https://github.com/deepleaper/deepbrain

🧠 Personal AI brain — 21 importers, hybrid search, knowledge graph, Web UI, MCP server, export. Import from Notion, Obsidian, Evernote, Roam, Flomo, Yuque, Feishu, and more.
https://github.com/deepleaper/deepbrain
agent ai brain chinese deepseek embedding evernote knowledge mcp memory notion obsidian ollama pglite pgvector search second-brain typescript vector
Last synced: about 1 month ago
JSON representation
🧠 Personal AI brain — 21 importers, hybrid search, knowledge graph, Web UI, MCP server, export. Import from Notion, Obsidian, Evernote, Roam, Flomo, Yuque, Feishu, and more.
Host: GitHub
URL: https://github.com/deepleaper/deepbrain
Owner: Deepleaper
License: apache-2.0
Created: 2026-04-12T23:53:46.000Z (about 1 month ago)
Default Branch: master
Last Pushed: 2026-04-16T12:58:21.000Z (about 1 month ago)
Last Synced: 2026-04-16T14:32:21.755Z (about 1 month ago)
Topics: agent, ai, brain, chinese, deepseek, embedding, evernote, knowledge, mcp, memory, notion, obsidian, ollama, pglite, pgvector, search, second-brain, typescript, vector
Language: TypeScript
Size: 304 KB
Stars: 3
Watchers: 0
Forks: 0
Open Issues: 0
Metadata Files:
- Readme: README.md
- Changelog: CHANGELOG.md
- Contributing: CONTRIBUTING.md
- License: LICENSE
Awesome Lists containing this project

README

          


# 🧠 DeepBrain

**Agent 记忆引擎 — 让 AI 越用越聪明**

[![npm](https://img.shields.io/npm/v/deepbrain)](https://www.npmjs.com/package/deepbrain)

[![License](https://img.shields.io/badge/License-Apache_2.0-blue.svg)](LICENSE)

[![TypeScript](https://img.shields.io/badge/TypeScript-5.7-blue)](https://www.typescriptlang.org/)

[快速开始](#快速开始) · [核心概念](#核心概念) · [API 文档](#api-文档) · [English](#english)



---

## 💡 一句话介绍

> **不是 RAG，不是知识库。DeepBrain 是 Agent 的大脑 — 会学习、会记忆、会进化。**

现有 Agent 框架的痛点：**每次对话都从零开始**。Agent 不记得昨天学到的，不记得上周犯的错，换个人接手一切归零。

DeepBrain 用三个 API 解决：

```

brain.learn()   → 把经验写入记忆

brain.recall()  → 需要时检索相关记忆  

brain.evolve()  → 定期提炼，知识越来越精而不是越来越多

```

## 🎯 和竞品的区别

| | Mem0 | LangChain Memory | Letta | **DeepBrain** |

|---|---|---|---|---|

| 定位 | 记忆存储 | 对话历史 | 有状态 Agent | **Agent 记忆引擎** |

| 核心能力 | CRUD | 窗口管理 | 状态机 | **learn/recall/evolve** |

| 知识提炼 | ❌ | ❌ | ❌ | ✅ **自动进化** |

| 记忆分层 | ❌ | ❌ | ✅ 3层 | ✅ **Core/Working/Archival** |

| 框架绑定 | 独立 | LangChain | Letta | **任意框架** |

| 本地运行 | 需要服务 | 内存 | 需要服务 | ✅ **PGLite 零依赖** |

**一句话：别人管存，DeepBrain 管"炼"。**

## 快速开始

```bash

npm install deepbrain

```

```typescript

import { Brain, AgentBrain } from 'deepbrain';

// 1. 初始化大脑

const brain = new Brain({ embedding_provider: 'openai' });

await brain.connect();

const agent = new AgentBrain(brain, 'my-sales-agent');

// 2. 学习 — Agent 每次交互后自动存储

await agent.learn({

  action: '客户问了退货政策',

  result: '7天无理由退货，需保留包装',

  context: { customer: 'VIP', channel: 'wechat' }

});

// 3. 回忆 — 需要时自动检索

const memories = await agent.recall('这个客户之前问过什么？');

// → [{ text: '客户问了退货政策...', score: 0.92 }]

// 4. 进化 — 定期运行，提炼知识

const report = await agent.evolve();

// → 把 50 条零散经验提炼成 3 条知识

```

## 核心概念

### 三个 API

| API | 做什么 | 什么时候调 |

|-----|-------|----------|

| `learn()` | 存储经验/知识 | 每次 Agent 完成任务后 |

| `recall()` | 语义检索记忆 | 每次 Agent 需要上下文时 |

| `evolve()` | 提炼+升级记忆 | 定时任务（如每天凌晨） |

### 记忆分层

```

Core     (≤20条)   — 始终在上下文中，最重要的知识

Working  (≤50条)   — 当前活跃，频繁访问的记忆

Archival (无限)    — 长期存储，按需检索

```

访问频率高 → 自动升级。长期不用 → 自动降级。

### 知识进化 (evolve)

这是 DeepBrain 最核心的差异化：

```

Day 1: 50条零散的客户对话记录

Day 7: evolve() → 提炼为 5 条客户偏好知识

Day 30: evolve() → 进一步精炼为 2 条核心洞察

```

**知识越来越精，而不是越来越多。**

### Memory Adapters

DeepBrain 不绑定任何框架。通过适配层对接：

```typescript

import { adapters } from 'deepbrain';

// OpenClaw 适配

const openclaw = adapters.openclaw;

// 原生适配（任何框架）

const native = adapters.native;

// 更多适配层开发中：LangChain, CrewAI, AutoGen...

