deepfish-ai

• English | 中文

Table of Contents

• Table of Contents
• 1. Introduction
• 2. Installation
  • Prerequisites
  • Installation via npm
  • Installation from Source
• 3. Quick Start
• 4. Configuration
  • Initial Setup
  • Configuration Commands
  • Configuration File Structure
• 5. Usage
  • Interactive Mode
  • Direct Command Mode
  • Usage Examples
• 6. Extension Development
  • Creating an Extension
  • Registering Extensions
• 7. Recommendations
  • AI Service Selection
• 8. Usage Notes
  • Using Relative Paths
  • Conversation History
• 9. Troubleshooting
  • Configuration Issues
  • AI Service Connection
  • Extension Not Loading
• 10. Contributing
• 11. License
• 12. Support

1. Introduction

An efficient and convenient AI-driven command-line tool designed to break down the barrier between natural language and operating system commands or file operation instructions. It enables non-professional developers to quickly generate directly executable operation instructions through simple natural language descriptions, significantly improving terminal operation efficiency. Core Features:

• Multi-model Compatibility: Seamlessly supports DeepSeek, Ollama, and all AI models that comply with the OpenAI API specification. It can be flexibly switched according to needs to adapt to instruction generation requirements in different scenarios.

• OpenClaw Skill Compatibility: Supports the OpenClaw Skill ecosystem. Skills can be installed, enabled, and managed through the existing Skill commands to quickly expand workflow capabilities.

• Natural Language to Instructions: Precisely parses natural language requirements and automatically converts them into corresponding operating system commands (such as Linux, Windows, and macOS terminal commands) and file operation instructions (such as creating, deleting, and modifying files/directories), eliminating the need to manually write complex commands.

• Highly Extensible: Supports expanding functional boundaries through an extension mechanism. In addition to basic terminal and file operations, it can easily implement complex tasks such as translation, novel writing, file format conversion, and data processing to meet diverse usage needs.

• AI Automatic Extension Generation: There is no need to manually develop complex extension tools. Custom extensions can be directly generated through AI, reducing the threshold for extension development and making function expansion more efficient and flexible.Extensions-Example

Suitable for various groups such as developers, operation and maintenance personnel, and daily terminal users. Whether it is quickly executing terminal operations, batch processing files, or realizing personalized needs through extensions, this tool can simplify the operation process, improve work efficiency, and empower every terminal operation with AI.

2. Installation

Prerequisites

• Node.js (v22.14.0 or higher)
• npm or yarn

Installation via npm

npm install -g deepfish-ai

Installation from Source

git clone https://github.com/qq306863030/deepfish-ai.git
cd deepfish
npm install
npm link

3. Quick Start

ai config add # Enter a name, then select deepseek, and enter your DeepSeek API key
ai use [the_name_you_entered]
ai "Help me write an article about future technology in the current directory, output in markdown format"

4. Configuration

Initial Setup

Run the configuration wizard to set up your AI service:

ai config add

This will prompt you to configure the following:

• AI Service Type: Choose DeepSeek, Ollama, or OpenAI
• API Base URL: Default URL provided for each service
• Model Name: Choose the AI model to use
• API Key: Required for DeepSeek and OpenAI
• Temperature: Controls response randomness (0-2)
• Max Tokens: Maximum response length
• Streaming Output: Enable/disable streaming response

Configuration Commands

# Configuration commands
ai config add # Add a new AI configuration
ai config ls # List all AI configurations
ai config use <name> # Set the specified AI configuration as the current one
ai config del <name> # Delete the specified AI configuration
ai config view [name] # View details of the specified AI configuration
ai config edit # Edit the configuration file manually
ai config dir # Open the configuration file directory
ai config reset # Reset configuration
ai config clear # Delete the configuration file

# Extension commands
ai ext add <filename> # Add an extension tool
ai ext del <filepath> # Remove an extension tool by file path
ai ext del <index> # Remove an extension tool by index
ai ext ls # List all extension tools

# Skill commands
ai skill ls # List all registered skills
ai skill add <name> # Add a local skill directory or zip file from the current directory
ai skill del <name|index> # Remove a skill by name or index, exp: ai skill del 1
ai skill install <url> # Install a skill from ClawHub，exp: ai skill install https://clawhub.ai/TheSethRose/agent-browser
ai skill enable <name|index> # Enable a skill by name or index, exp: ai skill enable 1
ai skill disable <name|index> # Disable a skill by name or index, exp: ai skill disable 1
ai skill dir # Open the skill directory

# History commands
ai history clear # Clear the history messages for the current directory
ai history output # Output the history messages to current directory
ai history dir # Open the history directory
ai history reset # Reset all history for all directories

Configuration File Structure

The configuration file (~/.deepfish-ai/config.js) has the following structure:

module.exports = {
  ai: [
    {
      name: "default", // AI configuration name
      type: "deepseek", // AI service type: "ollama", "deepseek", or "openai"
      baseUrl: "https://api.deepseek.com", // API base URL
      model: "deepseek-reasoner", // AI model name
      apiKey: "", // API key (required for DeepSeek and OpenAI)
      temperature: 1, // Response randomness (0-2)
      maxTokens: 8192, // Maximum response length
      stream: true, // Enable/disable streaming output
    }
  ],
  currentAi: "default", // Name of the currently active AI configuration
  maxIterations: -1, // Maximum iterations for AI to complete the workflow, -1 for unlimited
  maxMessagesLength: 150000, // Maximum compression length, -1 for unlimited
  maxMessagesCount: 100, // Maximum compression count, -1 for unlimited
  maxHistoryExpireTime: 30, // Maximum session expiration time in days, -1 for unlimited, 0 to disable recording
  maxLogExpireTime: 3, // Log expiration time in days, -1 for unlimited, 0 to disable recording
  maxBlockFileSize: 20, // Maximum block file size in KB; files exceeding this size need to be processed in blocks
  extensions: [], // List of extension file paths
  skills: [], // List of skill configurations
  encoding: "utf-8", // Command line encoding format, can be set to utf-8, gbk, etc., or auto/empty for auto-detection
};

