- 10
- November
"Claude Code for Non-Programmers: Managing Files, Summarizing Documents, and Working With Data on Your Own Computer — How to Install, How to Instruct It, and What to Watch Out For 2026" — the short answer is Claude Code is Claude running on your own computer. It reads the files in a folder you choose, writes new files, and runs commands, and it asks before every change. People who have never written code use it to tidy folders, merge spreadsheets, summarize documents in bulk and draft reports from exported data, without uploading files one at a time the way you do with Claude on the web. This guide covers which plans include it, the install path that never touches a terminal, how to set permission modes safely, and five real prompts for office work together with what to verify in each result.
In one line: Claude Code = Claude that does hands-on work with files on your computer (read, create, edit, run commands). It comes with every paid plan: Pro, Max, Ultra, every Team seat, and Enterprise, but not Free, and it draws on the same usage quota as Claude on the web. Install it either through the Claude Desktop app (the Code tab, no terminal needed) or with a single terminal command. The jobs non-programmers use it for most are tidying folders, batch-renaming files, merging CSV files, turning dozens of PDFs into one table, and drafting reports from ERP exports. The one rule to remember: start in Manual mode, which asks before every edit, and work in a copied folder, never the original.
The installation steps, plan list, permission modes and commands in this article were checked against Anthropic's official documentation at code.claude.com and support.claude.com on 10 November 2026. The program updates often; if your screen differs from what is described, follow the source documents listed under References. The five example tasks run on an ordinary computer with no coding knowledge, but every result needs a human check.
1. How Claude Code Differs From Claude on the Web
Claude on the web (claude.ai) takes the files you upload into a chat, analyzes them, and answers with text or a file to download. Claude Code works the other way round. It sits inside a folder you choose on your computer, reads whatever files the job needs, writes its output into the same folder, and when a task calls for a helper program, say merging 30 spreadsheets, it writes a short script and asks permission to run it. You do not need to read the script. You only decide whether to allow it.
| Aspect | Claude on the web / apps | Claude Code |
|---|---|---|
| Where the files live | Uploaded into the chat, one or a batch at a time | In a folder on your computer, no upload |
| Files per task | Limited by the upload size per conversation | The whole folder; it opens only the files it needs |
| Output | Text, tables, files to download | New or edited files placed straight into the folder |
| Can it run helper programs? | Yes, inside Anthropic's sandbox | Yes, on your machine, asking first for each command in Manual mode |
| Interface | The familiar chat window | The Code tab in the Desktop app, or a text screen in a terminal |
| Best for | Q&A, analyzing a few files, drafting documents | Repetitive work over many files, jobs whose output must be a file |
If your task is to analyze two or three Excel files and get an answer, Claude on the web is enough, and Using Claude to Read and Analyse Excel, CSV and PDF Files explains the limits of that route. If the task is "do the same thing to 40 files and put the results in a new file," Claude Code fits far better. For how Claude Code relates to plugins, MCP and Skills, see Claude Code vs Plugin vs MCP vs Skills.
Using Claude at work? Get a quote in Thai Baht with a full tax invoice
A short form — you get a request number by email right away, and our sales team follows up using that number · one form for Team and Enterprise
2. Which Plans Include It, and How Usage Is Counted
Anthropic's setup documentation states that Claude Code requires a Pro, Max, Ultra, Team, Enterprise or Console (pay-per-use API) account, and that the Free plan does not include it. For Team, the help center says plainly that "Claude Code is included with every Team plan seat," Standard or Premium. The seat types differ in how much usage you get, not in whether you can open Claude Code.
| Plan | Claude Code available? | Quota it draws on |
|---|---|---|
| Free | No | – |
| Pro / Max / Ultra | Yes | Same pool as Claude on the web (5-hour session + weekly · verified Ultra accounts have no 5-hour window) |
| Team Standard | Yes, every seat | 1.25x Pro per 5-hour session + weekly, per seat, not pooled |
| Team Premium | Yes, every seat | 6.25x Pro per session, for people who keep it open all day |
| Enterprise | Yes | Seat-based: quotas like Team · usage-based: no per-seat cap, billed at API rates |
| Console (API) | Yes | Pay per token used |
The point newcomers miss is that the quota is one pool across every channel. If you have Claude Code summarize 40 PDFs in the morning, the quota left for web chat in the afternoon shrinks by the same amount. Every plan's numbers and the options when the quota runs out, such as enabling usage credits, are in Every Claude Plan's Limits Compared. Plan prices and the development-team perspective are in What Is Claude Code? Pricing and How Development Teams Use It.
