docs: 更新README文档内容与结构

refactor(useThinkFlow): 改进节点布局算法以支持动态尺寸计算
style(i18n): 修改"deepDive"翻译为"回答"以更准确表达功能
This commit is contained in:
liuziting
2026-01-22 08:50:54 +08:00
parent c2970c3348
commit e442296757
5 changed files with 248 additions and 271 deletions

203
README.md
View File

@@ -2,164 +2,115 @@
English | [中文](./README.zh-CN.md)
ThinkFlow AI is a lightweight, local-first idea-to-structure workspace. You type a core idea, and the app expands it into a navigable knowledge graph (modules/submodules). Each node supports follow-up expansion, deep-dive details, and optional image generation.
ThinkFlow AI is a lightweight, local-first, AI-driven mind mapping workspace. It automatically transforms a simple idea into a structured and highly visual mind map using the divergent thinking power of AI.
This repository currently uses the package name `thinkflow` (historical) while the product name is **ThinkFlow AI**.
This project aims to provide a smooth, zero-backend-config thinking assistant that offers deep analytical capabilities, supporting users during brainstorming, knowledge organization, and creative expansion.
## Highlights
---
- **Idea expansion to graph**: generate a root node and expand into structured child nodes.
- **Context-aware follow-up**: expand a selected node with an additional requirement (follow-up input).
- **Deep dive**: generate detailed explanations for a node and show them inline.
- **Image generation (per node)**: generate/regenerate a visual for a node and preview it fullscreen.
- **Export to Markdown**: export the current graph as a Markdown outline (including deep-dive content).
- **Local-first settings**: API mode and endpoints are stored in `localStorage` (no backend needed).
- **Bilingual UI**: built-in English/Chinese UI with a one-click toggle.
## 🌟 Key Highlights
## Tech Stack
- **🚀 Idea to Graph**: Just enter a core concept, and the AI will automatically generate a root node and expand into multiple insightful child nodes.
- **🧠 Context-Aware Follow-up**: Support for follow-up questions on any node. The AI considers the "thinking path" from root to the current node to ensure logical consistency.
- **📖 Deep Answer (Deep Dive)**: More than just keywords—click "Answer" to have the AI generate detailed analysis, background, and applications for a node, rendered in Markdown.
- **🎨 Intelligent Visual Aid**: Integrated AI image generation creates realistic visuals based on node content, enhancing the visual impact and memory retention of your mind map.
- **📊 Session Summary**: One-click summary of the entire canvas, extracting core logic, main dimensions, and final insights.
- **📂 Structured Export**: Export the entire mind map as a standard Markdown outline, including node descriptions, answers, and hierarchical structure.
- **🔐 Local-First Configuration**: All API configurations (Key, Base URL, etc.) and language preferences are stored in the browser's `localStorage`, ensuring privacy and security.
- **Vue 3** (Composition API)
- **Vite 5**
- **TypeScript**
- **Tailwind CSS**
- **VueFlow** (`@vue-flow/core`, `background`, `controls`, `minimap`)
- **vue-i18n**
- **lucide-vue-next** (icons)
---
## Project Structure
## 🛠️ Tech Stack
Key files you will touch most often:
- **Framework**: [Vue 3](https://vuejs.org/) (Composition API)
- **Build Tool**: [Vite 5](https://vitejs.dev/)
- **Language**: [TypeScript](https://www.typescriptlang.org/)
- **Styling**: [Tailwind CSS](https://tailwindcss.com/)
- **Graph Engine**: [@vue-flow/core](https://vueflow.dev/)
- **I18n**: [vue-i18n](https://vue-i18n.intlify.dev/)
- **Icon Library**: [lucide-vue-next](https://lucide.dev/)
- **Markdown Rendering**: [markdown-it](https://github.com/markdown-it/markdown-it)
- `src/App.vue`: app shell, composes the layout and wires UI to business actions.
- `src/composables/useThinkFlow.ts`: core business logic (graph operations, API calls, persistence).
