从产品想法到完整文档，一键生成 | From idea to complete product documentation in one command
PM Agent 是一个基于 Google ADK 构建的多 Agent 产品经理助手。输入你的产品想法，它会自动生成一套完整的产品文档，包括市场研究、用户画像、功能列表、线框图和交互流程图。
- 层级化多 Agent 架构 - 三层嵌套设计：路由层 → 访谈层 → 文档生成流水线
- 智能访谈机制 - 通过多轮结构化问答收集产品信息，动态判断何时启动生成
- 真实市场研究 - 集成 Google Search，获取真实的市场数据和竞品信息
- 可视化输出 - 自动生成线框图 HTML、Mermaid 流程图、产品定位矩阵
- 结构化文档 - 输出专业的 Markdown 文档，可直接用于团队协作
三层嵌套的多 Agent 系统 (Hierarchical Multi-Agent Architecture)
┌─────────────────────────────────────────────────────────────────────────┐
│ Root Agent (PMAgent) │
│ ┌───────────────────────────────────────────────────────────────────┐ │
│ │ 智能路由：判断用户意图 │ │
│ │ • 普通问题 → 直接回答 │ │
│ │ • 产品想法 → 转发给 InterviewAgent │ │
│ └───────────────────────────────────────────────────────────────────┘ │
│ │ │
│ ▼ transfer_to_agent │
│ ┌───────────────────────────────────────────────────────────────────┐ │
│ │ Interview Agent │ │
│ │ ┌─────────────────────────────────────────────────────────────┐ │ │
│ │ │ 多轮访谈：结构化问答收集产品信息 │ │ │
│ │ │ • 动态判断信息是否充足 │ │ │
│ │ │ • 信息充足 → 触发文档生成流水线 │ │ │
│ │ └─────────────────────────────────────────────────────────────┘ │ │
│ │ │ │ │
│ │ ▼ transfer_to_agent (when ready) │ │
│ │ ┌─────────────────────────────────────────────────────────────┐ │ │
│ │ │ PM Document Pipeline (SequentialAgent) │ │ │
│ │ │ │ │ │
│ │ │ DecisionList → MarketResearch → UserResearch → │ │ │
│ │ │ ProductPositioning → FeatureList → Wireframe → │ │ │
│ │ │ InteractionFlow → ProductSummary │ │ │
│ │ │ │ │ │
│ │ └─────────────────────────────────────────────────────────────┘ │ │
│ └───────────────────────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────────────────┘
- 层级化 Agent 架构 - 而非简单的线性流水线，实现关注点分离
- 智能路由 - Root Agent 根据用户意图动态决策
- 访谈机制 - 通过多轮结构化问答收集信息，而非一次性输入
- 动态触发 - Interview Agent 判断信息充足后才启动文档生成
- 上下文传递 - 各 Agent 通过
output_key共享状态，下游 Agent 可访问上游所有输出
| 阶段 | Agent | 功能 | 输出 |
|---|---|---|---|
| 1 | DecisionListAgent | 整理用户想法为结构化决策列表 | decision_list.md |
| 2 | MarketResearchAgent | 市场规模、竞品分析、行业趋势 | market_research.md |
| 3 | UserResearchAgent | 用户画像、用户故事、验收标准 | user_research.md |
| 4 | ProductPositioningAgent | 产品定位、差异化策略、定位矩阵图 | product_positioning.md |
| 5 | FeatureListAgent | 功能规格、模块依赖图 | feature_list.md |
| 6 | WireframeAgent | 核心界面黑白线框图 | wireframe.html |
| 7 | InteractionFlowAgent | 关键交互时序图 | interaction_flow.png |
| 8 | ProductSummaryAgent | 产品摘要、最终报告 | final_report.html |
- 框架: Google ADK (Agent Development Kit)
- 模型: Gemini 2.5 Flash / Gemini 3 Flash Preview
- 搜索: Google Search API (市场研究)
- 图表: Mermaid (via mermaid.ink API)
- 运行时: Python 3.10+
# 1. 克隆项目
git clone https://github.com/YUHAO-corn/pm-agent.git
cd pm-agent
# 2. 安装依赖
pip install -r pm_agent/requirements.txt
# 3. 配置环境变量
cp pm_agent/.env.example pm_agent/.env
# 编辑 .env 填入你的 Google Cloud 配置
# 4. 运行
adk web pm_agent需要配置 Google Cloud Vertex AI：
# pm_agent/.env
GOOGLE_GENAI_USE_VERTEXAI=1
GOOGLE_CLOUD_PROJECT=your-project-id
GOOGLE_CLOUD_LOCATION=global运行后会在 pm_agent/outputs/<project_name>/ 下生成：
outputs/
└── your-product/
├── decision_list.md # 决策列表
├── market_research.md # 市场研究报告
├── user_research.md # 用户研究文档
├── product_positioning.md # 产品定位文档
├── feature_list.md # 功能列表
├── wireframe.html # 线框图
├── positioning_matrix.png # 定位矩阵图
├── interaction_flow.png # 交互流程图
└── final_report.html # 最终报告
PM Agent is a multi-agent product manager assistant built with Google ADK. Input your product idea, and it automatically generates a complete set of product documentation including market research, user personas, feature lists, wireframes, and interaction flow diagrams.