A billing caution: if the machine has an ANTHROPIC_API_KEY environment variable set (for example because IT once set it up for API testing), Claude Code asks whether to use that key. If you say yes, the work is billed at API rates instead of drawing on your plan quota. The monthly API credits bundled with Max, Ultra and Team cannot be used in Claude Code.
3. Two Ways to Install: the Desktop App With No Terminal, or One Terminal Command
For anyone who has never opened a terminal, the easiest route is the Claude Desktop app, the same app used for chat. Download it from claude.ai, install, sign in, and click the Code tab at the top of the sidebar. The app ships with Claude Code built in, so there is no Node.js or add-on to install. Then pick the Local environment, choose a folder (Open folder…), pick a model, and type your task. Permission requests appear as approve/decline buttons that show which file would change and where.
| Route | Best for | How | Requirements |
|---|---|---|---|
| Claude Desktop app, Code tab | Anyone who would rather not touch a terminal | Download for macOS / Windows (Linux beta) → sign in → Code tab → Open folder… | Paid plan · Windows x64 or ARM64 · macOS Intel/Apple Silicon |
| Native install (recommended) macOS / Linux / WSL | People comfortable with a terminal | curl -fsSL https://claude.ai/install.sh | bash | macOS 13.0+ · Ubuntu 20.04+ / Debian 10+ · 4 GB+ RAM · updates itself |
| Native install Windows (PowerShell) | Windows users | irm https://claude.ai/install.ps1 | iex | Windows 10 1809+ · no Administrator rights needed · Git for Windows optional |
| Homebrew (macOS) | People who already use Homebrew | brew install --cask claude-code | No auto-update; run brew upgrade claude-code |
| WinGet (Windows) | Organizations that manage software through WinGet | winget install Anthropic.ClaudeCode | No auto-update; run winget upgrade Anthropic.ClaudeCode |
| npm | Machines that already have Node.js 22+ | npm install -g @anthropic-ai/claude-code | Never with sudo · same native binary as the installer |
If you take the terminal route on macOS, press Cmd + Space, type Terminal, press Enter, and paste the one-line command above. On Windows, press Win + X and choose Windows PowerShell (the line starts with PS C:\; if there is no PS you are in CMD, which uses a different command). The screen shows no progress while it downloads. Wait for "Claude Code successfully installed!", then open a new terminal window and check:
claude --version # should print a version such as 2.1.xxx (Claude Code)
claude doctor # checks the installation and settings files
cd ~/Documents/trial-work # move into the folder you want Claude to work in
claude # start; the first run opens a browser window to sign in
Install from official channels only: in late October 2026 a trojanized Claude Code extension appeared on the Cursor Marketplace planting an SSH key, and 88 fake domains (more than 30 of them using the Claude Code name) lured people through Google Ads. The real install scripts live at claude.ai/install.sh and install.ps1, downloads come from downloads.claude.ai, and Anthropic's signing key has the fingerprint 31DD DE24 DDFA B679 F42D 7BD2 BAA9 29FF 1A7E CACE. Never install from an ad link or a forwarded file. Details and the machine checks are in the 12-point Claude user security checklist.
4. Three Things to Set Before Your First Prompt: the Folder, the Permission Mode, and the Stop Button
The folder. Claude Code sees only the folder you start it in (plus any you add with /add-dir). Reading files inside that folder needs no approval; editing files and running commands do. The safest habit for a beginner is therefore to create a new folder, copy the files you need into it, and start Claude Code there. The originals stay where they are, untouched.
The permission mode decides what Claude may do without asking. In the Desktop app, pick it from the selector under the prompt box. In the terminal, press Shift + Tab to cycle.
| Mode | Runs without asking | Still asks | Use it for |
|---|---|---|---|
| Manual (internal name default) | Reading files in the folder | Every edit, command and network action | Beginners and sensitive data. Start here. |
| Plan | Reading and planning | No edits until you approve the plan | Big jobs where you want to see the plan first |
| Accept edits (acceptEdits) | File edits and basic file commands (mkdir, mv, cp, rm) inside the working folder | Other commands, and anything outside the folder | Work you will review afterwards anyway |
| Auto | Almost everything, with a classifier model checking in the background | Whatever the classifier judges to exceed your request | Experienced users, not sensitive data |
| bypassPermissions | Everything, no prompts, no checks | – | Never on a work machine. The docs say to use it only in a container or VM without internet |
To make Claude's own file tools refuse every path outside the working folder in every mode, add these two keys to ~/.claude/settings.json (you can ask Claude to create it: "set this up so you can never read or write files outside this folder"), and, because these two keys do not cover shell commands, add deny rules for commands you never want run. Deny rules are checked before everything else and are enforced by the program, not by the model.
