Open Source Model Revolution: The Ultimate Beginner’s Guide to Claude Code

Have you ever imagined having a digital assistant that understands your every word and handles those tedious, repetitive tasks on your computer? Whether it’s splitting a hundred-line Excel payroll sheet, instantly turning ideas into runnable code or web pages, or even assembling scattered materials into a video?

Today, I’m introducing you to exactly that kind of revolutionary tool—Claude Code. It’s far more than just a code generator; it’s a versatile AI Agent that truly understands you and can directly operate your computer system. In the past, such capabilities were often blocked by high costs and strict usage limitations. But now, with the rise of domestic open-source models, we are witnessing a moment of technological democratization. This detailed guide of over 3,000 words will take you from zero to hero, showing you how to safely and affordably unlock what might be the most powerful productivity tool of 2024.

Executive Summary: What is Claude Code, and Why Can You Use It Now?

Claude Code (CC for short) is a command-line based AI agent framework. It can directly read, analyze, and manipulate files and systems on your computer through natural language instructions, automating complex tasks like programming, writing, data analysis, and file organization. Previously, its official service was difficult to access widely due to policy restrictions (described in the text as “banning accounts up to three times a month”) and its commercial scale of 10 billion USD per year. Now, leveraging domestically developed open-source large models whose performance has achieved a breakthrough (such as GLM 4.7, MiniMax M2.1, Kimi K2 mentioned in the text), users can experience a similar level of AI assistant capability at an extremely low cost (e.g., 54 Chinese Yuan for three months), realizing a qualitative change from a “toy” to a genuine “production tool.”

Chapter 1: Redefining Claude Code — A General-Purpose Agent Beyond Just Coding

Before diving into the installation steps, we must first dispel a misconception: although named “Code,” Claude Code’s capabilities extend far beyond programming.

What exactly is it?
You can think of Claude Code as an omnipotent digital butler residing in your computer. It has “eyes” (can read files and images you give it), “hands” (can create, modify, and move files within folders you specify), and a “brain” (a powerful AI model for understanding your instructions and planning execution steps). Your interaction with it is through the most natural medium: conversation.

What can it actually do?
Based on the author’s firsthand experience, Claude Code’s application scenarios include but are not limited to:

• Office Automation: For example, automatically splitting a master payroll spreadsheet into individual payslip files for each employee in 29 seconds.
• Content Creation: Conducting Q&A, writing, even quickly combining audio and images into a video.
• Software Development: Writing web pages and software. The author themselves used CC to develop an open-source launcher called “Claude Code Now,” which has garnered over 400 Stars on GitHub.
• Data Analysis: Reading various data files (like CSV, Excel) and performing in-depth analysis.
• Intelligent Design: Using Skills (skill packages), input a website URL to automatically redesign a highly sophisticated homepage for it.

One sentence from the source text perfectly summarizes its value: “Many tedious tasks on the computer are just one sentence for CC.” Its emergence has even amazed OpenAI’s renowned researcher, AK.

Why is now the best time to get started?
Previously, directly using the official Claude service faced high barriers to entry and significant account suspension risks. However, a turning point has arrived. As the author’s friend excitedly announced in a group chat: “This moment has arrived!” The coding capabilities of domestic open-source models have achieved a critical breakthrough, allowing us to use localized, low-cost, highly available AI models to power Claude Code. This means we can now access top-tier AI productivity, which once required high costs and technical expertise, for less than the price of a cup of coffee (the text notes a quarterly package for GLM 4.7 costs only 54 Yuan).

Chapter 2: From Zero to Hero — A Step-by-Step Guide to Installing and Configuring Your Claude Code

The beginning is the hardest part, but if you follow these steps, in about an hour you will have your own AI assistant. Please ensure all operations are performed within a stable and scientifically configured network environment.

Preparation

1. Install Node.js: Claude Code runs on Node.js. Please go to its official website to download and install the latest version.
2. (Windows Users Only) Install Git: You need to install Git for Windows to provide a complete command-line environment.

Formal Installation of Claude Code

1. Open the Terminal: Search for “Terminal” or “Command Prompt” (CMD) on your computer and open it. Don’t be intimidated by the black window; it’s just a dialog box for entering commands.
   