- Hierarchical Multi-Agent Architecture - Three-layer nested design: Routing → Interview → Document Pipeline
- Smart Interview Mechanism - Collects product info through multi-round structured Q&A, dynamically decides when to start generation
- Real Market Research - Integrated with Google Search for real market data and competitor analysis
- Visual Outputs - Auto-generates wireframe HTML, Mermaid diagrams, positioning matrices
- Structured Documentation - Professional Markdown documents ready for team collaboration
Three-Layer Hierarchical Multi-Agent System
┌─────────────────────────────────────────────────────────────────────────┐
│ Root Agent (PMAgent) │
│ ┌───────────────────────────────────────────────────────────────────┐ │
│ │ Smart Routing: Determine user intent │ │
│ │ • General questions → Answer directly │ │
│ │ • Product ideas → Transfer to InterviewAgent │ │
│ └───────────────────────────────────────────────────────────────────┘ │
│ │ │
│ ▼ transfer_to_agent │
│ ┌───────────────────────────────────────────────────────────────────┐ │
│ │ Interview Agent │ │
│ │ ┌─────────────────────────────────────────────────────────────┐ │ │
│ │ │ Multi-round Interview: Structured Q&A to collect info │ │ │
│ │ │ • Dynamically assess if info is sufficient │ │ │
│ │ │ • When ready → Trigger document generation pipeline │ │ │
│ │ └─────────────────────────────────────────────────────────────┘ │ │
│ │ │ │ │
│ │ ▼ transfer_to_agent (when ready) │ │
│ │ ┌─────────────────────────────────────────────────────────────┐ │ │
│ │ │ PM Document Pipeline (SequentialAgent) │ │ │
│ │ │ │ │ │
│ │ │ DecisionList → MarketResearch → UserResearch → │ │ │
│ │ │ ProductPositioning → FeatureList → Wireframe → │ │ │
│ │ │ InteractionFlow → ProductSummary │ │ │
│ │ │ │ │ │
│ │ └─────────────────────────────────────────────────────────────┘ │ │
│ └───────────────────────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────────────────┘
- Hierarchical Agent Architecture - Not a simple linear pipeline; separates concerns across layers
- Smart Routing - Root Agent dynamically decides based on user intent
- Interview Mechanism - Collects info through multi-round structured Q&A, not one-shot input
- Dynamic Triggering - Interview Agent decides when to start document generation
- Context Propagation - Agents share state via
output_key; downstream agents can access all upstream outputs
| Stage | Agent | Function | Output |
|---|---|---|---|
| 1 | DecisionListAgent | Organizes user ideas into structured decision list | decision_list.md |
| 2 | MarketResearchAgent | Market size, competitor analysis, industry trends | market_research.md |
| 3 | UserResearchAgent | User personas, user stories, acceptance criteria | user_research.md |
| 4 | ProductPositioningAgent | Product positioning, differentiation strategy | product_positioning.md |
| 5 | FeatureListAgent | Feature specifications, module dependencies | feature_list.md |
| 6 | WireframeAgent | Black-and-white wireframes for key screens | wireframe.html |
| 7 | InteractionFlowAgent | Sequence diagrams for key interactions | interaction_flow.png |
| 8 | ProductSummaryAgent | Product summary, final report | final_report.html |
- Framework: Google ADK (Agent Development Kit)
- Models: Gemini 2.5 Flash / Gemini 3 Flash Preview
- Search: Google Search API (for market research)
- Diagrams: Mermaid (via mermaid.ink API)
- Runtime: Python 3.10+
# 1. Clone the repository
git clone https://github.com/YUHAO-corn/pm-agent.git
cd pm-agent
# 2. Install dependencies
pip install -r pm_agent/requirements.txt
# 3. Configure environment variables
cp pm_agent/.env.example pm_agent/.env
# Edit .env with your Google Cloud configuration
# 4. Run
adk web pm_agentRequires Google Cloud Vertex AI setup:
# pm_agent/.env
GOOGLE_GENAI_USE_VERTEXAI=1
GOOGLE_CLOUD_PROJECT=your-project-id
GOOGLE_CLOUD_LOCATION=globalAfter running, outputs are generated in pm_agent/outputs/<project_name>/:
outputs/
└── your-product/
├── decision_list.md # Decision list
├── market_research.md # Market research report
├── user_research.md # User research document
├── product_positioning.md # Product positioning
├── feature_list.md # Feature list
├── wireframe.html # Wireframes
├── positioning_matrix.png # Positioning matrix
├── interaction_flow.png # Interaction flow diagram
└── final_report.html # Final report
MIT License