- `src/components/TopNav.vue`: top toolbar (layout, export, summary, settings, language toggle).
- `src/components/BottomBar.vue`: bottom input bar (core idea input + execute).
- `src/components/WindowNode.vue`: custom VueFlow node UI (expand/deep-dive/image/follow-up).
- `src/components/SettingsModal.vue`: API mode and custom endpoints configuration UI.
- `src/i18n/index.ts`: i18n bootstrap (default English; persists language in `localStorage`).
---
## How It Works (High-Level)
## 📂 Code Structure Guide
ThinkFlow AI is built around a single composable:
```text
src/
├── components/ # UI Components
│ ├── TopNav.vue # Top toolbar: Handles fit, layout, export, summary, and language toggle.
│ ├── SideNav.vue # Left side nav: Controls MiniMap, edge styles, background, and settings entry.
│ ├── BottomBar.vue # Bottom input bar: Initial entry for core ideas.
│ ├── WindowNode.vue # Custom Node: Core UI for expanding, answers, image generation, and follow-ups.
│ ├── SettingsModal.vue # API Settings: Configuration for Text and Image generation endpoints.
│ ├── SummaryModal.vue # Modal for displaying the session summary.
│ ├── ImagePreviewModal.vue # Modal for fullscreen node image preview.
│ └── ResetConfirmModal.vue # Confirmation modal for canvas reset.
├── composables/ # Core Logic Layer
│ └── useThinkFlow.ts # Core Business Hook: Manages node state, API calls, layout algorithms, and persistence.
├── i18n/ # Internationalization
│ ├── locales/ # Language packs (zh.json, en.json)
│ └── index.ts # I18n initialization
├── App.vue # App Entry: Composes components and configures the VueFlow environment.
└── main.ts # Program Entry Point
```
- `useThinkFlow` owns all reactive state for nodes/edges and UI flags.
- UI components are “thin”: they render and forward events (expand, deep-dive, image, export).
- The graph is rendered with VueFlow using a custom node type (`window`).
- Settings are stored in `localStorage`:
- language (`language`)
- API mode + chat/image endpoints/models/keys
### Core Design Principles
## Getting Started
- **State Management**: Uses `useThinkFlow` composable to centralize management of nodes, edges, and UI states, distributed to child components via `App.vue`.
- **Layout Algorithm**: Implements a horizontal tree layout algorithm based on dynamic subtree height calculations, automatically adapting to node size changes.
- **API Communication**: Compatible with OpenAI-style endpoints, supporting custom model names and Base URLs.
### Prerequisites
---
- Node.js **18+** (required by Vite 5)
- npm (or pnpm/yarn; examples use npm)
## 🚀 Quick Start
### Install
### 1. Prerequisites
- Node.js 18+
- npm or pnpm
### 2. Install Dependencies
```bash
npm install
```
### Run Dev Server
### 3. Run Development Server
```bash
npm run dev
```
Then open the URL printed by Vite (usually `http://localhost:5173`).
Access `http://localhost:5173` to start.
### Build
---
```bash
npm run build
```
## ⚙️ Configuration
### Preview Production Build
This project does not provide a backend; users must configure their own API endpoints:
```bash
npm run preview
```
1. Click the **Settings** icon at the bottom of the left sidebar.
2. **Default Mode**: Uses pre-configured trial endpoints (subject to rate limits).
3. **Custom Mode**:
- **Text Gen**: Enter OpenAI-compatible `Base URL`, `Model Name`, and `API Key`.
- **Image Gen**: Enter configuration for `dall-e` or other image generation APIs.
## Configuration
---
### Custom Mode (Bring Your Own Endpoint)
## 📖 Usage Manual
Open **Settings** in the UI:
1. **Start a Session**: Enter your idea in the bottom input (e.g., "Future of Urban Transportation") and press Enter.
2. **Node Actions**:
- **Expand**: Click the `+` on a node or enter a follow-up requirement to generate child nodes.
- **Answer**: Click the book icon for a deep analysis of that node.
- **Image**: Click the image icon to generate a visual reference for the idea.
