🧠 DeepBrain

Personal AI brain. Store, search, and grow your knowledge.

个人 AI 知识大脑。存储、搜索、生长你的知识。

English | 中文

English

DeepBrain is a personal knowledge engine inspired by GBrain, rebuilt from scratch with:

• 🔌 Multi-provider — 7 embedding providers via agentkits
• 🇨🇳 Chinese-friendly — DeepSeek, DashScope (Qwen), Zhipu (GLM) built-in
• 🏠 Local-first — Ollama + PGLite, zero cloud dependency
• ⚡ Lightweight — No bloat, just the essentials
• 🔍 Hybrid search — Vector (HNSW) + keyword (tsvector) + RRF fusion
• 📊 Compiled Truth + Timeline — Current facts vs. historical events
• 🔗 Knowledge graph — Bidirectional links between pages
• 💤 Dream Cycle — Automated knowledge maintenance
• 🔌 MCP Server — 12 tools for Claude / Cursor / OpenClaw
• 📥 Import — Notion & Obsidian vault importer

Quick Start

# Install
npm install -g deepbrain

# Initialize with your preferred provider
deepbrain init ollama          # Local, free
deepbrain init deepseek        # Cheap, fast
deepbrain init gemini          # Free tier available

# Add knowledge
deepbrain put meeting-notes notes.md
deepbrain put "john-doe" "John is a senior engineer at Google."

# Search
deepbrain query "who knows about distributed systems?"
deepbrain search "Google engineer"

# Import from Notion / Obsidian
deepbrain import notion ./notion-export/
deepbrain import obsidian ~/my-vault/

# Link & Timeline
deepbrain link john-doe google-company
deepbrain timeline john-doe "Promoted to Staff Engineer"

# Stats & maintenance
deepbrain stats
deepbrain dream

Programmatic API

import { Brain, importNotion, importObsidian } from 'deepbrain';

const brain = new Brain({
  embedding_provider: 'deepseek',
  api_key: process.env.DEEPSEEK_API_KEY,
});
await brain.connect();

// Store
await brain.put('ray', {
  type: 'person',
  title: 'Ray Wang',
  compiled_truth: 'CEO of Deepleaper. Expert in AI agents.',
});

// Semantic search
const results = await brain.query('who builds AI agents?');

// Import from Notion
const result = await importNotion(brain, './notion-export/');
console.log(`Imported ${result.imported} pages`);

// Import from Obsidian (auto-converts [[wikilinks]] to links)
const obsResult = await importObsidian(brain, '~/my-vault/');
console.log(`Imported ${obsResult.imported} pages`);

await brain.disconnect();

Providers

Import

Notion

Supports: Markdown export, HTML export, CSV database export.

# Export from Notion: Settings → Export → Markdown & CSV
deepbrain import notion ./notion-export/

import { importNotion } from 'deepbrain';
const result = await importNotion(brain, './notion-export/', {
  prefix: 'notion/',     // namespace slugs
  type: 'note',          // override type
  dryRun: true,          // preview without importing
  onProgress: (i, total, file) => console.log(`${i}/${total}: ${file}`),
});

Obsidian

Supports: Markdown, [[wikilinks]] → links, #tags, YAML frontmatter, nested folders, Chinese tags.

deepbrain import obsidian ~/my-vault/

import { importObsidian } from 'deepbrain';
const result = await importObsidian(brain, '~/my-vault/', {
  convertLinks: true,     // [[wikilinks]] → brain links (default)
  importTags: true,       // #inline-tags (default)
  ignoreFolders: ['.obsidian', '.trash'],
});

MCP Server (Claude / Cursor / OpenClaw)

DeepBrain ships with a built-in MCP server. Your AI assistant can read/write your brain directly.

{
  "mcpServers": {
    "deepbrain": {
      "command": "deepbrain-mcp",
      "env": { "DEEPBRAIN_CONFIG": "/path/to/deepbrain.json" }
    }
  }
}

12 MCP Tools

How It Works

Knowledge → Pages (Markdown) → Chunks → Embeddings → Vector DB (PGLite)
               ↓                                        ↓
            Links (Graph)                         Hybrid Search
               ↓                                (HNSW + tsvector + RRF)
           Timeline (append-only)                      ↓
               ↓                                   Results
         Dream Cycle (maintenance)

vs. GBrain

Roadmap

• Core engine (CRUD, search, graph, timeline)
• Hybrid search (vector + keyword + RRF)
• CLI
• Dream Cycle v1
• Multi-provider embedding (via agentkits)
• MCP Server (12 tools)
• Notion importer
• Obsidian importer
• 43 tests (brain, dream, import, types)
• Web UI (knowledge graph visualization)
• Import from WeChat / Feishu
• MRG layer (mission-centric knowledge graph)
• Smart chunking (semantic + LLM-assisted)
• Multi-brain sync

