简介

MaxKB​(Max Knowledge Brain)是一款开源的企业级智能体平台,专为构建和部署高性能AI代理而设计。由1Panel团队开发,MaxKB集成了先进的检索增强生成(RAG)管道、强大的工作流引擎和MCP工具使用能力,使企业能够轻松创建和部署智能客服、知识管理系统和学术研究工具。该平台支持多模态输入输出,兼容多种大语言模型,并提供无缝的第三方系统集成能力。

🔗 ​GitHub地址​:

https://github.com/1Panel-dev/MaxKB

⚡ ​核心价值​:

企业级智能体 · RAG增强 · 多模态支持


解决的企业痛点

传统企业AI实施痛点

MaxKB解决方案

知识管理效率低下

RAG管道自动处理文档,减少幻觉

系统集成复杂

零代码快速集成第三方系统

模型选择受限

支持私有和公有模型,灵活选择

多模态支持不足

原生支持文本、图像、音频、视频

部署和维护成本高

容器化部署,降低运维复杂度

定制化需求难以满足

强大工作流引擎支持复杂业务场景


核心功能架构

1. ​系统架构概览

2. ​功能矩阵

功能模块

核心能力

技术实现

RAG管道

文档上传、自动爬取、文本分割、向量化

自动处理 + 向量数据库

工作流引擎

可视化流程编排,复杂业务逻辑支持

拖拽界面 + 条件逻辑

模型管理

支持多种私有和公有模型

统一API接口 + 适配器模式

多模态支持

文本、图像、音频、视频输入输出

多媒体处理管道

第三方集成

零代码快速集成业务系统

REST API + Webhook

知识管理

企业知识库构建和维护

版本控制 + 权限管理

3. ​支持模型

  • 私有模型: DeepSeek, Llama, Qwen, ChatGLM, Baichuan

  • 公有模型: OpenAI GPT系列, Anthropic Claude, Google Gemini

  • 多模态模型: 支持图像、音频、视频处理的专用模型


安装与配置

1. ​Docker快速部署

# 使用官方Docker镜像一键部署
docker run -d \
  --name=maxkb \
  --restart=always \
  -p 8080:8080 \
  -v ~/.maxkb:/opt/maxkb \
  1panel/maxkb:latest

# 访问地址: http://your-server-ip:8080
# 默认登录凭据:
# 用户名: admin
# 密码: MaxKB@123..

2. ​高级配置选项

# 自定义端口和数据目录
docker run -d \
  --name=maxkb \
  -p 9090:8080 \
  -v /path/to/data:/opt/maxkb \
  -e MAXKB_HOST=0.0.0.0 \
  -e MAXKB_PORT=8080 \
  1panel/maxkb:latest

# 环境变量配置
MAXKB_HOST=0.0.0.0
MAXKB_PORT=8080
MAXKB_DATA_DIR=/opt/maxkb
MAXKB_LOG_LEVEL=INFO

3. ​Kubernetes部署

# maxkb-deployment.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: maxkb
spec:
  replicas: 3
  selector:
    matchLabels:
      app: maxkb
  template:
    metadata:
      labels:
        app: maxkb
    spec:
      containers:
      - name: maxkb
        image: 1panel/maxkb:latest
        ports:
        - containerPort: 8080
        volumeMounts:
        - name: data
          mountPath: /opt/maxkb
        env:
        - name: MAXKB_HOST
          value: "0.0.0.0"
        - name: MAXKB_PORT
          value: "8080"
      volumes:
      - name: data
        persistentVolumeClaim:
          claimName: maxkb-pvc
---
apiVersion: v1
kind: Service
metadata:
  name: maxkb-service
spec:
  selector:
    app: maxkb
  ports:
  - port: 80
    targetPort: 8080
  type: LoadBalancer

4. ​离线安装

对于中国用户或网络受限环境,可以使用离线安装方式:

# 下载离线安装包
wget https://github.com/1Panel-dev/MaxKB/releases/download/v2.0.0/maxkb-offline-installer.tar.gz

# 解压并安装
tar -zxvf maxkb-offline-installer.tar.gz
cd maxkb-offline-installer
./install.sh

# 按照提示完成安装

使用指南

1. ​初始设置

# 首次访问Web界面
http://your-server-ip:8080

# 使用默认凭据登录
用户名: admin
密码: MaxKB@123..

# 修改默认密码
1. 登录后进入"系统设置"
2. 选择"安全设置"
3. 更新密码并保存

2. ​知识库配置

# 通过API创建知识库
import requests

url = "http://localhost:8080/api/knowledge-base/"
headers = {"Content-Type": "application/json"}
data = {
    "name": "企业产品文档",
    "description": "公司产品手册和技术文档",
    "vector_type": "chroma",
    "embedding_model": "text2vec"
}

response = requests.post(url, json=data, headers=headers)
knowledge_base_id = response.json()["id"]

3. ​文档上传与处理

# 上传文档到知识库
files = {"file": open("product-manual.pdf", "rb")}
data = {
    "knowledge_base_id": knowledge_base_id,
    "process_type": "auto"
}

response = requests.post(
    "http://localhost:8080/api/document/upload",
    files=files,
    data=data
)

# 监控处理状态
document_id = response.json()["id"]
status_url = f"http://localhost:8080/api/document/{document_id}/status"

while True:
    status_response = requests.get(status_url)
    status = status_response.json()["status"]
    if status == "completed":
        break
    time.sleep(5)

4. ​模型配置

# 配置OpenAI模型
model_data = {
    "name": "GPT-4",
    "type": "openai",
    "api_key": "sk-your-openai-key",
    "model_name": "gpt-4",
    "base_url": "https://api.openai.com/v1"
}

response = requests.post(
    "http://localhost:8080/api/model/",
    json=model_data,
    headers=headers
)

# 配置私有模型
private_model_data = {
    "name": "本地Llama模型",
    "type": "llama",
    "model_path": "/models/llama-7b",
    "api_base": "http://localhost:8000/v1"
}

response = requests.post(
    "http://localhost:8080/api/model/",
    json=private_model_data,
    headers=headers
)

5. ​工作流创建

# 创建智能客服工作流
workflow_data = {
    "name": "智能客服流程",
    "description": "处理客户咨询的自动化流程",
    "nodes": [
        {
            "type": "start",
            "id": "start",
            "position": {"x": 100, "y": 100}
        },
        {
            "type": "llm",
            "id": "classify",
            "model": "gpt-4",
            "prompt": "分类用户问题类型: {user_input}",
            "position": {"x": 300, "y": 100}
        },
        {
            "type": "knowledge",
            "id": "retrieve",
            "knowledge_base_id": knowledge_base_id,
            "query": "{user_input}",
            "position": {"x": 500, "y": 100}
        },
        {
            "type": "llm",
            "id": "respond",
            "model": "gpt-4",
            "prompt": "基于以下信息回答用户问题: {knowledge_result}",
            "position": {"x": 700, "y": 100}
        }
    ],
    "edges": [
        {"source": "start", "target": "classify"},
        {"source": "classify", "target": "retrieve"},
        {"source": "retrieve", "target": "respond"}
    ]
}

response = requests.post(
    "http://localhost:8080/api/workflow/",
    json=workflow_data,
    headers=headers
)

应用场景实例

案例1:智能客服系统

场景​:电商公司需要处理大量客户咨询

解决方案​:

# 配置客服知识库
kb_data = {
    "name": "客服知识库",
    "description": "产品信息、退换货政策、常见问题",
    "sources": [
        {"type": "file", "path": "product-info.pdf"},
        {"type": "web", "url": "https://help.company.com"},
        {"type": "database", "connection": "mysql://user:pass@db/help_articles"}
    ]
}

# 创建客服工作流
workflow = {
    "name": "客户咨询处理",
    "steps": [
        {
            "name": "意图识别",
            "type": "llm",
            "model": "gpt-4",
            "prompt": "识别用户意图: {query}"
        },
        {
            "name": "知识检索",
            "type": "retrieval",
            "knowledge_base": "客服知识库",
            "query": "{identified_intent}"
        },
        {
            "name": "响应生成",
            "type": "llm",
            "model": "gpt-4",
            "prompt": "基于知识回答: {retrieved_knowledge}"
        },
        {
            "name": "满意度检查",
            "type": "llm",
            "model": "gpt-4",
            "prompt": "检查用户是否满意: {response}"
        }
    ]
}

