Contents
In short. Claude Code is an Anthropic program that, per its documentation, reads files in a folder you choose, edits them, runs commands and checks the result. It runs in the Claude app, a terminal, code editors and a browser. It needs a paid subscription or a Console account billed per token; the free plan does not include it. Availability depends on Anthropic's list of supported countries, so check it first. Start with copies of files without personal data and check the result by hand.
What is Claude Code, and how is it different from a regular AI chat?
Anthropic describes Claude Code as an agentic tool: it reads files in a folder you choose, edits them, runs commands and connects to services. A chat replies with text; an agent loops: gather context, act, verify, repeat until the task is done. The documentation adds that it is useful beyond code, for anything done from the command line.
So it can open an export, write a small program for a calculation, run it and show the result. The pages we read do not describe spreadsheet work separately, so this is the editors' inference: test it on your own files.
Scroll right to see all columns
| Option | What you need | Easiest for |
|---|---|---|
| Claude app, Code tab | A paid subscription, no terminal | People who have never used a terminal |
| Terminal | A subscription or Console account | People who want every setting |
| VS Code or JetBrains extension | A code editor and an account login | People who work in an editor |
| Browser, claude.ai/code | A subscription; tasks run on Anthropic servers | Long tasks without installing |
Is Claude Code available in your country?
The answer is on the official Anthropic page "Supported Regions Policy", which lists the countries and regions where its products, including Claude.ai, Claude Code and Cowork, are available. On 3 November 2026 Russia and Belarus are not on it, and the page says any country or region not listed is unsupported. It is marked "Effective Nov 12, 2026", a date after our check, so open it again before you decide.
The policy is not only about where a person sits. It also names organizations incorporated or headquartered in an unsupported region, and organizations majority-owned or controlled by persons or entities there. How this reads for your company is a question for a lawyer (see the data section). This journal gives no advice on working around restrictions.
What does it cost, and what does each plan include?
Prices come from the Anthropic pricing page on 3 November 2026, in US dollars before tax; Anthropic may change them. Pro costs $20 a month billed monthly or $17 billed annually ($200 up front): 20 × 12 = 240, so the year costs $40 less. Team seats are sold to teams of 2 to 150 people.
Pro and Max have limits per five hours and per week, shared between Claude chat and Claude Code. The other route is a Console account with prepaid credits, billed per token, a unit of processed text. The /usage command shows your own consumption.
Scroll right to see all columns
| Plan | Price per month | Claude Code | Usage limits on the pricing page |
|---|---|---|---|
| Free | $0 | Not included | Limited chat |
| Pro | $20, or $17 billed annually | Included | Five-hour and weekly windows |
| Max | $100 | Included | 5x Pro usage |
| Ultra | $200 | Included | 10x Pro usage per week |
| Team, per seat | Standard $25 or $20 billed annually, Premium $125 or $100 | Included | Premium 5x Standard |
Which business tasks is it good for, and when do you not need it?
Anthropic documentation has examples beyond code: rename screenshots by what they show, build a simple webpage from a description, search and edit files in a folder of notes. The other rows are the editors' judgments: test them on your own files.
Scroll right to see all columns
| Task | Do you need Claude Code? | Condition |
|---|---|---|
| Weekly report from a CSV export of orders | Yes, without personal data | Check the total against the source |
| Prototype of a page from a description | Yes, but not a live service | Payments and customer data need a developer |
| A one-off question, a letter | No | A regular chat is enough |
| Routine work with customer names and phones | Not yet | Read the data section first |
| Payments, mailings, database edits | Not at the start | File rollback does not undo them |
How does the working loop run, and where must you check?
Give it a check that comes back pass or fail
Per the documentation, you describe a task, Claude Code reads files, makes changes, runs a check and repeats. Before editing it snapshots a file, and pressing Esc twice rewinds the changes. Rewind does not cover external systems: databases, APIs and deployments cannot be restored that way.
The documentation's main advice: give it a check that comes back pass or fail, and ask for evidence, the command and its output, not the phrase "all done". For you the check is comparing the total with the source on a few rows.
- Describe the input and result: which file, what should come out, how you will know it is right.
- Switch to Plan mode (Shift+Tab): it studies the files and proposes a plan without edits.
- Correct the plan in words, allow it to run, approve edits one by one.
- Ask it to show the check and compare three rows with the source file.
- If it is wrong, press Esc twice, rewind the files and refine the task.
What should you allow, and how do you limit access?
In Manual mode Claude Code starts by reading and asks before edits and commands. In Auto mode a separate model blocks risky actions: since version 2.1.283 it is the starting mode in the terminal and VS Code, but the documentation says it does not guarantee safety. Choose Manual with Shift+Tab.
Hard limits come from rules, not requests in text. A deny rule beats an allow rule in every mode; the program enforces rules itself, while CLAUDE.md only influences the model. Read(./.env) blocks that file, including through cat, but not a script that opens it itself. The sandbox adds operating-system limits: off by default, available on macOS, Linux and WSL2, not on native Windows.
Scroll right to see all columns
| What | Decision | Basis in the documentation |
|---|---|---|
| Files in one working folder (copies) | Allow | Outside it the tool asks, but an approved command can write wherever your account can |
| Files with passwords and keys (.env) | Close with a rule | A Read deny rule holds in every mode |
| Running commands | Approve each one | In Manual mode non-read-only commands need approval |
| bypassPermissions mode | Do not enable | Only in isolated containers and virtual machines |
| External plug-in services (MCP) | Trusted ones only | Anthropic does not security-audit them |
| Payments, mailings, database edits | Do not delegate at the start | File rollback does not undo them |
What happens to your data, and can you share customer data?