{
"permissions": {
"defaultMode": "default",
"blockReadsOutsideWorkingDirectories": true,
"blockWritesOutsideWorkingDirectories": true,
"deny": ["Bash(rm *)", "Bash(curl *)"]
}
}
The stop button. The Desktop app has a stop button. In the terminal, press Esc to interrupt the current action, type a correction while it is still working, and leave with /exit or Ctrl + D twice. Other commands a beginner should know: /help lists everything, /clear starts a fresh conversation, /resume returns to an earlier one, /permissions shows and edits the rules, /add-dir adds a folder, /usage shows remaining quota, and /login switches accounts.
5. Five Real Office Tasks, With the Prompts and What to Check
A prompt that works has the same three parts every time: where the files are, what the output should look like, and where to stop. Type it in your own language. The examples below can be copied as they are.
# Task 1 Tidy a folder
Look at every file in this folder and propose how to sort them into subfolders by year and document type.
Do not move anything yet. Show a table of which file would go where.
# Task 2 Batch-rename files
Scan the PDFs in the invoice folder and rename them to YYYY-MM-DD_vendor_invoice-number.pdf.
Read the date and number from inside each document. If either is unclear, skip the file and list it.
Write the copies into renamed/ and never touch the originals.
# Task 3 Merge CSV files
Merge every CSV in the sales-2026 folder into one file called sales-2026-all.csv.
Check that the column headers match in every file. If they differ, stop and tell me which file differs and how.
Report the row count per file and the total after merging.
# Task 4 Summarize 40 PDFs into one table
Read every PDF in the quotations folder and build one Excel file with these columns:
file name / vendor / date / total before VAT / total after VAT / payment terms
Leave a cell blank if you cannot find the value. Never guess. Add a "page" column saying where each number came from.
# Task 5 Draft a report from an ERP export
Using budget-export.xlsx, draft the monthly budget status report as a Word document.
Structure: five-line summary / table of plan, reserved, committed, pending payment and paid by department / items above 90% used.
Every figure must come from the file. If you calculate anything, show the formula.
| Task | What Claude will do | What you must check |
|---|---|---|
| Tidy a folder | Read names and contents, propose a structure, move only when told | The table before any move · files it was unsure about |
| Batch-rename | Open each PDF, extract date and number, copy to the new name | Open 5 files at random and compare name to contents · the skipped list |
| Merge CSV | Write a merge script, ask to run it, report row counts | Total rows = sum of all files · non-English text not garbled |
| PDFs to a table | Read every file, build an Excel with a source-page column | Open the cited page for 10% of rows · the blanks it refused to guess |
| Report from an export | Read the Excel, calculate, draft the Word file to the structure | Totals match the report inside the ERP · the formulas it shows · the wording |
Two points to repeat. First, always have it write copies (renamed/, -all.csv, a new file). Do not ask it to "edit the original" until you trust the results. Second, when it asks permission to run a command, read the first line to see what it will do to which file. If you do not understand it, choose No and type "what does this command do?" It will explain in plain words before acting. For sharper prompts, see How to Write Prompts for Claude That Actually Work.
6. From Our Own Team: This Entire Website Is Produced With Claude Code
The non-coding work we hand to Claude Code every day: the articles on this site, including the one you are reading, are written and checked with Claude Code on our team's own machines. It researches the source documents, writes the Thai and English content to files, runs the quality script that checks more than 40 internal links per article, builds the cover image from our stock-photo library, and updates the sitemap and RSS index. The team acts as the reviewer: reading every paragraph, correcting facts, and pressing approve before anything goes live. We have no figure for the percentage of time saved, because we never measured it in a controlled way. What we can say is that it lets one person maintain nearly 400 bilingual articles in a way that was not possible before.
The lesson from our own use that you can apply at once: the jobs that go best are the ones with rules already written down for it to read. We keep all of ours in a single file in the site's folder, which is the next section.
