🧠 DeepBrain

Your personal AI knowledge brain. Import everything, find anything.

你的个人 AI 知识大脑。导入一切，找到一切。

📖 中文文档 / Chinese Documentation

English | 中文

The Problem

You have knowledge scattered everywhere — Notion, Obsidian, Evernote, Flomo, Feishu, WeChat articles, PDFs, EPUBs. When you need to find something, you have to remember where you put it.

DeepBrain brings everything into one place, understands it with AI, and lets you search across all your knowledge with natural language.

What Makes It Different

Quick Start

Install

npm install -g deepbrain

Initialize with your preferred AI provider

# Google Gemini (recommended — free tier available)
deepbrain init gemini

# Local with Ollama (zero cost, zero cloud)
deepbrain init ollama

# Chinese providers
deepbrain init dashscope    # 阿里通义千问
deepbrain init deepseek     # DeepSeek
deepbrain init zhipu        # 智谱 GLM

Add your first knowledge

deepbrain put my-first-note "# Hello Brain
I learned that AI agents work best with structured knowledge bases.
Tags: #AI #agents #knowledge"

deepbrain put meeting-notes "# Team Meeting 2026-04-13
- Decided to use TypeScript for the new API
- Deadline: end of month
- Action: Ray to review architecture doc"

Search with natural language

deepbrain query "what did we decide about the API?"
# → Returns: meeting-notes (score: 0.89)
#   "Decided to use TypeScript for the new API..."

deepbrain query "what do I know about AI?"
# → Returns: my-first-note (score: 0.92)
#   "AI agents work best with structured knowledge bases..."

Open Web UI

deepbrain serve
# → 🧠 DeepBrain Web UI running at http://localhost:3000

📥 Import from Anywhere (21 Sources)

The most comprehensive importer coverage in the open-source ecosystem.

One command to import

# Import your Obsidian vault
deepbrain import obsidian ./my-vault

# Import Notion export
deepbrain import notion ./notion-export

# Import Evernote
deepbrain import evernote ./my-notes.enex

Or use the API

import { importNotion, importObsidian, importFlomo, importWechat } from 'deepbrain';

// Import with progress tracking
const pages = await importObsidian({
  dir: './my-vault',
  onProgress: (current, total, name) => {
    console.log(`[${current}/${total}] ${name}`);
  },
});
console.log(`Imported ${pages.length} pages`);

Supported platforms

International (12): Notion · Obsidian · Evernote · Roam Research · Logseq · Bear · Apple Notes · Google Keep · OneNote · Joplin · Readwise · Day One

Chinese (8): 语雀 · 飞书 · 石墨 · 微信公众号 · Flomo · 我来(Wolai) · 息流(FlowUs) · 思源笔记

Formats (2): EPUB · PDF

Each importer handles platform-specific quirks:

• Obsidian: [[wikilinks]], #nested/tags, YAML frontmatter
• Evernote: ENML→Markdown, checkbox conversion
• Roam: [[page refs]], ((block refs)), TODO/DONE
• Flomo: #tag/subtag hierarchy, memo timestamps
• 微信: js_content extraction, og:title parsing

📤 Export (5 Formats)

Your data is yours. Export anytime, anywhere.

deepbrain export --format markdown --output ./backup
deepbrain export --format obsidian --output ./my-vault     # Open with Obsidian
deepbrain export --format logseq --output ./my-graph       # Open with Logseq
deepbrain export --format json --output ./data.json        # For programmatic use
deepbrain export --format html --output ./site             # Static website

🏷️ Tag Graph — Knowledge Auto-Discovery

This is where DeepBrain goes beyond simple search. Tags aren't just labels — they form a knowledge graph that automatically discovers relationships.