3. **Organize Canvas**:
- **Layout**: Click the "Layout" button at the top to automatically organize all nodes.
- **Export**: Click the "Export" button at the top to get a Markdown copy of your work.
- Switch to **Custom**
- Configure **Text Generation** (chat completion)
- `baseUrl` (POST endpoint)
- `model`
- `apiKey`
- Configure **Image Generation**
- `baseUrl` (POST endpoint)
- `model`
- `apiKey`
---
These fields are saved into `localStorage` automatically.
## 📄 License
## Features Guide
### 1) Expand From a Core Idea
1. Type a core idea in the bottom input.
2. Press Enter or click **Execute**.
3. A root node appears and the app generates child nodes.
### 2) Expand a Specific Node (Follow-Up)
1. Click into a nodes follow-up input.
2. Type an additional requirement.
3. Press Enter to expand using the context path (root → current node).
### 3) Deep Dive a Node
Click **Deep Dive** on a node to generate a detailed explanation. The content will be stored on the node and can be reopened without re-requesting (unless you expand further).
### 4) Generate/Preview Node Image
- Click **IMG** to generate an image for a node.
- Click the image to open fullscreen preview.
- Use the regenerate action to request a new image.
### 5) Export Markdown
Use **Export** in the top toolbar:
- Root becomes the Markdown title
- Child nodes become an indented list
- Deep-dive text is exported as blockquote content under the node
## Third-Party Integrations
- **Microsoft Clarity**: included in `index.html` for analytics.
- **VueFlow plugins**:
- Background (Dots/Lines)
- Controls (zoom/fit)
- MiniMap (overview)
## Internationalization (i18n)
- Default locale: **English**
- Supported: `en`, `zh`
- Language choice is stored in `localStorage` key `language`
To add a new language:
1. Add a JSON file in `src/i18n/locales/`
2. Register it in `src/i18n/index.ts`
3. Add a toggle option in the UI (currently a simple EN/ZH switch)
## Common Troubleshooting
- **CORS / Failed to fetch**: if you use a custom endpoint, ensure it allows browser requests and supports proper CORS headers.
- **401/403**: verify the API key and `Authorization: Bearer ...` format.
- **429**: rate limited by the provider; retry later.
## Scripts
- `npm run dev`: start Vite dev server
- `npm run build`: production build
- `npm run preview`: preview the build output locally
MIT License

View File

@@ -2,165 +2,115 @@
[English](./README.md) | 中文