2. Execute the Installation Command: In the terminal, paste and execute the following command:

npm install -g @anthropic-ai/claude-code

• Potential Issues: If you have already installed other AI programming tools (like Cursor), it is strongly recommended to execute this command within their built-in terminal, as they usually handle environment dependencies better. If the installation fails, you can directly copy the error message and ask any AI assistant (like ChatGPT, Wenxin Yiyan, etc.) for help diagnosing and solving the problem.
• Alternative Installation Methods:

  • macOS, Linux, WSL Systems:

  curl -fsSL https://raw.githubusercontent.com/anthropics/claude-code/main/install.sh | bash
  

  • Windows Systems:

  irm https://raw.githubusercontent.com/anthropics/claude-code/main/install.ps1 | iex
  

3. Verify Installation: After installation, enter the following command. If a version number is displayed (e.g., claude-code/1.0.0), the installation was successful.

claude --version

Configuring a Domestic Large Model Server (Using GLM 4.7 as an Example)

To avoid the limitations of the official service, we will use high-performance domestic models. Here is the configuration process using Zhipu AI’s (GLM 4.7) service.

1. Register and Obtain an API Key:

   • Visit the Zhipu AI Open Platform to register an account.
   • After logging in, find “API Key” in the “Personal Center,” create a new key, and copy and save it securely.

2. Subscribe to a Cost-Effective Plan: For stable and worry-free use, it is recommended to subscribe to a “Coding” plan. The text specifically recommends a “New Year special” at the time—a quarterly Coding Lite plan for only 54 Chinese Yuan. This provides ample quota for heavy usage.
   

3. Use the Configuration Tool to Import the API:

   • Run the following command in the terminal to start a Chinese configuration helper:

   npx @anthropic-ai/claude-code-config-helper
   

   • After the tool starts, you will see a friendly Chinese interface. Simply follow the prompts, paste the API Key copied in the previous step, and complete the subsequent simple configuration steps.
     

Launch Your AI Assistant

After completing all configurations, the exciting moment arrives! In the terminal, simply enter one short command and press Enter:

claude

You will see a clean interactive interface, meaning your Claude Code is ready. Now, you can say “Hello” to it just like chatting with a friend!

Pro Tip: Using a Launcher (Claude Code Now)
If you find it inconvenient to open the terminal and type commands each time, you can install the launcher “Claude Code Now” developed by the article’s author. After installation, you can right-click in any folder to quickly launch Claude Code and automatically load the current folder, greatly improving efficiency.

Chapter 3: Core Principles and Efficient Onboarding Tips

To truly master Claude Code, you need to understand its three core operational concepts. It’s like giving your new assistant an operations manual and a clear workspace.

1. The Folder (Workspace): AI’s “Playground for Work”

Claude Code does not have free rein over your entire computer. All its operations are confined to a single folder you specify. This folder is the context and operational boundary for its current task.

• Best Practice: Create a main folder named “Claude Code,” then create subfolders within it for different projects, such as “Data_Analysis_2024,” “Personal_Blog_Project,” “Reading_Notes,” etc. This maintains organization and enables task isolation.
• Methods to Specify a Folder:

  • Use the “Claude Code Now” launcher, which automatically loads the current folder.
  • Simply drag and drop a folder or file into the Claude Code terminal window; it will automatically read its contents.
    

2. The .claude File: AI’s “Long-Term Memory and Constitution”

Within your project folder, you can create a hidden file named .claude. This file is Claude Code’s core configuration file, acting as its “long-term memory” and the “constitution” for the current project.

• Function: Every time CC starts within that folder, it automatically reads this file. You can write the project goals, your preferences, rules to follow, etc., inside (e.g., “This project aims to develop a calculator. Please use Python and ensure code comments are detailed.”).
• How to Create It: You can create it manually, or simply tell CC: “Please create a .claude configuration file for me, stating that this project is for analyzing sales data.”
  

3. Dangerous Mode: The Trade-off Between Efficiency and Control

By default, CC will ask for your confirmation before performing operations that might modify or delete files. To improve the efficiency of automated workflows, you can enable “Dangerous Mode.”

• Enabling Command:

  claude --dangerously-skip-permissions
  

• Warning: Once enabled, CC will execute all operations without asking. This is extremely risky! You must ensure CC only operates within the designated project folder and develop the habit of regularly backing up important data. The “Claude Code Now” launcher mentioned in the text has this mode integrated by default for convenience, but it must be used with caution.

Practical Tip: How to Input Images to CC?

