Kode: Claude Code Open Source Optimized Version

Kode is a powerful AI assistant that runs directly in your computer terminal. You can think of it as a programmer working with you, it understands your codebase, edits files for you, and executes command line instructions to handle the entire development process. Unlike other tools, Kode can use many different big models at the same time, such as DeepSeek, Kimi, GLM, Qwen Coder, and others. This allows it to choose the most appropriate model to solve a problem depending on the task. For example, invoke a model that specializes in abstract thinking when designing system architecture, and switch to another model that is proficient in code generation when writing concrete code.Kode does this through an intelligent@The auto-completion feature that mentions systematic and fuzzy matching allows you to easily call these models, agents (agents) or project files, dramatically improving development efficiency. It also supports deployment via Docker and provides a safe mode to add an insurance policy of manual confirmation when working on important projects.

Related items that have lapsed:analysis_claude_code: a library for reverse engineering Claude Code.,Anon-Kode: command-line AI code assistant (Claude Code code decompilation)

Study:What makes Claude Code so great? Deconstructing the inner magic of its Agent design,Claude Code Complete Hands-On Guide: From Introductory Configuration to Advanced Automation Workflows

Function List

• Multi-model collaboration:: Support for the simultaneous configuration and use of multiple large language models, which can be flexibly switched according to the task requirements, giving full play to the strengths of different models.
• Intelligent Agent System: Can be used@run-agent-nameformatted instructions that delegate specific tasks to specialized subagents.
• Expert model consulting:: Ready for use in dialogues@ask-model-nameto request specific AI models to give professional analysis for difficult problems.
• Code comprehension and editing:: Ability to analyze code structure and file relationships throughout a project and make changes to code files directly from instructions.
• command execution: You can execute shell commands directly in the terminal environment and return results in real time.
• Intelligent auto-completion: A powerful complementation system that allows you to quickly enter commands, file paths, and model names through a variety of algorithms such as fuzzy matching, abbreviation support, and more.
• Automatic Document Generation: SupportAGENTS.mdcriteria, you can use the#Open-ended natural language requests to automatically generate and maintain project documentation.
• Cross-platform support: A variety of mounting options are available, includingBuncap (a poem)npmand support for the development and implementation of the program through theDockerContainerized Deployment.

Using Help

Kode, as a powerful endpoint AI assistant, provides flexible ways to install and use it. Here is a detailed guide to help you get started quickly.

Installation process

RecommendedBunPerform the installation as it is the fastest. If you don't have theBun, which can be installed first with the following command:

curl -fsSL https://bun.sh/install | bash

set upBunAfter that, run the following command to install Kode globally:

bun add -g @shareai-lab/kode

If you are more accustomed to usingnpm, which can also be installed with the following command:

npm install -g @shareai-lab/kode

After successful installation, you can use any of the three commands to start Kode:kode(Master command),kwa(Abbreviation for Kode With Agent) orkd(ultra-short alias).

Basic use

Kode supports two main modes of operation: interactive and non-interactive.

1. interactive mode
   Entering the main command directly into the terminal enters an interactive session of continuous dialog:

   kode
   

   In this mode, you can have a continuous conversation with the AI, which will memorize the context and is suitable for complex tasks that require multiple steps.

2. non-interactive mode
   If you just want to get a quick answer to a question or execute a simple command, you can use the-pparameter, this mode exits immediately after execution:

   kode -p "解释一下这个函数的作用" src/main.js
   

Core Function Operation

1. Utilization@References to the system

@The mention system is a central feature of Kode that greatly simplifies interaction with models, agents, and files.

• Model for consulting experts: When you are faced with a difficult question, you can specify a particular model to answer.Kode's Smart Completion will suggest available models.

  @ask-gpt-5 这个身份验证方法的安全隐患是什么？
  @ask-o1-preview 分析一下这个算法的复杂度。
  

• Calling specialized agents:: Delegate complex tasks to pre-defined specialized agents.

  @run-agent-architect 为这个系统设计一个微服务架构。
  @run-agent-test-writer 为这些模块创建全面的测试。
  

• Referencing Project Documents: Conveniently refer to a project file or directory in a dialog, Kode automatically complements the path.

  @src/components/Button.tsx
  @docs/api-reference.md
  

2. Automatic document generation

Kode followsAGENTS.mdspecification that can help you automate the generation of project documentation. Simply use the#Use it as an opening to ask a question or give a command.