Hierarchical tags

deepbrain tag my-note "AI/Agent/RAG"
# Automatically creates: AI → AI/Agent → AI/Agent/RAG

Auto-recommendations

import { Brain, TagGraph } from 'deepbrain';

const brain = new Brain({ embedding_provider: 'gemini', api_key: 'AIza...' });
await brain.connect();
const tg = new TagGraph(brain);

// "What else should I read based on this page?"
const related = await tg.recommend('my-note', 5);
// → [{ slug: 'agent-design', title: 'Agent Design Patterns', score: 0.78, sharedTags: ['AI', 'AI/Agent'] }]

// "What tags should this page have?"
const suggested = await tg.suggestTags('untitled-note');
// → ['TypeScript', 'API', 'architecture']

// "Show me the full tag landscape"
const graph = await tg.getGraph();
// → { nodes: [...], edges: [...], stats: { totalTags: 42, mostConnected: 'AI' } }

// "Group my knowledge into clusters"
const clusters = await tg.cluster();
// → [{ name: 'AI', tags: ['AI', 'LLM', 'Agent'], pages: [...], size: 15 }]

Tag operations

// Rename across all pages
await tg.renameTag('machine-learning', 'ML');

// Merge duplicates
await tg.mergeTags(['js', 'javascript', 'JS'], 'JavaScript');

// Find lonely tags (only 1 page)
const orphans = await tg.findOrphanTags();

🔍 Hybrid Search

Not just vector similarity. DeepBrain combines three search strategies:

// Semantic: finds conceptually related content
const results = await brain.query('how to build AI agents');

// Keyword: exact match
const results2 = await brain.search('TypeScript');

📊 Dual-Track Storage

Every piece of knowledge has two components:

// Store current facts (overwritable)
await brain.put({
  slug: 'project-alpha',
  type: 'project',
  title: 'Project Alpha',
  compiled_truth: 'Status: In Progress. Team: 5 people. Stack: TypeScript + React.',
});

// Add historical events (append-only)
await brain.addTimeline('project-alpha', {
  date: '2026-04-01',
  content: 'Project kickoff meeting',
});
await brain.addTimeline('project-alpha', {
  date: '2026-04-13',
  content: 'First prototype delivered',
});

🔌 MCP Server — Connect to AI Assistants

Works with Claude Desktop, Cursor, OpenClaw, and any MCP-compatible client:

deepbrain-mcp

12 tools available:

🌐 Web UI

Built-in browser interface — no separate frontend:

deepbrain serve --port 3000

Pages:

• 📚 /pages — Browse all knowledge
• 🔍 /search — Semantic + keyword search
• 🏷️ /tags — Tag cloud, tree, graph
• 📊 /stats — Knowledge base statistics

API endpoints:

• GET /api/pages — List all pages (JSON)
• GET /api/search?q=... — Search (JSON)
• GET /api/tags — Full tag graph (JSON)
• GET /api/tags/tree — Tag hierarchy (JSON)
• GET /api/tags/recommend?slug=... — Related pages (JSON)
• GET /api/tags/clusters — Knowledge clusters (JSON)

💤 Dream Cycle — Self-Maintaining Knowledge

Your brain maintenance runs on autopilot:

deepbrain dream

🧠 DeepBrain Health Report
   Pages: 156
   Chunks: 892 (890 embedded)
   Links: 234
   Tags: 67
   Timeline entries: 412
   ⚠️  2 chunks missing embeddings — re-embedding...
   ⚠️  3 orphan chunks found — cleaning up...
   ✅ Health check complete

✂️ Browser Clipper

Save any web page to your brain:

import { generateBookmarklet } from 'deepbrain';

const code = generateBookmarklet('http://localhost:3000');
// Drag to bookmark bar → click on any page → saved

🏗️ Architecture

deepbrain/
├── core/          # 🧠 Brain engine (PGLite + pgvector)
│   ├── brain.ts   #    CRUD, search, links, tags, timeline
│   ├── schema.ts  #    Database schema
│   └── types.ts   #    TypeScript interfaces
├── import/        # 📥 21 importers
├── export/        # 📤 5 export formats
├── tag-graph/     # 🏷️ Tag relationship engine
├── web/           # 🌐 Built-in Web UI + API
├── clipper/       # ✂️ Browser clipper
├── dream/         # 💤 Automated maintenance
├── mcp/           # 🔌 MCP server (12 tools)
└── cli            # 🖥️ Command-line interface

Stack: TypeScript · PGLite (in-process PostgreSQL) · pgvector · agentkits (multi-provider AI)

Design principles:

• Local-first — Everything runs in-process. No external database, no cloud dependency.
• Provider-agnostic — 7 embedding providers. Switch with one config change.
• Import everything, lock in nothing — 21 importers in, 5 export formats out.
• Knowledge grows — Tag Graph discovers relationships automatically. Dream Cycle maintains health.