```

## 全部特性

| 类别 | 特性 |

|------|------|

| 🧠 **记忆** | learn/recall/evolve 三件套、Memory Tiers 自动升降级 |

| 🔍 **搜索** | 向量语义搜索 + 全文检索、混合排序 |

| 🔄 **维护** | evolve 知识提炼、dream 自动维护周期 |

| 🔌 **适配** | OpenClaw / Native 适配层、可扩展接口 |

| 📥 **导入** | 24+ importers（Markdown/Notion/Obsidian/Logseq/Readwise...） |

| 🤖 **AI** | 7 providers（OpenAI/Gemini/DeepSeek/通义千问/智谱/Moonshot/Ollama） |

| 🛠️ **集成** | MCP Server、REST API、Web UI |

| 💾 **存储** | PGLite 本地零依赖、向量索引 |

## CLI 命令

```bash

deepbrain learn "客户偏好：周一上午开会"    # 学习

deepbrain recall "客户什么时候开会？"        # 检索

deepbrain evolve                            # 知识进化

deepbrain query "退货政策"                  # 语义搜索

deepbrain serve                             # Web UI + API

deepbrain import notion ./export            # 导入 Notion

deepbrain dream                             # 维护周期

deepbrain stats                             # 查看统计

```

## 架构

```

┌─────────────────────────────────────────┐

│         你的 Agent 框架                   │

│   (OpenClaw / LangChain / 自研 / ...)    │

├─────────────────────────────────────────┤

│         Memory Adapters                  │

│   (openclaw / langchain / native)        │

├─────────────────────────────────────────┤

│         AgentBrain API                   │

│   learn() · recall() · evolve()          │

├─────────────────────────────────────────┤

│         Brain Engine                     │

│   向量搜索 · FTS · 知识图谱 · 分层        │

├─────────────────────────────────────────┤

│         PGLite + Embeddings              │

│   本地存储 · 零依赖 · 零成本              │

└─────────────────────────────────────────┘

```

## 🔗 生态

DeepBrain 是跃盟开源四件套之一：

| 项目 | 定位 | 关系 |

|------|------|------|

| **DeepBrain** | Agent 记忆引擎 | ← 你在这里 |

| [opc-agent](https://github.com/Deepleaper/opc-agent) | Agent OS | Traces → learn() |

| [agentkits](https://github.com/Deepleaper/agentkits) | 带记忆的 OpenRouter | 自动 recall + learn |

| [agent-workstation](https://github.com/Deepleaper/agent-workstation) | 虚拟工位模板 | brain-seed.md → L1 |

```

opc-agent 收集 Traces → DeepBrain 学习 → agentkits 调用时自动 recall → Agent 越来越聪明

```

## License

Apache-2.0 — 免费商用，欢迎贡献。

---



## English

## 💡 What Is DeepBrain?

> **Not RAG. Not a knowledge base. DeepBrain is the brain for your AI Agent — it learns, remembers, and evolves.**

The pain point with existing Agent frameworks: **every conversation starts from scratch**. The agent doesn't remember what it learned yesterday, doesn't remember last week's mistakes, and when someone else takes over, everything resets to zero.

DeepBrain solves this with three APIs:

```

brain.learn()   → Write experience into memory

brain.recall()  → Retrieve relevant memories when needed

brain.evolve()  → Periodically refine — knowledge gets sharper, not bigger

```

## 🎯 Competitive Comparison

| | Mem0 | LangChain Memory | Letta | **DeepBrain** |

|---|---|---|---|---|

| Positioning | Memory storage | Conversation history | Stateful Agent | **Agent Memory Engine** |

| Core capability | CRUD | Window management | State machine | **learn/recall/evolve** |

| Knowledge refinement | ❌ | ❌ | ❌ | ✅ **Auto-evolve** |

| Memory tiers | ❌ | ❌ | ✅ 3 tiers | ✅ **Core/Working/Archival** |

| Framework lock-in | Standalone | LangChain | Letta | **Any framework** |

| Local-first | Requires service | In-memory | Requires service | ✅ **PGLite, zero dependencies** |