7. CLAUDE.md: State the Rules Once Instead of Retyping Them Every Day
Claude Code reads a file called CLAUDE.md in the working folder every time it starts. It is a plain-text file for the rules of that folder: the file-naming pattern, the language you want, the standard columns for tables, and what it must not do. You can have Claude create it by typing "create a CLAUDE.md that describes the rules of this folder," then edit it as you like. An example for a document-work folder:
# Rules for the folder: Purchasing 2026
- Files here are copies exported from the ERP every Monday. Never edit files starting with export-
- Write every output into the output/ folder only
- Dates in the format 10 November 2026 · money with 2 decimal places
- Leave a value blank if it cannot be found. Never estimate
- Before any command that deletes or moves files, list them first
CLAUDE.md is guidance, not a wall: Anthropic's documentation says directly that instructions in CLAUDE.md "shape what Claude tries to do, but they don't change what Claude Code allows." When something must be forbidden, pair it with the deny rules and permission modes in section 4.
8. ERP Data on Your Computer Is a Safe Sandbox, but the Live System Stays Where It Is
Tasks 3 to 5 above all start from files exported out of the ERP. That is the way we recommend ordinary users apply Claude Code to company data: an export is a copy at a point in time, with a clear boundary (which report, which period, whose permissions), and if Claude gets something wrong the live system is unaffected. Saeree ERP remains the system of record. Figures Claude calculates are a draft for a human reviewer, not numbers to quote in place of the report inside the system.
The limits of this approach are equally clear. The data is stale the moment it is exported, someone has to export it every cycle, and nobody controls where the file is copied next. For organizations that want Claude to read live ERP data under each user's own permissions, we build that over MCP (Model Context Protocol), defining which tool may read or write what, binding it to roles in the system, and keeping an audit trail of every request. Grand Linux used this approach on our own internal support work before offering it to customers; details are at Claude Integration & Solutions, and What is Claude MCP? explains the protocol itself. For Claude Team and Enterprise procurement with a Thai tax invoice, see Claude for Organizations.
9. Suitable / Not Suitable
| A good fit for Claude Code | Not yet a job for Claude Code |
|---|---|
| Repetitive work over dozens or hundreds of files you are tired of doing by hand | Editing data directly in a live system, whether ERP, accounting or HR |
| Jobs whose output must be a file (Excel, Word, CSV), not just an answer in a chat | Files with personal data not yet assessed under PDPA (Personal Data Protection Act) for AI processing |
| Work with written rules and a named person checking the result | Actions that cannot be undone, such as emailing an entire customer list |
| Drafting reports from data exported from the ERP | A shared machine with no separate working folders |
| People willing to read the permission request before pressing Yes | People planning to switch on the skip-permissions mode on day one |
Conclusion
Claude Code is not reserved for programmers. It is Claude that can work with the files on your computer, it comes with every paid plan including every Team seat, and the Desktop app lets you use it without a terminal. What makes it safe is not the model's intelligence but three things you set yourself: a copied folder separate from the originals, Manual mode that asks before every edit, and deny rules for commands that must never run. Start with Task 1 in this article today, and move toward Task 5 once you can read its permission requests with confidence.
The best Claude Code user in the office is not the one who can write code. It is the one who can describe their own work most clearly, and knows what to check.
- Saeree ERP team
References
Checked 10 November 2026. All are official Anthropic documentation.
- Claude Code docs: Advanced setup — system requirements, every install method, updates and uninstall
- Claude Code docs: Quickstart — first session and essential commands
- Claude Code docs: Terminal guide for new users — opening a terminal on macOS and Windows
- Claude Code docs: Get started with the desktop app — the Code tab in Claude Desktop
- Claude Code docs: Permission modes — every mode and its warnings
- Claude Code docs: Configure permissions — working directories, allow/deny rules, blocking outside the folder
- Claude Help Center: Using Claude Code with your Pro, Max or Ultra plan — shared limits and usage credits
- Claude Help Center: Using Claude Code with your Team or Enterprise plan — "included with every Team plan seat"
Want your non-technical team to use Claude Code safely?
Grand Linux supplies Claude Team and Enterprise with a Thai tax invoice, and helps set up folder rules, permission modes and a working pattern for ERP exports so your team can start for real.
Get advice / request a quoteTel 02-347-7730 | sale@grandlinux.com




