude Code for occasional coding sprints, Research mode, or Opus. Predictable $20/month.

Choose Max 5x if...

Hitting Pro limits 3+ times per week; Claude Code is part of your daily workflow; need reliable peak-hour access.

Choose Max 20x if...

Running Claude Code in sustained agentic loops all day; parallel agent teams; replacing API usage costing $150+/month.

Choose Teams if...

5-150 people who need centralized billing, SSO, and admin controls. Standard seats for non-coding roles; Premium for engineers.

Choose Enterprise if...

150+ users; need SCIM, audit logs, Compliance API, HIPAA, IP allowlisting, 500K context, or negotiated pricing.

Good Habits for Better Results

These practices apply across all Claude products. The most impactful habits are about how you frame requests and manage context -- not about finding tricks.

Prompt Best Practices

Assign a role upfront: "You are a senior technical writer at a B2B software company..." This shapes tone, depth, and vocabulary for every response without re-explaining yourself.
Specify the output format: "Respond in bullet points with a one-sentence summary at the top" produces much more usable output than an open-ended request.
Give Claude an audience: "Explain this to a non-technical executive" versus "Explain this to a senior engineer" will produce completely different outputs from the same content.
Ask for honest feedback explicitly: Claude defaults to being helpful and affirming. Say "Be critical. Point out every weakness. Do not soften your feedback" to get real critique.
Use a "do not include" instruction: When output keeps missing the mark, add one more constraint -- audience, format, length, or a specific thing to exclude.
Iterate over 2-3 rounds rather than restarting: Claude retains full conversation context and adjusts on refinement instructions. Starting fresh loses that context.

Context Management

Use Projects for all ongoing work. Set custom instructions once (keep under 400 words); Claude reads them at the start of every conversation in that project.
Keep Project memory scoped: enable "Memory: Scope per Project" so project memories don't bleed into unrelated chats.
Shorter, focused conversations are more efficient. Claude reprocesses the entire conversation history on every message -- a 100-message thread costs dramatically more per response than a fresh chat.
For Claude Code: run /compact at 70% context fill and /clear at 90%+. Ignoring this leads to hallucinations and precision loss.
Turn off tools you are not using. Web search, code execution, and active connectors add token overhead on every message, even when not invoked.

Extended Thinking Guidance

Mode	When to use it	When to skip it
Off	Simple lookups, formatting, rewrites, translation	Any task requiring reasoning
Light	Everyday tasks where some reasoning helps -- Claude decides when to think deeply	Already-simple tasks; wastes tokens
Full	Math, multi-step debugging, complex decisions, strategy with tradeoffs, hard coding problems	Summarization, translation, formatting -- adds 3-12 seconds and 20-40% more tokens for no benefit

Default to Light mode. Switch to Full only when you know the task is genuinely hard. This can reduce Opus token spend by 18-25% over time.

Common Mistakes

Treating Claude output as final: Always review numbers, names, dates, and any factual claim before sharing externally. Claude can make confident-sounding errors.
Using one long thread for everything: Long threads get expensive and context-heavy. Use Projects and start fresh conversations for new tasks.
Not giving Claude a role: Open-ended prompts produce generic output. Role + audience + format = consistently better results.
Skipping CLAUDE.md in code projects: Without it, Claude asks more clarifying questions and makes more assumptions. A 30-minute CLAUDE.md setup saves hours.
Ignoring context fill warnings in Claude Code: Running to 90%+ context without /clear leads to hallucinated code paths and missed dependencies.
Using Extended Thinking Full mode on simple tasks: Adds latency and token cost with no quality benefit on summarization, formatting, or translation tasks.
Keeping sensitive data in memory: Use the "forget what you remembered about [topic]" command or the memory exclusion list in settings to keep sensitive context out of persistent memory.

Tips by Product

Claude.ai Tips

Projects over isolated chats

Projects give Claude persistent context about your work. Set instructions once; they apply to every conversation in that project automatically.

Research mode for cited reports

Use Research mode (Pro+) when you need Claude to gather and cite sources rather than rely on its training data alone.

Lean into the 200K context

Upload multiple long documents in one conversation. Claude can compare, synthesize, and answer questions across all of them simultaneously.