# 如何设置开发环境？
# 这个项目的测试流程是怎样的？

Kode will automatically format the answer and append it to the project root directory in theAGENTS.mdDocumentation.

3. Security model

By default, Kode runs onYOLOmode, which automatically performs file modifications and commands for maximum efficiency. However, when working on critical projects, it is recommended to enable Safe Mode. In safe mode, all hazardous operations require your manual approval.

kode --safe

4. Internal orders

In interactive mode, you can use the/commands at the beginning to control Kode's behavior:

• /help: Displays all available internal commands.
• /model: Toggle or configure the currently used AI model. You can set different default models here for primary tasks, sub-agent tasks, etc.
• /config: Open the global configuration file for more detailed settings.
• /cost: View the number of tokens and costs consumed by each model in the current session.
• /clear: Clear the current dialog history and start a new session.

5. Docker deployment

If you want to use Kode in an isolated environment or easily deploy it on different machines, you can use Docker.

First, clone Kode's source repository and enter the directory:

git clone https://github.com/shareAI-lab/Kode.git
cd Kode

Then, build the Docker image:

docker build --no-cache -t kode .

Once the build is complete, start the Kode container by executing the following command in your own project record. This command will mount your project directory, Kode's configuration directory, and configuration files into the container to ensure data persistence.

cd your-project
docker run -it --rm \
-v $(pwd):/workspace \
-v ~/.kode:/root/.kode \
-v ~/.kode.json:/root/.kode.json \
-w /workspace \
kode

Kode vs. Claude

application scenario

1. Automated Code Refactoring
   When it's time to modernize or refactor an old codebase, developers can instruct Kode: "Use the@run-agent-refactorRefactor all the class components in the project into React Hooks फंक्शन components", Kode scans the entire code base, locates the target files and modifies them one by one.
2. Quickly Generate Unit Tests
   Completing test cases for existing functions. A developer can point to a complex function file and instruct Kode: "Use the@run-agent-test-writerbecause of@src/utils/calculation.jsKode will automatically analyze the function logic and create corresponding test files.
3. System Architecture Design
   During the startup phase of a new project, a product manager or architect can use Kode to brainstorm ideas. For example, type, "Use@ask-o3-previewDesigning a technical solution for our e-commerce system for a high concurrency spike campaign", Kode can provide a preliminary but comprehensive architectural design document.
4. Cross-language project development
   In a project containing multiple programming languages, a developer may not be familiar with one of them. He can instruct Kode: "Using the Qwen Coder model, help me write a script in Python for parsing the log files generated by this Java project".Kode can utilize the code capabilities of the particular model to accomplish the task.

QA

1. How is Kode different from tools like GitHub Copilot?
   GitHub Copilot is primarily used as a plugin within the code editor to provide code auto-completion and suggestions. Kode, on the other hand, is a more standalone AI assistant that runs in the terminal and is capable of doing more than writing code; it can execute shell commands, manage files, automate entire workflows, and collaborate with a number of different AI models to work on tasks in parallel, acting more like an AI developer that can talk to you and execute complex commands.
2. Is the default "YOLO mode" safe? When should I use--safePatterns?
   "YOLO Mode" skips all permission checks and performs the operation directly in order to increase efficiency. This is efficient in trusted development environments and on non-critical projects. However, if you are working on production projects that contain important data or sensitive information, or are using a model from an insufficiently reliable source, it is highly recommended to use thekode --safecommand to start, so that all dangerous operations (e.g., modifying files, executing commands) require you to manually confirm before they can be executed, providing additional security.
3. What AI models does Kode support? Can I add my own models?
   Kode natively supports collaboration with a wide range of mainstream AI models, such as DeepSeek V3.1, Kimi2, GLM-4.5, Qwen Coder, and so on. Its architecture is designed to be open, and as long as the model provides an API interface compatible with OpenAI, you can use it by modifying the configuration file (using the/configcommand or direct editing~/.kode.jsonfile) to plug their models into Kode for use.
4. Why does Kode recommend installing with Bun instead of npm?
   Bun is a modern JavaScript runtime with a package manager that is usually much faster than npm when installing dependencies. For faster installation and setup, Kode officially recommends using Bun. Of course, if you're used to the npm ecosystem, using thenpm installThe installation can be completed equally well and there is no difference in functionality.