busy-agent 0.1.3

模拟 ReAct Agent 工作过程 - Simulate ReAct Agent working process

Busy Agent 🤖

English | 中文

Pretend to be busy in the LLM Agent era!

This is a fun project that reads trajectories from the react-llama dataset and prints the Agent's thinking and execution process in ReAct Agent style, making you look like you're running a real AI Agent.

Features

• 🎨 Colorful Output: Uses ANSI color codes for beautiful terminal output
• ⌨️ Typewriter Effect: Character-by-character printing to simulate real thinking process
• ⏳ Loading Animations: Displays "Thinking...", "Executing..." and other animation effects
• 🔄 ReAct Style: Complete display of Thought → Action → Observation reasoning loop
• 🎲 Random Selection: Randomly selects from 3538 trajectories for display
• 🚀 Multiple Modes: Supports fast mode, loop mode, and more
• 🎯 Smart Answer Generation: Intelligently generates correct or incorrect answers based on configured success rate and incident events
• 🤖 LLM as Judge: Simulates LLM scoring system to evaluate answer quality
• 📊 Observable Metrics: Real-time display of success rate, time spent, step count, incidents, and other key metrics
• 🌍 Multi-language Support: Supports Chinese and English interface switching

Installation

Method 1: Install from PyPI (Recommended)

pip install busy-agent

After installation, you can directly use the busy-agent command.

Method 2: Install from Source

git clone https://github.com/demouo/busy-agent.git
cd busy-agent
pip install -e .

Method 3: Development Mode

If you want to modify the code or contribute:

git clone https://github.com/demouo/busy-agent.git
cd busy-agent
pip install -r requirements.txt
python busy_agent.py  # Run script directly

Usage

Basic Usage

Run the program to continuously display random trajectories (loop mode by default):

busy-agent

Fast Mode

Skip animation effects for quick display (suitable for testing):

busy-agent --fast

Single Run Mode

Display a single trajectory and exit:

busy-agent --once

Specify Index

Display a specific trajectory by index (automatically enters single run mode):

busy-agent --index 0

Loop Mode Settings

Customize the delay between loops (seconds):

busy-agent --delay 5.0

Model Selection

Choose different AI models (similar to Claude's three-tier model system):

qwen-flash (Fast model):

busy-agent --model qwen-flash

qwen-plus (Balanced model, default):

busy-agent --model qwen-plus

qwen-max (Most powerful model):

busy-agent --model qwen-max

The program displays the current model at startup, with different models distinguished by different colors.

Language Selection

Switch between Chinese and English interface:

busy-agent --language en  # English
busy-agent --language zh  # Chinese

Configuration File

The program uses the config.json configuration file to manage delay times and display parameters. You can adjust these parameters to control the Agent's "busyness" level.

Configuration Options

Language Configuration (language):

• default: Default language, options: zh (Chinese) or en (English)

Model Configuration (model):

• default: Default model, options: qwen-flash, qwen-plus, qwen-max
• available_models: List of available models and their configurations

Delay Time Configuration (delays):

• thinking.min / thinking.max: Thinking delay time range (seconds), default 4.0-10.0 seconds
• executing.min / executing.max: Action execution delay time range (seconds), default 6.0-12.0 seconds

Typewriter Effect Configuration (typewriter):

• thought_speed: Typing speed for thought content (delay per character), default 0.02 seconds
• action_speed: Typing speed for action content, default 0.015 seconds
• observation_speed: Typing speed for observation content, default 0.005 seconds

Display Configuration (display):

• observation_max_length: Maximum display length for observation content, default 500 characters

Print Mode Configuration (print_modes):

• switch_interval: Switch print mode every N steps, default 10 steps
• modes: Available print modes (smooth, chunky, slow, instant)

Incident Configuration (incidents):

• model_disconnect: Model disconnection configuration
  • enabled: Whether to enable, default true
  • probability: Trigger probability, default 0.03 (3%)
  • max_retries: Maximum retry count, default 8
• action_timeout: Action timeout configuration
  • enabled: Whether to enable, default true
  • probability: Trigger probability, default 0.02 (2%)
  • max_retries: Maximum retry count, default 8

Success Rate Configuration (success_rate):

• target_rate: Target success rate, default 0.60 (60%)
• incident_penalty: Incident penalty coefficient, default 0.3 (increases failure probability by 30% when incidents occur)
• wrong_answer_strategies: Wrong answer generation strategies
  • unable_to_determine: Unable to determine type answers (weight 0.4)
  • reasoning_failed: Reasoning failed type answers (weight 0.3)
  • wrong_guess: Random wrong answers (weight 0.3)

LLM Scoring Configuration (llm_judge):

• enabled: Whether to enable LLM as Judge scoring, default true
• correct_answer_score: Correct answer score range, default 8.5-10.0
• wrong_answer_score: Wrong answer score range, default 2.0-6.0

Metrics Tracking Configuration (metrics):

• track_success_rate: Whether to track success rate, default true
• track_time: Whether to track time spent, default true
• track_steps: Whether to track step count, default true
• track_retries: Whether to track retry count, default true
• track_incidents: Whether to track incident events, default true

Output Example

The program outputs the Agent's reasoning process in ReAct style:

🤖 ReAct Agent Working...
================================================================================

❓ Question:
Since 2017 Nick Ayers has been Chief of Staff to a man that served as governor of what state?