ThinkFlow AI 是一个轻量、偏本地化(Local-first)的“从想法到结构化图谱”的工作台:输入一个核心想法,应用会把它扩展为可浏览的知识图谱(模块/子模块)。每个节点支持继续追问扩展、深挖详情,以及可选的配图生成
ThinkFlow AI 是一个轻量、Local-first(本地优先)的 AI 驱动思维导图工作台。它能够将你的一个简单想法,通过 AI 的发散性思维,自动化地扩展为结构清晰、可视化程度高的思维图谱
## 亮点功能
本项目旨在提供一个极致流畅、无需复杂后端配置、且具备深度解析能力的思维助手,帮助用户在头脑风暴、知识梳理和创意发散时获得持续的灵感支持。
- **想法扩展为图谱**:生成根节点并扩展出结构化子节点。
- **上下文追问扩展**对选中节点追加需求follow-up 输入)并基于路径上下文扩展。
- **深挖详情**:为节点生成更详细的解释/补充内容并在节点内展开展示。
- **节点配图(可重试)**:为单个节点生成/重新生成图片,并支持全屏预览。
- **导出 Markdown**:将当前图谱导出为 Markdown 大纲(包含深挖内容)。
- **本地配置**API 模式与配置保存到 `localStorage`,无需额外后端。
- **中英双语**:内置英文/中文 UI一键切换。
---
## 技术栈
## 🌟 核心亮点
- **Vue 3**Composition API
- **Vite 5**
- **TypeScript**
- **Tailwind CSS**
- **VueFlow**`@vue-flow/core` + background/controls/minimap
- **vue-i18n**
- **lucide-vue-next**(图标)
- **🚀 从想法到图谱**只需输入一个核心概念AI 会自动生成根节点并向外发散多个具有深度的子节点。
- **🧠 上下文关联追问**支持针对任意节点进行追问AI 在生成时会参考“从根到当前节点”的思考路径,确保生成的逻辑一致性。
- **📖 深度回答Deep Dive**:不再仅仅是关键词,点击“回答”即可让 AI 针对该节点生成详尽的解析、背景及应用场景,支持 Markdown 渲染。
- **🎨 智能视觉辅助**:集成 AI 生图能力,可根据节点内容生成写实风格的配图,提升思维导图的视觉冲击力与记忆点。
- **📊 全篇智能总结**:一键梳理整张画布的所有想法,提取核心逻辑、主要维度和最终洞察,生成精炼的全局总结。
- **📂 结构化导出**:支持将整个思维图谱导出为标准的 Markdown 大纲,包含节点描述、回答内容及层级结构。
- **🔐 本地化配置**:所有的 API 配置Key、Base URL 等)和语言偏好均保存在浏览器 `localStorage` 中,隐私且安全。
## 项目结构
---
你最常关注的文件:
## 🛠️ 技术栈
- `src/App.vue`:应用壳层,组合布局与事件转发。
- `src/composables/useThinkFlow.ts`:核心业务逻辑(节点/边操作、API 调用、持久化)。
- `src/components/TopNav.vue`:顶部工具栏(布局、导出、总结、设置、语言切换)。
- `src/components/BottomBar.vue`:底部输入条(核心想法输入 + 执行)。
- `src/components/WindowNode.vue`:自定义节点 UI扩展/深挖/配图/follow-up
- `src/components/SettingsModal.vue`API 模式与自定义端点配置 UI。
- `src/i18n/index.ts`i18n 初始化(默认英文;语言写入 `localStorage`)。
- **框架**: [Vue 3](https://vuejs.org/) (Composition API)
- **构建工具**: [Vite 5](https://vitejs.dev/)
- **语言**: [TypeScript](https://www.typescriptlang.org/)
- **样式**: [Tailwind CSS](https://tailwindcss.com/)
- **画布引擎**: [@vue-flow/core](https://vueflow.dev/)
- **国际化**: [vue-i18n](https://vue-i18n.intlify.dev/)
- **图标库**: [lucide-vue-next](https://lucide.dev/)
- **Markdown 渲染**: [markdown-it](https://github.com/markdown-it/markdown-it)
## 核心原理(高层)
---
应用围绕一个 composable 构建:
## 📂 代码结构指南
- `useThinkFlow` 统一持有:节点/边数据、UI 状态、网络请求、错误处理、导出能力。
- 组件尽量“轻”负责渲染与事件转发expand、deep-dive、image、export 等)。
- 画布由 VueFlow 渲染,节点使用自定义类型 `window`
- 配置持久化到 `localStorage`
- 语言:`language`
- API 模式与 chat/image 的 baseUrl/model/apiKey
```text
src/
├── components/ # UI 组件库
│ ├── TopNav.vue # 顶部工具栏:处理适配、布局、导出、总结、语言切换等全局动作
├── SideNav.vue # 左侧配置栏:控制小地图、连线样式、背景网格及 API 设置入口
├── BottomBar.vue # 底部输入框:核心想法的初始入口
│ ├── WindowNode.vue # 自定义节点:核心 UI包含扩展、回答、生图及追问逻辑