# 集成到现有系统
integration = {
    "type": "webhook",
    "endpoint": "https://crm.company.com/api/ai-responses",
    "events": ["new_query", "response_ready"]
}

成效​:

  • 客服效率 ​提升300%​

  • 响应准确率 ​达到95%​

  • 客户满意度 ​提升40%​

案例2:企业内部知识库

场景​:大型企业需要集中管理分散的知识资源

工作流​:

# 知识库配置
knowledge_base:
  name: "企业综合知识库"
  sources:
    - type: "confluence"
      url: "https://confluence.company.com"
      spaces: ["TECH", "HR", "FINANCE"]
    - type: "sharepoint"
      site: "company.sharepoint.com"
      libraries: ["Documents", "Policies"]
    - type: "github"
      repos: ["company/docs", "company/wiki"]
  processing:
    chunk_size: 1000
    overlap: 200
    embedding_model: "text2vec-large"

# 访问控制
access_control:
  groups:
    - name: "全体员工"
      permissions: ["read"]
      sources: ["HR", "Policies"]
    - name: "技术团队"
      permissions: ["read", "write"]
      sources: ["TECH", "Documents"]
    - name: "管理层"
      permissions: ["read", "write", "admin"]
      sources: ["*"]

# 自动化维护
automation:
  sync_schedule: "0 2 * * *"  # 每天凌晨2点同步
  dead_link_check: true
  content_validation: true

价值​:

  • 知识查找时间 ​从小时级→秒级

  • 信息一致性 ​100%保证

  • 新员工培训 ​效率提升60%​

案例3:学术研究助手

场景​:研究机构需要辅助文献调研和论文写作

配置方案​:

# 学术知识库配置
academic_config = {
    "name": "学术研究库",
    "sources": [
        {
            "type": "arxiv",
            "categories": ["cs.AI", "cs.LG", "cs.CL"],
            "max_papers": 10000
        },
        {
            "type": "pubmed",
            "keywords": ["machine learning", "deep learning"],
            "years": ["2020-2024"]
        },
        {
            "type": "ieee",
            "topics": ["natural language processing", "computer vision"]
        }
    ],
    "processing": {
        "chunk_size": 500,
        "overlap": 100,
        "embedding_model": "all-mpnet-base-v2"
    }
}

# 研究助手工作流
research_workflow = {
    "name": "文献调研助手",
    "steps": [
        {
            "name": "主题分析",
            "type": "llm",
            "model": "gpt-4",
            "prompt": "分析研究主题并提取关键词: {research_topic}"
        },
        {
            "name": "文献检索",
            "type": "retrieval",
            "knowledge_base": "学术研究库",
            "query": "{keywords}",
            "filters": {"year": "2020-2024", "citation_count": ">100"}
        },
        {
            "name": "摘要生成",
            "type": "llm",
            "model": "gpt-4",
            "prompt": "生成文献综述摘要: {retrieved_papers}"
        },
        {
            "name": "参考文献整理",
            "type": "llm",
            "model": "gpt-4",
            "prompt": "格式化参考文献: {papers}"
        }
    ]
}

效益​:

  • 文献调研时间 ​减少70%​

  • 研究质量 ​显著提高

  • 论文引用 ​准确性100%​


高级功能与定制

1. ​多模态处理

# 图像处理配置
image_config = {
    "processors": [
        {
            "type": "ocr",
            "engine": "tesseract",
            "languages": ["eng", "chi_sim"]
        },
        {
            "type": "object_detection",
            "model": "yolov8",
            "confidence_threshold": 0.7
        }
    ]
}

# 音频处理配置
audio_config = {
    "processors": [
        {
            "type": "speech_to_text",
            "model": "whisper",
            "version": "large-v3"
        },
        {
            "type": "audio_analysis",
            "features": ["speech_rate", "emotion", "keywords"]
        }
    ]
}