🔄 Starting reasoning process...

💭 Thought 1: I need to search Nick Ayers, find who he is chief of staff to...
⚡ Action 1: Search[Nick Ayers]
📊 Observation 1: James Nicholas Ayers (born August 16, 1982) is an American...

💭 Thought 2: Nick Ayers is chief of staff to Mike Pence...
⚡ Action 2: Search[Mike Pence]
📊 Observation 2: Michael Richard Pence (born June 7, 1959) is an American...

💭 Thought 3: Mike Pence was governor of Indiana...
⚡ Action 3: Finish[Indiana]

✅ Final Answer: Indiana
================================================================================

📊 Observable Metrics:

🤖 LLM as Judge Score: 9.27/10.0
✅ Overall Success Rate: 100.0% (1/1)
⏱️  Time Spent: 45.32s
📝 Total Steps: 8

Project Structure

busy-agent/
├── busy_agent/           # Package directory
│   ├── __init__.py      # Package initialization
│   ├── agent.py         # Core BusyAgent class
│   ├── cli.py           # CLI entry point
│   └── data/            # Data files
│       ├── config.json
│       └── datasets/
│           └── react-llama.parquet
├── setup.py             # Packaging configuration
├── MANIFEST.in          # Data file manifest
├── README.md            # Chinese documentation
├── README_EN.md         # English documentation
├── requirements.txt     # Python dependencies
└── LICENSE              # License file

Dataset

The project uses the react-llama dataset, containing 3538 ReAct-style trajectories. Each data entry includes:

• question: Question
• correct_answer: Correct answer
• trajectory: Complete reasoning process (Thought → Action → Observation)

Technical Implementation

• Parsing: Uses regular expressions to parse trajectory text
• Typewriter Effect: Character-by-character printing to simulate real input
• Loading Animation: Uses Unicode characters to create rotating animations
• Color Output: ANSI escape sequences for colorful terminal output
• Smart Answer Generation: Probabilistic determination based on success rate and incidents
• LLM as Judge: Simulated scoring system with realistic score ranges
• Observable Metrics: Real-time tracking and display of key performance indicators

License

MIT License

---

Note: This project is for entertainment and demonstration purposes. It simulates an AI Agent's working process to make your terminal look busy and professional.

For more information, visit: https://github.com/demouo/busy-agent

---

Busy Agent 🤖

English | 中文

在 LLM Agent 时代假装很忙！

这是一个有趣的项目，通过读取 react-llama 数据集的 trajectory，以 ReAct Agent 的风格逐步打印 Agent 的思考和执行过程，让你看起来像是在运行一个真实的 AI Agent。

特性

• 🎨 彩色输出: 使用 ANSI 颜色代码，让输出更加美观
• ⌨️ 打字机效果: 逐字符打印，模拟真实的思考过程
• ⏳ 加载动画: 显示"思考中..."、"执行中..."等动画效果
• 🔄 ReAct 风格: 完整展示 Thought → Action → Observation 的推理循环
• 🎲 随机选择: 从 3538 条 trajectory 中随机选择展示
• 🚀 多种模式: 支持快速模式、循环模式等
• 🎯 智能答案生成: 基于配置的成功率和意外事件，智能生成正确或错误答案
• 🤖 LLM as Judge: 模拟 LLM 评分系统，对答案质量进行评分
• 📊 可观测指标: 实时显示成功率、耗时、步骤数、意外事件等关键指标
• 🌍 多语言支持: 支持中文和英文界面切换

安装

方式一：从 PyPI 安装（推荐）

pip install busy-agent

安装后可直接使用 busy-agent 命令。

方式二：从源码安装

git clone https://github.com/demouo/busy-agent.git
cd busy-agent
pip install -e .

方式三：开发模式

如果你想修改代码或贡献代码：

git clone https://github.com/demouo/busy-agent.git
cd busy-agent
pip install -r requirements.txt
python busy_agent.py  # 直接运行脚本

使用方法

基本使用

运行程序，持续显示随机 trajectory（默认循环模式）：

busy-agent

单次运行模式

显示一次 trajectory 后退出：

busy-agent --once

快速模式

跳过动画效果，快速显示（适合测试）：

busy-agent --fast

指定索引

显示特定索引的 trajectory（自动进入单次模式）：

busy-agent --index 0

循环模式设置

自定义循环间隔时间（秒）：

busy-agent --delay 5.0

模型选择

选择不同的 AI 模型（类似 Claude 的三档模型系统）：

qwen-flash（快速模型）：

busy-agent --model qwen-flash

qwen-plus（平衡模型，默认）：

busy-agent --model qwen-plus

qwen-max（最强模型）：

busy-agent --model qwen-max

程序会在启动时显示当前使用的模型，不同模型使用不同颜色区分。

配置文件

程序使用 config.json 配置文件来管理延迟时间和显示参数。你可以根据需要调整这些参数来控制 Agent 的"忙碌"程度。

配置项说明

语言配置 (language):