5. Usage

Interactive Mode

Start an interactive session (multi-turn dialogue):

ai

Or explicitly specify:

ai -i or ai -interactive

Direct Command Mode

Execute a single command:

ai "Create a file named hello.txt with the content 'Hello World'"

Usage Examples

File Operations:

ai "Create 10 text documents, each with 100 random characters"
ai "Clear the current directory"

Code Generation:

ai "Create a simple Express server with a /hello endpoint"
ai "Create a browser-based airplane shooting game"

System Commands:

ai "List all files in the current directory with their sizes"
ai "Check disk usage for the current directory"

Extension Tool Generation:

ai "Create a weather.js extension tool for querying weather"
ai ext add weather.js

Skill Management:

ai skill install https://clawhub.ai/TheSethRose/agent-browser
ai skill install https://clawhub.ai/steipete/weather
ai skill ls
ai skill enable 1
ai skill disable 0

OpenClaw Skill Generation:

ai "Create a weather query skill"
ai skill add weather-query
ai skill enable weather-query

Media Processing:

ai "I have ffmpeg5 installed on my system, help me convert all MP4 files in the directory to AVI format"

File Organization:

ai "Classify all files under the 'model' directory into the 'model2' directory by month, date format YYYY-MM"

Task List Execution:

ai "Create a task list: 1.xxxx; 2.xxxx; ..."
ai "Execute task list" # Start execution

ai "I want to implement an extension tool for long-form novel writing that supports large-scale writing, maintains contextual logic coherence, and avoids AI context explosion issues. This extension tool may be a bit complex to implement. You need to carefully read the extension tool generation rules first, then create a task list"
ai "Execute task list" # Start execution

6. Extension Development

Extensions allow you to add custom functions that AI can use in its workflows. For complex processes, you can develop them yourself or try generating extensions using this program, then register the extension with the program and use the command line to complete tasks.

Creating an Extension

• Extensions should export an object containing descriptions (an array of tool descriptions) and functions (an object of functions).
• Tip: You can generate extensions via AI, e.g., ai "Create a weather.js extension tool for querying weather"
• View example extensions.

// Example extension: Weather Extension
const axios = require('axios');

const descriptions = [
  {
    type: 'function',
    function: {
      name: 'getWeather',
      description: 'Get current weather information for a city',
      parameters: {
        type: 'object',
        properties: {
          city: { type: 'string' }
        },
        required: ['city']
      }
    }
  }
];

const functions = {
  async getWeather(city) {
    // Implement weather API call
    const response = await axios.get(`https://api.weatherapi.com/v1/current.json?key=YOUR_KEY&q=${city}`);
    return `${city} weather: ${response.data.current.temp_c}°C, ${response.data.current.condition.text}`;
  }
};

module.exports = {
  descriptions,
  functions
};

Registering Extensions

Method 1: Using Command Line

ai ext add <filename> # ai ext add weather.js
ai ext add . # Traverse the current directory, automatically scan and add extensions

Method 2: Manual Configuration

• ai config edit
• Add it to your configuration:

module.exports = {
  // ... other configurations
  extensions: [
    '/path/to/weather-extension.js'
  ],
};

Method 3: Automatic Scanning

Rules for automatic scanning of extension modules upon program startup:

• Scanning locations:

  • node_modules in the npm root directory
  • node_modules in the command execution directory
  • the command execution directory itself

• Scanned files:

  • Extension packages under the @deepfish-ai directory
  • Extension packages starting with "deepfish-"
  • JavaScript extension files in the command execution directory, which are automatically loaded if they contain the strings 'module.exports', 'descriptions', and 'functions'

7. Recommendations

AI Service Selection

Recommendation: Use online AI services (DeepSeek/OpenAI) for best results

While local AI services (like Ollama) offer privacy and offline capabilities, they may have the following limitations:

• Response Accuracy: Local models may not be as rigorous and precise as online models
• Code Quality: Generated code may require more manual review and correction
• Complex Task Handling: May encounter difficulties with multi-step or complex operations
• Language Understanding: Online services provide better language models

For production environments or complex tasks, we recommend using DeepSeek, OpenAI services, or cloud services within Ollama for more reliable and accurate results.

8. Usage Notes

Using Relative Paths

AI always uses paths relative to the current working directory.

Conversation History

Conversation history is created on a per-directory basis — each execution directory corresponds to its own Agent context. This means that conversations started in different directories are independent of each other.

Conversation history will be automatically cleared after a configurable period (controlled by the maxHistoryExpireTime field in the configuration file, default is 30 days). You can also manage it manually:

• ai history dir — Open the history directory to view stored conversation contexts
• ai history clear — Manually clear the conversation history for the current directory
• ai history output — Export the conversation history to the current directory
• ai history reset — Reset all conversation history for all directories

9. Troubleshooting

Configuration Issues

If you encounter configuration errors, try resetting:

ai config reset

AI Service Connection

• Ollama: Ensure Ollama is running locally on port 11434
• DeepSeek/OpenAI: Verify your API key is correct and you have sufficient quota

Extension Not Loading

• Check the file path in the configuration
• Ensure the file exports the correct object
• Verify the file has no syntax errors

10. Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

11. License

This project is licensed under the MIT License - see the LICENSE file for details.

12. Support

For issues and questions, please submit an issue on the GitHub repository.