│ ├── SettingsModal.vue # API 配置弹窗:支持文本/图片生成的 Base URL 和 Key 设置
│ ├── SummaryModal.vue # 全篇总结展示弹窗
│ ├── ImagePreviewModal.vue # 节点配图全屏预览弹窗
│ └── ResetConfirmModal.vue # 画布重置确认弹窗
├── composables/ # 核心逻辑层
│ └── useThinkFlow.ts # 核心业务 Hook管理节点状态、API 调用、布局算法、持久化逻辑等
├── i18n/ # 国际化
│ ├── locales/ # 语言包 (zh.json, en.json)
│ └── index.ts # i18n 初始化
├── App.vue # 应用入口:组合各组件并配置 VueFlow 画布环境
└── main.ts # 程序起点
```
## 快速开始
### 核心逻辑设计
### 环境要求
- **状态管理**: 采用 `useThinkFlow` 组合式函数统一管理画布节点、连线及 UI 状态。通过解构方式在 `App.vue` 中分发给各子组件。
- **布局算法**: 实现了基于子树高度动态计算的“横向树形”排版算法,能够自动适配节点展开后的尺寸变化,确保图谱整齐。
- **API 通信**: 适配 OpenAI 格式接口,支持自定义模型名称及 Base URL。
- Node.js **18+**Vite 5 要求)
- npm也可以用 pnpm/yarn示例使用 npm
---
### 安装依赖
## 🚀 快速开始
### 1. 环境准备
- Node.js 18+
- npm 或 pnpm
### 2. 安装依赖
```bash
npm install
```
### 启动开发环境
### 3. 启动开发服务器
```bash
npm run dev
```
打开 Vite 输出的地址(通常是 `http://localhost:5173`
访问 `http://localhost:5173` 即可开始使用
### 构建
---
```bash
npm run build
```
## ⚙️ 配置说明
### 预览构建产物
本项目不提供后端服务,需用户自行配置 API 接口:
```bash
npm run preview
```
1. 点击左侧工具栏底部的 **设置** 图标。
2. **默认模式**: 预设了部分体验接口(可能有请求限制)。
3. **自定义模式**:
- **文本生成**: 填入 OpenAI 兼容格式的 `Base URL``Model Name``API Key`
- **图片生成**: 填入支持 `dall-e` 格式或其他生图 API 的配置。
## 配置说明
---
### 自定义模式自带端点BYO Endpoint
## 📖 功能使用手册
在界面里打开 **Settings**
1. **开启新会话**: 在底部输入框输入您的想法(如“未来城市的交通方式”),按回车。
2. **节点操作**:
- **扩展**: 点击节点右侧的 `+` 或输入后续要求AI 会继续在该分支下生成子节点。
- **回答**: 点击书本图标,获取该节点的深度解析。
- **生图**: 点击图片图标,为该想法生成视觉参考。
3. **整理画布**:
- **布局**: 点击顶部“布局”按钮,自动按层级整理所有节点。
- **导出**: 点击顶部“导出”按钮,获取 Markdown 格式的内容副本。
- 切换到 **Custom**
- 配置 **文本生成chat completion**
- `baseUrl`POST 接口地址)
- `model`
- `apiKey`
- 配置 **图片生成**
- `baseUrl`POST 接口地址)
- `model`
- `apiKey`
---
这些字段会自动保存到 `localStorage`
## 📄 开源协议
## 功能使用说明
### 1从核心想法生成图谱
1. 在底部输入框输入核心想法
2. 回车或点击 **Execute**
3. 生成根节点并自动扩展出子节点
### 2对某个节点继续扩展Follow-Up
1. 在节点的 follow-up 输入框填写“追加需求”
2. 回车触发扩展
3. 扩展时会携带“从根到当前节点”的上下文路径,保证更贴合当前分支
### 3深挖节点详情
点击节点上的 **Deep Dive**
- 生成更长的解释/拓展内容
- 内容会写回节点并可重复打开(不必每次都重新请求)
### 4节点配图与预览
- 点击节点上的 **IMG** 生成配图
- 点击图片打开全屏预览
- 支持在预览层或节点上进行重新生成
### 5导出 Markdown
顶部工具栏点击 **Export**
- 根节点作为一级标题
- 子节点按缩进列表输出
- 深挖内容作为引用块输出到对应节点下方
## 第三方集成与插件
- **Microsoft Clarity**:在 `index.html` 中引入,用于分析与统计。
- **VueFlow 插件**
- BackgroundDots/Lines
- Controls缩放/适配)
- MiniMap总览
## 国际化i18n
- 默认语言:**英文**
- 已支持:`en``zh`
- 语言选择写入 `localStorage``language`
新增语言建议步骤:
1.`src/i18n/locales/` 添加对应 JSON
2.`src/i18n/index.ts` 注册 messages
3. 在 UI 中扩展语言切换逻辑(目前为 EN/ZH 双向切换)