• default: 默认语言，可选 zh (中文) 或 en (English)，默认 en

模型配置 (model):

• default: 默认使用的模型，可选 qwen-flash、qwen-plus、qwen-max
• available_models: 可用的模型列表及其配置

延迟时间配置 (delays):

• thinking.min / thinking.max: 思考延迟时间范围（秒），默认 4.0-10.0 秒
• executing.min / executing.max: 执行动作延迟时间范围（秒），默认 6.0-12.0 秒

打字机效果配置 (typewriter):

• thought_speed: 思考内容的打字速度（每字符延迟），默认 0.02 秒
• action_speed: 动作内容的打字速度，默认 0.015 秒
• observation_speed: 观察内容的打字速度，默认 0.005 秒

显示配置 (display):

• observation_max_length: 观察内容的最大显示长度，默认 500 字符

打印模式配置 (print_modes):

• switch_interval: 每N步切换一次打印模式，默认 10 步
• modes: 可用的打印模式（smooth、chunky、slow、instant）

意外事件配置 (incidents):

• model_disconnect: 模型断连配置
  • enabled: 是否启用，默认 true
  • probability: 触发概率，默认 0.03 (3%)
  • max_retries: 最大重试次数，默认 8
• action_timeout: 动作超时配置
  • enabled: 是否启用，默认 true
  • probability: 触发概率，默认 0.02 (2%)
  • max_retries: 最大重试次数，默认 8

成功率配置 (success_rate):

• target_rate: 目标成功率，默认 0.60 (60%)
• incident_penalty: 意外惩罚系数，默认 0.3 (发生意外时增加30%失败概率)
• wrong_answer_strategies: 错误答案生成策略
  • unable_to_determine: 无法确定类答案（权重 0.4）
  • reasoning_failed: 推理失败类答案（权重 0.3）
  • wrong_guess: 随机错误答案（权重 0.3）

LLM评分配置 (llm_judge):

• enabled: 是否启用 LLM as Judge 评分，默认 true
• correct_answer_score: 正确答案评分范围，默认 8.5-10.0
• wrong_answer_score: 错误答案评分范围，默认 2.0-6.0

指标跟踪配置 (metrics):

• track_success_rate: 是否跟踪成功率，默认 true
• track_time: 是否跟踪耗时，默认 true
• track_steps: 是否跟踪步骤数，默认 true
• track_retries: 是否跟踪重试次数，默认 true
• track_incidents: 是否跟踪意外事件，默认 true

配置示例

{
  "model": {
    "default": "qwen-plus",
    "available_models": {
      "qwen-flash": {
        "display_name": "Qwen-Flash",
        "tier": "flash"
      },
      "qwen-plus": {
        "display_name": "Qwen-Plus",
        "tier": "plus"
      },
      "qwen-max": {
        "display_name": "Qwen-Max",
        "tier": "max"
      }
    }
  },
  "delays": {
    "thinking": {
      "min": 2.0,
      "max": 5.0
    },
    "executing": {
      "min": 3.0,
      "max": 6.0
    }
  },
  "typewriter": {
    "thought_speed": 0.02,
    "action_speed": 0.015,
    "observation_speed": 0.005
  },
  "display": {
    "observation_max_length": 500
  }
}

输出示例

程序会以 ReAct 风格输出 Agent 的推理过程：

🤖 ReAct Agent 工作中...
================================================================================

❓ 问题:
Since 2017 Nick Ayers has been Chief of Staff to a man that served as governor of what state?

🔄 开始推理过程...

💭 Thought 1: I need to search Nick Ayers, find who he is chief of staff to...
⚡ Action 1: Search[Nick Ayers]
📊 Observation 1: James Nicholas Ayers (born August 16, 1982) is an American...

💭 Thought 2: Nick Ayers is chief of staff to Mike Pence...
⚡ Action 2: Search[Mike Pence]
📊 Observation 2: Michael Richard Pence (born June 7, 1959) is an American...

💭 Thought 3: Mike Pence was governor of Indiana...
⚡ Action 3: Finish[Indiana]

✅ 最终答案: Indiana

项目结构

busy-agent/
├── busy_agent.py          # 主程序
├── config.json            # 配置文件（延迟时间、打字速度等）
├── datasets/
│   └── react-llama.parquet  # ReAct trajectory 数据集
├── explore_dataset.py     # 数据集探索脚本
├── view_trajectory.py     # 查看 trajectory 示例
├── requirements.txt       # Python 依赖
└── README.md             # 项目说明

数据集

项目使用 react-llama 数据集，包含 3538 条 ReAct 风格的 trajectory。每条数据包括：

• question: 问题
• correct_answer: 正确答案
• trajectory: 完整的推理过程（Thought → Action → Observation）

技术实现

• 解析: 使用正则表达式解析 trajectory 文本
• 打字机效果: 逐字符打印，模拟真实输入
• 加载动画: 使用 Unicode 字符创建旋转动画
• 颜色输出: ANSI 转义序列实现彩色终端输出

许可证

MIT License