Before you start
- A Claude account, created on claude.ai with an email address
- For tasks on your local files, the Claude desktop app (Windows or macOS)
In plain words
Claude is Anthropic's AI assistant. You can ask it a question as you would a colleague, but you can also hand it a whole task: "compare these three quotes and make me a table", "write the procedure from my notes", "prepare a presentation of this report". It searches, reads, calculates and gives you back a usable result: an answer, a document, a spreadsheet, a presentation. This page shows how to find your way around and get the best out of it from the first days. To understand how this kind of AI works and its limits, first see Generative AI explained simply.
The features described here change quickly and some depend on the plan (Free, Pro, Max, Team, Enterprise). The Claude Help Centre is the authority: links at the bottom of the page. Information checked on 20 October 2026.
1. Where to use Claude
| Access | What for |
|---|---|
| Website claude.ai | Do everything from a browser, with nothing to install |
| Desktop app (Windows, macOS) | The same features, plus access to your local folders, a built-in browser and, in beta on some plans, control of the computer |
| Mobile app (Android, iOS) | Ask questions, follow a task started on the computer, reply to it or redirect it |
| Claude in Chrome | An extension that lets Claude act in your Chrome browser |
| Claude for Microsoft 365 | Claude built into the Microsoft 365 apps |
| Claude Code | The agentic programming tool, in the terminal, the code editor or on the web (see Claude Code: installing and getting started) |
| API | For developers integrating Claude into their applications (documentation) |
A conversation started on one device can be picked up on the others: you can start a task on the computer and follow its progress on your phone.
2. Choosing your model
Claude comes in several models, of different sizes and speeds. You choose in the menu below the input box; the models offered depend on your plan.
| Range | Profile | When to use it |
|---|---|---|
| Haiku | The fastest and lightest | Simple questions, rewording, sorting text |
| Sonnet | The balance between quality and speed | Most everyday uses |
| Opus | The most powerful of the regular range | Complex problems, long analyses, difficult code |
At the time of writing, the most recent versions are Claude Haiku 5.5, Claude Sonnet 5.5 and Claude Opus 5.5. Anthropic also offers a higher tier, Mythos, of which Claude Fable 5.1 is the publicly available version, with additional safeguards.
Start simple
Keep the model offered by default, and move to a more powerful model when an answer seems superficial or a task is genuinely complex. A more powerful model uses up more of your usage allowance.
3. Asking a question, handing over a task
Within the same conversation, Claude decides by itself whether an answer is enough or whether to carry out a task: search the web, read the attached files, run code to calculate, then produce a file. There's no mode to choose.
| You write | Claude… |
|---|---|
| "What's the difference between RAID 5 and RAID 6?" | Answers directly in the conversation |
| "Here's the CSV export of my fleet: list the PCs with no updates for 30 days, with a chart" | Reads the file, runs code to analyse it, and gives back a table and a chart |
| "Prepare a 10-slide presentation from this report for Monday's meeting" | Creates a presentation, editable and exportable to PowerPoint or PDF |
| "Turn my notes into an on-call procedure" | Creates a structured document that you can touch up yourself |
Long tasks keep running in the cloud even if you close the tab or the computer, and several tasks can run in parallel.
The level of autonomy
A setting in the input box determines whether Claude asks for your approval before acting:
| Mode | Behaviour | When to use it |
|---|---|---|
| Manual (default) | Claude asks for your approval before each action | To begin with, and whenever it touches your files or your applications |
| Automatic | Claude chains actions together, with automatic safety checks | For tasks you know well, within a controlled scope |
By default, Claude always asks before permanently deleting files.
4. Projects: one space per subject
A project groups conversations, reference documents and instructions on the same subject. Examples: "IT fleet", "Proxmox migration", "Website documentation".
| Project element | What it's for |
|---|---|
| Knowledge | The reference documents (procedures, inventories, diagrams, exports) that Claude consults in every conversation in the project |
| Instructions | The standing instructions: tone, format, context. Example: "I'm a system administrator. Answer with exact commands and full paths, and flag the risks." |
| Project memory | What Claude remembers across the project's conversations, kept separate from other projects |
Create a project as soon as a subject keeps coming back: you'll no longer have to explain the context every time. On Team and Enterprise plans, a project can be shared with colleagues, read-only or with editing rights.
5. Memory
Claude can remember information from one conversation to the next: your job, your tools, your format preferences.
| Setting (Settings > Memory) | Effect |
|---|---|
| Generate memory from chats | Turns memory on; on by default on individual plans |
| Search and reference chats | Lets Claude find a past conversation when you ask ("what did we decide for…?") |
| Topics | A list of everything Claude has remembered: you can read, edit or delete each item |
You can also tell it directly: "Remember that my environment mixes Proxmox and Hyper-V", or "Forget what I told you about…".