中文文档

DeepBrain 是一个个人 AI 知识大脑，灵感来自 GBrain，从零重建，特点：

• 🔌 多模型支持 — 7 个 Embedding 提供商，通过 agentkits
• 🇨🇳 中文友好 — 内置 DeepSeek、通义千问 (DashScope)、智谱 (GLM)
• 🏠 本地优先 — Ollama + PGLite，零云端依赖
• ⚡ 轻量 — 无臃肿依赖，只保留核心功能
• 🔍 混合搜索 — 向量 (HNSW) + 关键词 (tsvector) + RRF 融合排序
• 📊 编译真相 + 时间线 — 当前事实 vs. 历史事件，分开管理
• 🔗 知识图谱 — 页面间双向链接
• 💤 梦境循环 — 自动化知识库维护（清理孤页、刷新过期嵌入）
• 🔌 MCP 服务器 — 12 个工具，让 Claude / Cursor 直接读写你的知识库
• 📥 导入 — 支持 Notion 和 Obsidian 笔记库

快速上手

# 安装
npm install -g deepbrain

# 初始化（选一个你喜欢的模型）
deepbrain init ollama          # 本地运行，免费
deepbrain init deepseek        # 便宜又快
deepbrain init gemini          # 有免费额度

# 存入知识
deepbrain put 会议纪要 notes.md
deepbrain put "张三" "张三是谷歌的资深工程师，擅长分布式系统。"

# 搜索
deepbrain query "谁懂分布式系统？"
deepbrain search "谷歌工程师"

# 从 Notion / Obsidian 导入
deepbrain import notion ./notion-export/
deepbrain import obsidian ~/my-vault/

# 关联与时间线
deepbrain link 张三 谷歌
deepbrain timeline 张三 "晋升为 Staff Engineer"

# 统计与维护
deepbrain stats
deepbrain dream

编程接口

import { Brain, importNotion, importObsidian } from 'deepbrain';

const brain = new Brain({
  embedding_provider: 'deepseek',    // 或 'ollama', 'gemini', 'dashscope', 'zhipu'
  api_key: process.env.DEEPSEEK_API_KEY,
});
await brain.connect();

// 存入
await brain.put('ray', {
  type: 'person',
  title: '王冉',
  compiled_truth: '跃盟科技 CEO，AI 情景智能专家。',
});

// 语义搜索
const results = await brain.query('谁做 AI Agent？');
console.log(results[0].slug); // → 'ray'

// 从 Notion 导入
const result = await importNotion(brain, './notion-export/');
console.log(`导入了 ${result.imported} 页`);

// 从 Obsidian 导入（自动将 [[双链]] 转为知识图谱链接）
const obsResult = await importObsidian(brain, '~/my-vault/');
console.log(`导入了 ${obsResult.imported} 页`);

await brain.disconnect();

支持的模型

导入功能

从 Notion 导入：

1. Notion → 设置 → 导出 → Markdown & CSV 格式
2. deepbrain import notion ./导出目录/
3. 支持 Markdown、HTML、CSV 三种格式

从 Obsidian 导入：

1. deepbrain import obsidian ~/我的知识库/
2. 自动将 [[双链]] 转为知识图谱链接
3. 自动提取 #标签（支持中文标签如 #人工智能）
4. 解析 YAML frontmatter
5. 忽略 .obsidian 等系统目录

MCP 服务器

让 Claude / Cursor / OpenClaw 直接读写你的知识库：

{
  "mcpServers": {
    "deepbrain": {
      "command": "deepbrain-mcp"
    }
  }
}

配置好后，你可以直接对 Claude 说：

你： 我关于 Ray 知道什么？

Claude： 调用 brain_query("Ray") → 找到：王冉，跃盟科技 CEO，AI 情景智能专家。

你： 记录一下他昨天在 AI 峰会上做了演讲。

Claude： 调用 brain_timeline("ray", "AI 峰会演讲") → ✅ 已添加到时间线。

核心概念

与 GBrain 对比

Architecture / 架构

deepbrain
├── src/
│   ├── index.ts         # 主入口
│   ├── cli.ts           # CLI 命令行工具
│   ├── mcp.ts           # MCP 服务器 (12 tools)
│   ├── core/
│   │   ├── brain.ts     # 核心引擎 (CRUD + 搜索 + 图谱)
│   │   ├── schema.ts    # PGLite 数据库 schema
│   │   └── types.ts     # TypeScript 类型定义
│   ├── dream/
│   │   └── index.ts     # 梦境循环 (自动维护)
│   └── import/
│       ├── index.ts     # 导入模块入口
│       ├── notion.ts    # Notion 导入器
│       └── obsidian.ts  # Obsidian 导入器
├── tests/               # 43 个测试
├── package.json
└── tsconfig.json

License / 开源协议

MIT © Magicray1217

灵感来自 GBrain by Garry Tan。基于 AgentKits 构建。