CLI Reference

deepbrain init <provider>              # Initialize brain
deepbrain put <slug> <body>            # Add/update page
deepbrain get <slug>                   # Read page
deepbrain query <text>                 # Semantic search
deepbrain search <keyword>             # Keyword search
deepbrain tag <slug> <tag>             # Add tag
deepbrain link <from> <to>             # Create link
deepbrain timeline <slug> <date> <text> # Add timeline entry
deepbrain list [--type note]           # List pages
deepbrain stats                        # Show statistics
deepbrain dream                        # Run maintenance
deepbrain serve [--port 3000]          # Start Web UI
deepbrain export --format <fmt> --output <dir>  # Export

📄 License

MIT © Magicray1217

中文文档

问题

你的知识散落在各处——Notion、Obsidian、Evernote、Flomo、飞书、语雀、微信公众号、PDF……当你需要找某个东西时，你得先想起来放在哪了。

DeepBrain 把所有知识汇集到一处，用 AI 理解它们，让你用自然语言搜索全部知识。

为什么选 DeepBrain

⚡ 快速上手

# 安装
npm install -g deepbrain

# 初始化（选你喜欢的 AI 提供商）
deepbrain init gemini       # Google Gemini（推荐，有免费额度）
deepbrain init ollama       # 本地运行，零费用
deepbrain init dashscope    # 阿里通义千问
deepbrain init deepseek     # DeepSeek

# 写入你的第一条知识
deepbrain put 第一条笔记 "# 你好，大脑！
AI Agent 需要结构化的知识库才能工作得好。
标签: #AI #Agent #知识库"

# 用自然语言搜索
deepbrain query "AI Agent 需要什么？"
# → 返回: 第一条笔记 (score: 0.92)

# 打开 Web 界面
deepbrain serve
# → 🧠 http://localhost:3000

📥 从任何平台导入（21 个数据源）

开源生态中覆盖最全的知识导入器。

# 导入 Obsidian 笔记库
deepbrain import obsidian ./my-vault

# 导入 Notion 导出
deepbrain import notion ./notion-export

# 导入 Evernote
deepbrain import evernote ./my-notes.enex

import { importFlomo, importWechat, importObsidian } from 'deepbrain';

// 导入 Flomo（带进度）
const memos = await importFlomo({
  file: './flomo-export.html',
  onProgress: (current, total, name) => console.log(`[${current}/${total}] ${name}`),
});
console.log(`导入了 ${memos.length} 条 memo`);

支持的平台

国际: Notion · Obsidian · Evernote · Roam Research · Logseq · Bear · Apple Notes · Google Keep · OneNote · Joplin · Readwise · Day One

国内: 语雀 · 飞书 · 石墨 · 微信公众号 · Flomo · 我来(Wolai) · 息流(FlowUs) · 思源笔记

格式: EPUB · PDF

每个导入器处理平台特有的格式：

• Obsidian: [[双向链接]]、#嵌套/标签、YAML frontmatter
• Evernote: ENML→Markdown、checkbox 转换
• Flomo: #标签/子标签 层级、memo 时间戳
• 微信公众号: js_content 提取、og:title 解析
• 语雀: Admonition 清理、lakebook 格式

📤 导出（5 种格式）

数据是你的，随时带走：

deepbrain export --format markdown --output ./备份
deepbrain export --format obsidian --output ./我的笔记库     # 直接用 Obsidian 打开
deepbrain export --format logseq --output ./我的图谱         # 直接用 Logseq 打开
deepbrain export --format json --output ./data.json          # 程序处理用
deepbrain export --format html --output ./网站               # 静态网站

🏷️ 标签图谱 — 知识自动关联

标签不只是标记，它们构成了一张知识关系图，自动发现知识之间的联系。

import { Brain, TagGraph } from 'deepbrain';

const brain = new Brain({ embedding_provider: 'gemini', api_key: 'AIza...' });
await brain.connect();
const tg = new TagGraph(brain);

// "基于这篇笔记，我还应该看什么？"
const related = await tg.recommend('我的笔记', 5);
// → [{ slug: 'agent-设计', title: 'Agent 设计模式', score: 0.78, sharedTags: ['AI', 'AI/Agent'] }]