## 常见问题排查
- **CORS / Failed to fetch**:使用自定义端点时,确保支持浏览器跨域请求并正确返回 CORS 响应头。
- **401/403**:检查 API Key 是否正确,以及是否使用 `Authorization: Bearer ...`
- **429**:触发限流,稍后重试。
## Scripts
- `npm run dev`:启动开发服务器
- `npm run build`:构建生产包
- `npm run preview`:本地预览构建产物
MIT License

View File

@@ -510,35 +510,112 @@ export function useThinkFlow({ t, locale }: { t: Translate; locale: Ref<string>
/**
* 布局:从根节点开始按“横向树形”重新排布节点位置
* - 主要用于整理节点过多导致的视觉拥挤
* - 动态计算节点宽高与子树高度,确保在节点展开(回答/生图)后仍能整齐排布
*/
const resetLayout = () => {
const rootNode = flowNodes.value.find(n => n.data.type === 'root')
if (!rootNode) return
const nodeGapX = 150 // 节点层级间的横向间距
const nodeGapY = 40 // 同级节点间的纵向间距
/**
* 获取节点的实际尺寸,优先使用已测量尺寸,否则使用默认值
*/
const getNodeSize = (node: any) => ({
width: node.dimensions?.width ?? node.measured?.width ?? 280,
height: node.dimensions?.height ?? node.measured?.height ?? 180
})
// 存储每个节点及其子树所需的总高度
const subtreeHeights = new Map<string, number>()
/**
* 第一遍遍历:递归计算每个节点及其子树占用的总高度
*/
const calculateSubtreeHeight = (nodeId: string): number => {
const node = flowNodes.value.find(n => n.id === nodeId)
if (!node || node.hidden) return 0
const size = getNodeSize(node)
const children = flowEdges.value
.filter(e => e.source === nodeId)
.map(e => e.target)
.filter(id => {
const childNode = flowNodes.value.find(n => n.id === id)
return childNode && !childNode.hidden
})
if (children.length === 0) {
subtreeHeights.set(nodeId, size.height)
return size.height
}
let childrenTotalHeight = 0
children.forEach((childId, index) => {
childrenTotalHeight += calculateSubtreeHeight(childId)
if (index < children.length - 1) {
childrenTotalHeight += nodeGapY
}
})
// 节点自身高度与子树高度取较大值
const totalHeight = Math.max(size.height, childrenTotalHeight)
subtreeHeights.set(nodeId, totalHeight)
return totalHeight
}
// 开始计算
calculateSubtreeHeight(rootNode.id)
const visited = new Set<string>()
const layoutNode = (nodeId: string, x: number, y: number) => {
/**
* 第二遍遍历:根据计算出的子树高度,递归设置节点位置
* @param nodeId 当前节点 ID
* @param x 当前起始 X 坐标
* @param ySubtreeTop 当前子树顶部的 Y 坐标
*/
const layoutNode = (nodeId: string, x: number, ySubtreeTop: number) => {
if (visited.has(nodeId)) return
visited.add(nodeId)
const node = flowNodes.value.find(n => n.id === nodeId)
if (node) {
node.position = { x, y }
if (!node || node.hidden) return
const childEdges = flowEdges.value.filter(e => e.source === nodeId)
childEdges.forEach((edge, index) => {
const offsetX = 450
const totalHeight = (childEdges.length - 1) * 280
const startY = y - totalHeight / 2