Incognito chat
For an exchange that should leave no trace in memory or history, open an incognito chat (ghost icon in a new conversation). It can't create files or run code.
6. Artifacts: documents you keep
An artifact is content that Claude creates for you and that opens beside the conversation: you can edit it, come back to it and share it.
| Type | Examples | Export |
|---|---|---|
| Documents | Procedure, minutes, memo, report | Word, PDF, Markdown, Google Docs |
| Presentations | Meeting slides, project presentation | PowerPoint, PDF, Google Slides |
| Designs | Screen mock-up, poster, landing page | PDF, image |
| Dashboards (beta) | Indicators built from your data | — |
| Other | Diagram, small interactive tool, web page, code | Depends on the type |
Just ask for it ("make it into a document"), or use the output button in the input box to choose a template. In a document, you can highlight a passage and ask Claude to reword it.
An artifact is private by default. If you share it, check that it contains nothing confidential.
7. Connectors, skills and scheduled tasks
| Feature | What it is | Example |
|---|---|---|
| Connectors | Connecting Claude to your applications (email, calendar, cloud storage, ticketing tools) so it can read or write there; they're built on the open MCP protocol | "Summarise the tickets opened this week and prepare the team update" |
| Skills | Procedures you teach Claude once so that it then applies them the same way | Your incident report template, your naming conventions |
| Plugins | Ready-made bundles of skills and connectors, by profession | A package for support, finance, legal |
| Scheduled tasks | A task that reruns by itself, at a set time, in the cloud | "Every Monday at 8 am, give me a summary of the week's monitoring alerts" |
Connectors are added in Customise > Connectors. Only connect what you need, with the minimum permissions.
Connectors and booby-trapped content
An email, a web page or a document read by Claude can contain hidden instructions intended to hijack it (prompt injection). Keep manual mode for actions that send, publish or delete, and review before approving.
8. Privacy: the settings to check
| Point | What to know |
|---|---|
| Model training | On individual plans (Free, Pro, Max), you choose in Settings > Privacy whether your conversations can be used to improve the models (the Help improve Claude setting). This choice also changes how long data is kept |
| Business plans | On Team and Enterprise plans and through the API, conversations aren't used to train models, under Anthropic's commercial terms |
| Sensitive data | Don't paste passwords, keys or third parties' personal data. Replace real values with examples (spo, 192.0.2.10, example.com) |
| Memory | Regularly review and clean up the remembered topics (Settings > Memory > Topics) |
| In a business | Use the business plan chosen by your organisation, and follow its AI usage policy |
9. Good habits
| Habit | Why |
|---|---|
| Give the context: system, version, what you've already tried, the exact error message | A suitable answer first time |
| Say what you want to get and in what form (table, numbered steps, commented script, document) | Fewer back-and-forths |
| Attach the files rather than describing them: logs, exports, screenshots, configurations | Claude works on the real thing |
| One conversation per subject, and a project for recurring subjects | The context stays clear, and you use less of your allowance |
| Ask it to flag doubts and sources | Spot what needs checking |
| Test before production: scripts, commands, configuration changes | Claude can be wrong, and doesn't know every quirk of your environment |
| Correct and clarify rather than starting again: "keep the structure, but…" | The answer improves as the conversation goes on |
Example requests for an IT professional
I manage a fleet of 40 Windows 11 PCs and 3 Proxmox VE servers.
Here's the CSV export of the inventory (attached). Give me:
1. the list of PCs whose warranty expires within 6 months;
2. a table by model with the number of PCs;
3. a chart of the breakdown by year of purchase.
Deliver it all as a spreadsheet.
Here's my /etc/nginx/sites-available/site.conf file (attached).
Review it like a security auditor: list the problems in order of severity,
propose the corrected configuration, and explain each change in one sentence.
Flag anything you're not sure about.
10. The limits to keep in mind
- Claude can be wrong with confidence: check what matters, especially commands, figures and references.
- Usage limits depend on the plan; long tasks using many tools consume more than a simple question.
- Access to local files, the built-in browser and computer control only works while the desktop app is open.
- Some features are in beta or reserved for certain plans: check in the Help Centre.
Key points
- Claude answers questions and carries out complete tasks: research, file analysis, documents, presentations.
- Projects for recurring subjects, with their documents and instructions; memory viewable and editable in the settings.
- Artifacts can be edited, exported (Word, PDF, PowerPoint…) and stay private until you share them.
- Connectors and scheduled tasks sparingly, in manual mode for sensitive actions.
- Check the training setting in privacy, and never paste secrets.