**In one line: others store memories; DeepBrain *refines* them.**

## Quick Start

```bash

npm install deepbrain

```

```typescript

import { Brain, AgentBrain } from 'deepbrain';

// 1. Initialize the brain

const brain = new Brain({ embedding_provider: 'openai' });

await brain.connect();

const agent = new AgentBrain(brain, 'my-sales-agent');

// 2. Learn — automatically store after each interaction

await agent.learn({

  action: 'Customer asked about return policy',

  result: '7-day no-questions-asked return, original packaging required',

  context: { customer: 'VIP', channel: 'wechat' }

});

// 3. Recall — retrieve when needed

const memories = await agent.recall('What did this customer ask before?');

// → [{ text: 'Customer asked about return policy...', score: 0.92 }]

// 4. Evolve — run periodically to refine knowledge

const report = await agent.evolve();

// → Distills 50 scattered experiences into 3 pieces of knowledge

```

## Core Concepts

### Three APIs

| API | What it does | When to call |

|-----|-------------|-------------|

| `learn()` | Store experience/knowledge | After each Agent task completion |

| `recall()` | Semantic memory retrieval | When the Agent needs context |

| `evolve()` | Refine + upgrade memories | Scheduled task (e.g., daily at midnight) |

### Memory Tiers

```

Core     (≤20 items)  — Always in context, most important knowledge

Working  (≤50 items)  — Currently active, frequently accessed memories

Archival (unlimited)  — Long-term storage, retrieved on demand

```

High access frequency → auto-promoted. Long unused → auto-demoted.

### Knowledge Evolution (evolve)

This is DeepBrain's core differentiator:

```

Day 1:  50 scattered customer conversation records

Day 7:  evolve() → Refined into 5 customer preference insights

Day 30: evolve() → Further distilled into 2 core insights

```

**Knowledge gets sharper, not bigger.**

### Memory Adapters

DeepBrain is framework-agnostic. Connect via adapters:

```typescript

import { adapters } from 'deepbrain';

// OpenClaw adapter

const openclaw = adapters.openclaw;

// Native adapter (any framework)

const native = adapters.native;

// More adapters in development: LangChain, CrewAI, AutoGen...

```

## Full Feature Set

| Category | Features |

|----------|----------|

| 🧠 **Memory** | learn/recall/evolve trio, Memory Tiers with auto-promotion/demotion |

| 🔍 **Search** | Vector semantic search + full-text search, hybrid ranking |

| 🔄 **Maintenance** | evolve knowledge refinement, dream auto-maintenance cycle |

| 🔌 **Adapters** | OpenClaw / Native adapter layer, extensible interface |

| 📥 **Import** | 24+ importers (Markdown/Notion/Obsidian/Logseq/Readwise...) |

| 🤖 **AI** | 7 providers (OpenAI/Gemini/DeepSeek/Qwen/GLM/Moonshot/Ollama) |

| 🛠️ **Integration** | MCP Server, REST API, Web UI |

| 💾 **Storage** | PGLite local zero-dependency, vector indexing |

## CLI Commands

```bash

deepbrain learn "Customer prefers Monday morning meetings"  # Learn

deepbrain recall "When does the customer have meetings?"     # Recall

deepbrain evolve                                             # Knowledge evolution

deepbrain query "return policy"                              # Semantic search

deepbrain serve                                              # Web UI + API

deepbrain import notion ./export                             # Import from Notion

deepbrain dream                                              # Maintenance cycle

deepbrain stats                                              # View statistics

```

## Architecture

```

┌─────────────────────────────────────────┐

│         Your Agent Framework             │

│   (OpenClaw / LangChain / Custom / ...) │

├─────────────────────────────────────────┤

│         Memory Adapters                  │

│   (openclaw / langchain / native)        │

├─────────────────────────────────────────┤

│         AgentBrain API                   │

│   learn() · recall() · evolve()          │

├─────────────────────────────────────────┤

│         Brain Engine                     │

│   Vector Search · FTS · Knowledge Graph  │

├─────────────────────────────────────────┤

│         PGLite + Embeddings              │

│   Local Storage · Zero Deps · Zero Cost  │

└─────────────────────────────────────────┘

```

## 🔗 Ecosystem

DeepBrain is part of Deepleaper's open-source suite of four:

| Project | Role | Relationship |

|---------|------|-------------|

| **DeepBrain** | Agent Memory Engine | ← You are here |

| [opc-agent](https://github.com/Deepleaper/opc-agent) | Agent OS | Traces → learn() |

| [agentkits](https://github.com/Deepleaper/agentkits) | OpenRouter with Memory | Auto recall + learn |

| [agent-workstation](https://github.com/Deepleaper/agent-workstation) | Virtual Role Templates | brain-seed.md → L1 |

```

opc-agent collects Traces → DeepBrain learns → agentkits auto-recalls during calls → Agents get smarter over time

```

## License

Apache-2.0 — Free for commercial use. Contributions welcome.
ecosyste.ms

Data

Tools

Indexes

Applications

Experiments

Awesome

https://github.com/deepleaper/deepbrain

Awesome Lists containing this project

README
