Claude Desktop setup
A few minutes of setup, then a tour of the main screen you will use every day
You can ask a quick question or hand over a whole office task in one Claude conversation. The capabilities previously called Claude Cowork are becoming part of Claude itself. This course uses Claude Desktop to connect local practice files; web and mobile are useful too. First, install the app and find the message box, permission setting, and model picker.
1 · Install it (a couple of minutes)
- Go to claude.com/download and download the app for macOS, Windows, or Linux (beta).
- Open the downloaded file and install it like any other app, then launch it (Applications on a Mac, the Start menu on Windows).
- Sign in with your Anthropic account — the same login you use on claude.ai. You can ask for quick answers and multi-step work from the app.
Linux is in beta for Ubuntu and Debian (x64 and arm64). Computer use and dictation are not yet available there; the screenshots below show the macOS interface.
Prefer another language? Look for Language in your profile menu and choose an available interface language. The menu language and the language of your work are separate: you can still write to Claude in your preferred language. Available choices and menu positions can change.
2 · Where the work runs
Checked on 1 November 2026: new tasks on Pro, Max, and Ultra run in Anthropic's cloud. Older local tasks stay on their original computer. Keep Claude Desktop open and online for access to linked local files, the built-in browser, or computer use.
Unified Claude Enterprise has been generally available since 23 October. For eligible existing organizations that hadn’t already chosen a setting, the unified experience has been on by default since 16 November. See the admin guide. Admins control access; some regulated configurations stay on their existing setup. Team has the unified experience in beta where enabled.
Plan names: Ultra is the former Max 20x plan. Verified Ultra accounts have no five-hour limit; weekly limits still apply. Check your own usage screen rather than assuming unlimited work.
3 · The project screen, top to bottom
Once you're in, you point Claude at a folder to work in (that folder, the working folder, is the whole of lesson 1.3) and you land on the project screen. It splits into three calm regions.
- Left rail — Home and Recents bring your conversations and longer tasks together; Code is for software development. Start a new conversation and describe the result you need. Below that are your Projects and other tools.
- Middle — where you work. The project name sits up top (here colearn); under it, the message box ("New session in colearn") — you brief Claude here in plain language, just like writing to a colleague. If the message box shows a Chat · Cowork selector, choose Cowork for these file-based exercises.
- Right rail — the project's context. Context shows the working folder on your computer (lesson 1.3) and the project's Memory; Instructions is where you set house rules for how Claude works here. You can ignore these at the start — Claude "picks up your project context automatically."
- New — start a fresh conversation (if your message box offers Chat and Cowork, pick Cowork); one task = one focused thread of work.
- Projects — every project you've set up, each with its own working folder, instructions, and memory.
- Scheduled — tasks that run by themselves on a timer; you'll build one in lesson 2.10.
- Artifacts — saved documents, presentations, and interactive tools Claude builds for you. A static snapshot does not refresh itself; for a connected dashboard, check its source, refresh status, and timestamp.
- Plugins / Customize — where skills, connectors, and plugins are managed; lesson 1.11 is all about it.
- Recents — your latest tasks, so you can jump back into anything you were doing.
Now look closer at the slim bar right under the message box — it carries the controls you'll touch most often.
- The
+(bottom-left of the box) — attach extra files or context to a single message. - The microphone (bottom-right) — dictate your request instead of typing.
- The working folder (the colearn chip) — a connected local folder the session can read and save into (lesson 1.3). Check other linked folders, attachments, and enabled tools separately.
- Mode — how much Claude does without stopping to ask. The next section.
- Model · effort (here Opus · Medium) — which Claude does the work, and how hard it thinks. Section 5.
4 · The mode selector — how much leash you give it
Click the mode selector (the chip showing the current mode) and you get two ways for Claude to work. This is the single most important setting for feeling in control, especially at the start. It does not change the folder you granted access to; it changes how often Claude pauses when a step needs explicit approval.
| Mode | What happens | Use it when… |
|---|---|---|
| Manual ✋ | Claude asks before using a new tool or opening a new site. Read each request and choose Allow or Deny. Existing permissions can already allow file changes; Manual is not an approval for every individual action. | you're new, the task touches important files, or you want to watch the plan and approvals closely. |
| Auto 🛡️ | Claude runs on its own without stopping at every step. Automated safety checks run before each action, and it pauses to ask you if anything looks unsafe. | the task is low-risk, well understood, and supervised — ideally a routine you've already watched it handle well. |
Some versions of the app use longer names ("Manually approve" and "Automatically approve"), and older ones called Manual "Ask before acting". Some accounts and organizations also offer a third mode, "Skip all approvals" (formerly "Act without asking"), which lets Claude act without asking at all. Don't use it for this course: pick the mode that asks before acting.
5 · Choosing a model (and effort)
Open the model control on the right of the bottom bar and you can pick which Claude does the work — plus how hard it thinks. There's no wrong choice; it's a trade-off between speed and depth.
| Model choice | Think of it as… | Best for |
|---|---|---|
| Opus: demanding work | the senior specialist | the hardest, most tangled jobs: long multi-step work, careful reasoning, big or messy documents. Often heavier on usage; availability and the default model can change. |
| Sonnet: everyday work | the reliable all-rounder | everyday tasks. Fast, capable, and lighter on your limit than Opus — a smart switch for routine work. |
| Haiku: quick tasks | the quick assistant | simple, short, or repetitive tasks where speed matters and you want to spend the least of your limit. |
Effort is a separate dial: how thoroughly Claude thinks before it acts. Higher effort gives it more room to reason and may help on tricky problems, at the cost of time and usage; you still check the result. For this course, start with the default model at its default effort, raise effort for genuinely hard reasoning, and lower it for obvious tasks. More models simply lets you reach other available models if you ever need a specific one.
Mini-exercise (3 minutes)
Install Claude Desktop if you haven't, sign in, and complete any setup your app requests. Then just open the mode selector and the model menu to see the options — don't run a task yet. Set it to the default model · default effort and Manual. That's the whole setup — in the next lesson we look at what Claude actually is and the kinds of work you can hand it.