On personal plans (Free, Pro, Max, Ultra) you choose: if you allow data use for improving models, Anthropic trains on it, including Claude Code data, and keeps it 5 years; if not, 30 days. Under Team, Enterprise and API terms there is no training without your opt-in. Locally, transcripts sit in plain text in ~/.claude/projects, and /feedback sends Anthropic a copy of the conversation.
An export of orders with names and phones is personal data. The European Commission defines it as any information relating to an identified or identifiable living individual, and says a processor working for a controller needs a binding contract. Whether Anthropic's terms satisfy GDPR or another privacy law, this article cannot say.
Scroll right to see all columns
| Terms | Model training | Retention |
|---|---|---|
| Personal plans, training allowed by you | Yes | 5 years |
| Personal plans, training not allowed | No | 30 days |
| Team, Enterprise, API | No, unless you opt in | 30 days |
What can replace Claude Code or work alongside it?
The comparison uses only developers' own documentation on 3 November 2026. We did not compare quality: no open method exists. The table answers "what is it meant for", not "which is better".
Scroll right to see all columns
| Option | What the documentation says | What is not confirmed |
|---|---|---|
| Claude Cowork | The same approach beyond code, no terminal; access only to folders you connect; beta | Prices not stated separately |
| Cursor | Called a coding agent; the Claude Code extension installs into it | Countries, systems, prices |
| OpenAI Codex | Lists the ChatGPT app, web, a command-line tool, an IDE extension, cloud | Definition, country availability |
| Regular Claude chat | The free plan includes chat, file creation and running code, but not Claude Code | Does not replace local folders |
What to do in the first two hours, and what did we not check?
The plan below is an editorial method, not an Anthropic requirement: one small task on copies of files, 120 minutes in total (15 + 15 + 20 + 10 + 40 + 20). If eligibility is not confirmed at step one, stop.
What this article lacks. The editors claim no hands-on experience with the tool and did not test it: every fact comes from open dated pages. Not confirmed: Cursor and Codex country availability, Cursor prices, current privacy-law wording, payment terms by country. Disclosure: the journal is published by a project connected with the marketing company WEBNOSTRA; if you hire any automation contractor, including it, ask the same questions.
- Eligibility and plan, 15 minutes: country, company, owners, prices, date of the check.
- Folder with copies, 15 minutes: files without names, phones, passwords and keys.
- Install and sign in, 20 minutes: the app or the terminal, per official instructions.
- Permissions, 10 minutes: Manual mode, a ban on reading files with keys, the sandbox if available.
- First task, 40 minutes: Plan mode, execution, then compare three rows with the source file.
- Wrap-up, 20 minutes: /usage, a note on the result, the decision to continue or not.
Key decisions
- before installing, check your country, company, owners and plan on official pages and note the date
- start with copies of files without personal data, with approval mode on and a manual check of the result
- a regular chat covers a one-off question; an agent earns its place with file actions and repeated routine work
What to discuss with your team
- Country, company, owners and plan are checked, date noted
- A folder holds copies without personal data, passwords and keys
- Manual mode is selected and reading files with keys is closed
- The first task has a check against source figures
- Consumption was viewed with /usage and the decision is noted
- The lawyer question on eligibility and customer data was asked before using such data
Terms in plain language
- Agentic tool
- A program that reads files, runs commands and checks the result itself while you supervise
- Permission mode
- What the tool may do without asking: Manual asks, Plan builds a plan, Auto relies on another model
- Sandbox
- An operating-system limit keeping commands inside chosen folders and network addresses
- CLAUDE.md
- A file of instructions read at the start of a session; guidelines, not bans
Sources
- Anthropic: Claude Code overview (documentation)code.claude.com · checked 03/11/2026
- Anthropic: Claude Code quickstartcode.claude.com · checked 03/11/2026
- Anthropic: How Claude Code works, loop, tools, checkpointscode.claude.com · checked 03/11/2026
- Anthropic: Configure permissions in Claude Codecode.claude.com · checked 03/11/2026
- Anthropic: Claude Code permission modescode.claude.com · checked 03/11/2026
- Anthropic: Claude Code securitycode.claude.com · checked 03/11/2026
- Anthropic: Claude Code sandboxingcode.claude.com · checked 03/11/2026
- Anthropic: Claude Code data usage, training and retentioncode.claude.com · checked 03/11/2026
- Anthropic: Manage costs effectively in Claude Codecode.claude.com · checked 03/11/2026
- Anthropic: Best practices for Claude Codecode.claude.com · checked 03/11/2026
- Anthropic: Advanced setup and system requirementscode.claude.com · checked 03/11/2026
- Anthropic: Terminal guide for new userscode.claude.com · checked 03/11/2026
- Anthropic: Get started with the Claude desktop appcode.claude.com · checked 03/11/2026
- Anthropic: Claude Code common workflowscode.claude.com · checked 03/11/2026
- Anthropic: Claude Code feature availability by plan and providercode.claude.com · checked 03/11/2026
- Anthropic: Supported Regions Policyanthropic.com · checked 03/11/2026
- Anthropic: Claude pricingclaude.com · checked 03/11/2026
- Claude Help Center: Using Claude Code with your Pro or Max plansupport.claude.com · checked 03/11/2026
- Claude Help Center: Get started with Claude Coworksupport.claude.com · checked 03/11/2026
- Cursor: documentationcursor.com · checked 03/11/2026
- OpenAI: Codex documentationlearn.chatgpt.com · checked 03/11/2026
- Federal Law 152-FZ on Personal Data, Article 18 part 5 (ConsultantPlus, Russian)consultant.ru · checked 03/11/2026
- Federal Law 152-FZ on Personal Data, Article 12 cross-border transfer (ConsultantPlus, Russian)consultant.ru · checked 03/11/2026
- European Commission: What is a data controller or a data processor?commission.europa.eu · checked 03/11/2026