Claude Code Tips

CLAUDE.md is non-negotiable

Add CLAUDE.md to every project root. It is the single most impactful thing you can do to improve Claude Code quality and reduce clarifying questions.

Verify all generated code

Claude Code produces strong first drafts but has a higher rate of logic errors than human-written code. Treat output as a starting point requiring review and testing.

Vet MCP servers before enabling

Third-party MCP servers have access to your filesystem and commands. Review any server's permissions and source before adding it to Claude Code.

Which Tier to Use for Which Task

Task	Recommended tier	Why
Occasional writing and research, light use	Free	The 5-hour window is sufficient for casual users with low frequency
Daily productivity, Projects, occasional Claude Code	Pro ($20/mo)	5x more usage, all models including Opus, Research mode, chat search
Heavy Claude Code daily use, solo developer	Max 5x ($100/mo)	Full Code access with high rate limits; hitting Pro limits regularly
Sustained agentic Claude Code all day, parallel agents	Max 20x ($200/mo)	Zero-latency priority; 100x baseline usage; replaces expensive API usage
Team of 5-150 people, centralized billing, SSO needed	Teams Standard + Premium mix	Standard for non-coding roles, Premium seats for engineers needing Code
Large org, compliance requirements, HIPAA, 500K context	Enterprise	Only Enterprise includes SCIM, Compliance API, audit logs, HIPAA, IP allowlisting

✨ Claude.ai Resources

Official Anthropic documentation, learning resources, and tutorials for getting started with Claude.ai.

Anthropic -- Official 📄 Product Overview · Free Claude.ai Official Product Overview Anthropic's official Claude product page -- covers capabilities, model tiers, use cases, and how to get started with Claude for individuals and organizations. Explore Claude → Anthropic -- Official Docs 📄 Documentation · Free · Reference Claude API and Developer Documentation The complete Anthropic developer documentation hub -- covers the Claude API, SDKs, prompting guides, model selection, extended thinking, file handling, and interactive courses. Read Documentation → Anthropic -- Official 🎓 Learning Resource · Free Claude for Work: AI Solutions for Business Anthropic's official guide to using Claude in a professional and organizational context -- covers workflows, integrations, and best practices for business users of Claude.ai. Explore Work Guide → Anthropic Engineering -- Official 📄 Best Practices Guide · Free Claude Best Practices: Getting More from Claude Anthropic Engineering's official guide on prompting patterns, context management, and productivity techniques that apply across Claude.ai, Claude Code, and API usage. Read Best Practices → YouTube · May 2026 📹 Video · Full Tutorial · Beginner Claude AI Tutorial for Beginners (Step-by-Step) A comprehensive beginner tutorial covering Claude.ai from the basics through advanced features -- includes writing, file uploads, artifacts, projects, connectors, and skills. Watch on YouTube → Anthropic -- Official 📄 Help Center · Free Claude Help Center Anthropic's official Claude Help Center -- covers account setup, plan features, troubleshooting, integrations, and frequently asked questions for all Claude tiers. Open Help Center →

>_ Claude Code Resources

Official Anthropic documentation and guides for Claude Code -- the agentic coding tool that works in your terminal, IDE, and CI/CD pipelines.

Anthropic Academy -- Official 🎓 Learning Hub · Free · Multi-level Build with Claude: Anthropic Developer Academy Anthropic's official developer learning hub -- includes API guides, interactive prompt engineering tutorials, tool use courses, and a Claude Code quickstart video. Covers API and Claude Code setup. Start Learning → Anthropic Engineering -- Official 📄 Best Practices Guide · Free Claude Code: Best Practices for Agentic Coding Anthropic's in-depth engineering guide for Claude Code -- covers CLAUDE.md setup, using bash tools, writing effective instructions, multi-agent verification patterns, and iterating on complex tasks. Read Guide → Anthropic API Docs -- Official 📄 API Reference · Free Claude API: Introduction and Quickstart The official Anthropic API introduction -- covers the Messages API, multi-turn conversations, system prompts, model selection, and the developer tools including Workbench and SDKs for integrating Claude Code. View API Docs → GitHub -- Anthropic Official 📄 Code Samples · Free · Open Source Anthropic Quickstarts: Deployable Starter Apps Official Anthropic GitHub repository of deployable starter applications built with the Claude API -- practical code samples and implementation patterns for developers building with Claude. View on GitHub →