// "这篇笔记应该打什么标签？"
const suggested = await tg.suggestTags('无标题笔记');
// → ['TypeScript', 'API', '架构']

// "我的知识全景是什么样的？"
const graph = await tg.getGraph();
// → { nodes: [...], edges: [...], stats: { totalTags: 42, mostConnected: 'AI' } }

// "把知识自动分组"
const clusters = await tg.cluster();
// → [{ name: 'AI', tags: ['AI', 'LLM', 'Agent'], pages: [...], size: 15 }]

// 重命名标签（所有页面同步更新）
await tg.renameTag('机器学习', 'ML');

// 合并重复标签
await tg.mergeTags(['js', 'javascript', 'JS'], 'JavaScript');

层级标签自动展开：

deepbrain tag 我的笔记 "AI/Agent/RAG"
# 自动创建: AI → AI/Agent → AI/Agent/RAG

🔍 混合搜索

不只是向量相似度，三种搜索策略融合：

// 语义搜索：找概念相关的内容
const results = await brain.query('如何构建 AI Agent');

// 关键词搜索：精确匹配
const results2 = await brain.search('TypeScript');

📊 双轨存储

// 当前事实（可覆写）
await brain.put({
  slug: '项目-alpha',
  type: 'project',
  title: 'Project Alpha',
  compiled_truth: '状态: 进行中。团队: 5人。技术栈: TypeScript + React。',
});

// 历史事件（只追加）
await brain.addTimeline('项目-alpha', { date: '2026-04-01', content: '项目启动会议' });
await brain.addTimeline('项目-alpha', { date: '2026-04-13', content: '第一个原型交付' });

🔌 MCP Server（12 个工具）

接入 Claude Desktop、Cursor、OpenClaw：

deepbrain-mcp

工具: put · get · query · search · link · unlink · get-links · timeline-add · get-timeline · list · stats · dream

🌐 Web UI

内置浏览器界面，一行启动：

deepbrain serve --port 3000

• 📚 页面浏览 — 所有知识一览
• 🔍 搜索 — 语义 + 关键词
• 🏷️ 标签图谱 — Tag Cloud + 树形结构
• 📊 统计 — 页面/chunk/链接/标签数量
• 🔌 JSON API — /api/pages, /api/search, /api/tags

💤 梦境循环 — 自动维护

deepbrain dream

自动检测：过期页面、缺失 embedding 的 chunk、孤儿 chunk、死链。自动修复并报告。

🏗️ 架构

deepbrain/
├── core/          # 🧠 Brain 引擎（PGLite + pgvector）
├── import/        # 📥 21 个导入器
├── export/        # 📤 5 种导出格式
├── tag-graph/     # 🏷️ 标签关系引擎
├── web/           # 🌐 内置 Web UI + API
├── clipper/       # ✂️ 网页剪藏
├── dream/         # 💤 自动维护
├── mcp/           # 🔌 MCP Server（12 个工具）
└── cli            # 🖥️ 命令行工具

技术栈: TypeScript · PGLite（进程内 PostgreSQL）· pgvector · agentkits（多提供商 AI）

设计理念:

• 本地优先 — 一切在进程内运行，无外部数据库，无云依赖
• 提供商无关 — 7 个 embedding 提供商，一行配置切换
• 导入一切，不锁定任何 — 21 个导入器进，5 种格式出
• 知识会生长 — Tag Graph 自动发现关系，Dream Cycle 自动维护健康

CLI 命令参考

deepbrain init <provider>              # 初始化
deepbrain put <slug> <body>            # 添加/更新
deepbrain get <slug>                   # 读取
deepbrain query <text>                 # 语义搜索
deepbrain search <keyword>             # 关键词搜索
deepbrain tag <slug> <tag>             # 打标签
deepbrain link <from> <to>             # 创建链接
deepbrain timeline <slug> <date> <text> # 添加时间线
deepbrain list [--type note]           # 列表
deepbrain stats                        # 统计
deepbrain dream                        # 维护
deepbrain serve [--port 3000]          # 启动 Web UI
deepbrain export --format <格式> --output <目录>  # 导出

📄 开源协议

MIT © Magicray1217