const offsetY = index * 280
const size = getNodeSize(node)
const subtreeHeight = subtreeHeights.get(nodeId) || size.height
layoutNode(edge.target, x + offsetX, startY + offsetY)
// 将节点放置在子树区域的垂直中心
node.position = {
x,
y: ySubtreeTop + (subtreeHeight - size.height) / 2
}
const children = flowEdges.value
.filter(e => e.source === nodeId)
.map(e => e.target)
.filter(id => {
const childNode = flowNodes.value.find(n => n.id === id)
return childNode && !childNode.hidden
})
if (children.length > 0) {
const nextX = x + size.width + nodeGapX
let currentY = ySubtreeTop
children.forEach(childId => {
const childSubtreeHeight = subtreeHeights.get(childId) || 0
layoutNode(childId, nextX, currentY)
currentY += childSubtreeHeight + nodeGapY
})
}
}
layoutNode(rootNode.id, 50, 300)
// 从根节点开始排版
layoutNode(rootNode.id, 50, 100)
// 延迟执行 fitView 确保位置更新已应用
setTimeout(() => {
fitView({ padding: 0.2, duration: 800 })
}, 100)
@@ -990,4 +1067,3 @@ export function useThinkFlow({ t, locale }: { t: Translate; locale: Ref<string>
expandIdea
}
}

View File

@@ -72,7 +72,7 @@
"coreIdea": "Core Idea",
"followUp": "Ask a follow-up...",
"imgAction": "IMG",
"deepDive": "Detail",
"deepDive": "Answer",
"view": "View",
"regenerate": "Regenerate",
"mainTitle": "Root Node",

View File

@@ -72,14 +72,14 @@
"coreIdea": "核心想法",
"followUp": "输入后续问题...",
"imgAction": "生图",
"deepDive": "详情",
"deepDive": "回答",
"view": "查看",
"regenerate": "重新生成",
"mainTitle": "主节点",
"moduleTitle": "子节点"
},
"prompts": {
"system": "你是一个思维发散助手,帮助用户将想法逐层展开,构建思维树。\n\n工作流程\n1. 用户给出一个初始想法(或选择一个已有节点继续追问)。\n2. 你需要根据【思考上下文路径】(即从根节点到当前节点的思考链路)来理解用户的意图。\n3. 生成 3-5 个更深层或相关维度的子想法。\n4. 每个子想法包含简短名称和简描述。\n\n返回格式必须为严格 JSON\n{'{'}\n \"nodes\": [\n {'{'} \"text\": \"子想法1名称\", \"description\": \"一句话描述\" {'}'},\n {'{'} \"text\": \"子想法2名称\", \"description\": \"一句话描述\" {'}'}\n ]\n{'}'}\n\n注意只返回 JSON不附加解释。",
"system": "你是一个思维发散助手,帮助用户将想法逐层展开,构建思维树。\n\n工作流程\n1. 用户给出一个初始想法(或选择一个已有节点继续追问)。\n2. 你需要根据【思考上下文路径】(即从根节点到当前节点的思考链路)来理解用户的意图。\n3. 生成 3-5 个更深层或相关维度的子想法。\n4. 每个子想法包含简短名称和简描述。\n\n返回格式必须为严格 JSON\n{'{'}\n \"nodes\": [\n {'{'} \"text\": \"子想法1名称\", \"description\": \"一句话描述\" {'}'},\n {'{'} \"text\": \"子想法2名称\", \"description\": \"一句话描述\" {'}'}\n ]\n{'}'}\n\n注意只返回 JSON不附加解释。",
"image": "这是一张高质量写实摄影风格的图片。上下文(从根到当前的概念链路):{context}。主体/概念:{topic}。详情/特征:{detail}。画面要求:真实材质与细节、准确透视与比例、自然光或电影级布光、干净背景与高级构图、色彩克制,适合作为思维导图的视觉核心。质量:超清、锐利、噪点控制自然、层次分明的景深。严禁:任何文字/水印/Logo卡通、插画、绘本、手绘、漫画、动漫、3D 渲染感、过度滤镜、变形、模糊、低清。",
"continue": "请继续深入发散",
"coreIdeaPrefix": "核心想法",