Some problems (like “Please analyze this chart”) require visual information. In the terminal, the paste shortcut for images is usually not Cmd/Ctrl + V, but:

• macOS: Cmd + Shift + V
• Windows/Linux: Ctrl + Shift + V
  After pasting, CC can “see” the image and analyze it.
  

Chapter 4: Real-World Use Cases and Advanced Features

Real Case Study: Splitting Payroll Slips in 29 Seconds

Let’s experience the power of Claude Code through a real work scenario shared by the author.
Task: Split a master payroll sheet (Excel) containing all employee information into individual payslip files for each employee.
Process:

1. Drag the master payroll file into CC.
2. Enter the instruction: “This is a payroll sheet. Help me split it into individual payslips, one file per person.”
   Result: The entire process, from start to finish, took only 29 seconds. CC automatically completed the entire workflow: reading, analyzing, splitting, and generating individual files.
   
   This is just one example. Any repetitive, rule-based file processing, data conversion, or content generation task can be attempted with CC.

Loading Skills: Directly Reusing Expert-Level Workflows

Skills are pre-packaged, plug-and-play workflows within the Claude Code ecosystem. You can think of them as “skill plugins” installed for CC, pre-tested and optimized by community experts or developers.

• How to Install a Skill?
  For example, to install a Skill named “Frontend Design,” just use a one-line command:

  npx @anthropic-ai/skills-installer install @anthropics/skills/frontend-design --client claude-code
  

• How to Use a Skill?
  After installation, simply tell CC to use that skill. For example: “Use the @anthropics/skills/frontend-design skill to redesign the homepage of https://example.com.”
• Effect: CC will invoke that skill package to quickly generate a well-designed, non-templated webpage proposal. This significantly lowers the professional barrier to entry.
  
  

Chapter 5: Frequently Asked Questions (FAQ) and Conclusion

FAQ

Q1: Why choose domestic models over the official Claude?
A: There are three main reasons: 1) Accessibility: The official service is not friendly to Chinese users, with high account suspension risk; 2) Cost: Domestic model plans are highly competitive (e.g., 54 Yuan per quarter), while the official service has an annual revenue of 10 billion USD, targeting different customer segments; 3) Performance: Current top domestic open-source models (like GLM 4.7) have crossed the critical threshold in coding capability, making them fully capable of handling complex tasks driven by CC.

Q2: What if I encounter errors during installation?
A: This is the most common issue for beginners. First, try executing the installation command within tools like Cursor that already have a configured AI programming environment. If errors persist, copy the complete error message and ask any common AI conversational assistant you use; they can usually provide precise troubleshooting guidance. This is also your first practice in collaborating with AI to solve problems.

Q3: Is enabling “Dangerous Mode” really safe?
A: It is not safe, but sometimes necessary for smooth automation. The core of safety is “isolation.” You must only assign CC to work within clearly defined project folders and absolutely never let it operate on critical system directories or your personal documents without backups. It is recommended that beginners initially keep this mode off, familiarize themselves with how CC works, and then consider enabling it.

Q4: Are there other model options besides GLM 4.7?
A: Yes. According to the source content, other domestic models currently well-supported for Claude Code include MiniMax’s M2.1 and Kimi’s K2 models. Their configuration process is similar to GLM’s—simply obtain an API Key from their respective open platforms and subscribe to the corresponding coding plan.

Conclusion: The Beginning of Technological Democratization and Ecosystem Prosperity

By following this guide, you have successfully crossed the most difficult threshold to using a top-tier AI agent. From now on, a partner that understands you, executes for you, and extends your digital capabilities resides in your computer. Any repetitive work, creative challenge, or learning difficulty you encounter now has a first point of assistance.

The author’s reflection at the end of the text is thought-provoking: “When one whale falls, a myriad of creatures thrive.” In the past, “whales” like Claude monopolized the feast of AI productivity. Now, the qualitative change in open-source models, especially the rise of domestic capabilities, is breaking that monopoly. Technological democratization is happening—every cloud provider can deploy and offer these open-source models, while users can enjoy nearly equivalent intelligent convenience for one percent or even less of the previous cost.

This is not just a tutorial on using a tool; it is the starting point for embracing a new era. The prosperity of the open-source ecosystem will propel “Vibe Coding” (the concept of development through natural language intent) from an interesting toy to genuine productivity. The whale still swims in the deep sea, but in broader waters, a myriad of creatures are thriving.

Now, take a deep breath and give your Claude Code its first task instruction. Your “Wow Moment” might be just a second away.