# 视频处理配置
video_config = {
    "processors": [
        {
            "type": "frame_extraction",
            "interval": 1  # 每秒提取一帧
        },
        {
            "type": "scene_detection",
            "threshold": 0.5
        }
    ]
}

2. ​高级RAG优化

# RAG优化配置
rag_optimization = {
    "chunking_strategy": "semantic",
    "embedding_models": [
        {"name": "text2vec", "dimension": 768},
        {"name": "all-mpnet", "dimension": 768},
        {"name": "bge-large", "dimension": 1024}
    ],
    "retrieval_strategies": [
        {"type": "dense", "weight": 0.7},
        {"type": "sparse", "weight": 0.2},
        {"type": "hybrid", "weight": 0.1}
    ],
    "reranking": {
        "enabled": true,
        "model": "bge-reranker-large",
        "top_k": 50
    },
    "query_expansion": {
        "enabled": true,
        "model": "gpt-3.5-turbo",
        "max_expansions": 3
    }
}

3. ​企业级安全

# 安全配置
security:
  authentication:
    providers:
      - type: "ldap"
        server: "ldap://company.com"
        base_dn: "dc=company,dc=com"
      - type: "saml"
        idp_metadata_url: "https://sso.company.com/metadata"
      - type: "oauth2"
        providers: ["google", "microsoft", "github"]
  
  authorization:
    rbac: true
    roles: ["viewer", "editor", "admin", "superadmin"]
    permissions:
      viewer: ["read"]
      editor: ["read", "write"]
      admin: ["read", "write", "manage"]
      superadmin: ["read", "write", "manage", "admin"]
  
  data_protection:
    encryption:
      at_rest: true
      in_transit: true
    masking: true
    anonymization: true
  
  audit_logging:
    enabled: true
    retention: "365d"
    events: ["login", "query", "modification"]

生态系统集成

1. ​第三方工具集成

# MCP工具配置
mcp_tools = {
    "calculator": {
        "type": "math",
        "operations": ["add", "subtract", "multiply", "divide"]
    },
    "web_search": {
        "type": "search",
        "engines": ["google", "bing", "duckduckgo"]
    },
    "database": {
        "type": "sql",
        "connections": [
            {"name": "customer_db", "url": "mysql://user:pass@customer-db"},
            {"name": "product_db", "url": "postgresql://user:pass@product-db"}
        ]
    },
    "api_client": {
        "type": "http",
        "endpoints": [
            {"name": "crm_api", "base_url": "https://crm.company.com/api"},
            {"name": "erp_api", "base_url": "https://erp.company.com/api"}
        ]
    }
}

2. ​CI/CD集成

# GitHub Actions工作流
name: Deploy MaxKB

on:
  push:
    branches: [main]
  pull_request:
    branches: [main]

jobs:
  test:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Run tests
        run: docker-compose run --rm maxkb pytest tests/
  
  deploy:
    runs-on: ubuntu-latest
    needs: test
    steps:
      - uses: actions/checkout@v4
      - name: Deploy to production
        run: |
          docker-compose -f docker-compose.prod.yml up -d
        env:
          MAXKB_VERSION: ${{ github.sha }}

3. ​监控与告警

# Prometheus监控配置
monitoring:
  enabled: true
  port: 9090
  metrics:
    - name: "requests_total"
      type: "counter"
      help: "Total number of requests"
    - name: "request_duration_seconds"
      type: "histogram"
      help: "Request duration in seconds"
    - name: "knowledge_base_size"
      type: "gauge"
      help: "Size of knowledge base in documents"
  
  alerts:
    - alert: "HighErrorRate"
      expr: "rate(requests_errors_total[5m]) / rate(requests_total[5m]) > 0.05"
      for: "5m"
      labels:
        severity: "critical"
      annotations:
        summary: "High error rate detected"
    
    - alert: "SlowResponse"
      expr: "histogram_quantile(0.95, rate(request_duration_seconds_bucket[5m])) > 2"
      for: "5m"
      labels:
        severity: "warning"
      annotations:
        summary: "Slow response time detected"

🚀 ​GitHub地址​:

https://github.com/1Panel-dev/MaxKB

📊 ​性能数据​:

支持1000+并发请求 · RAG准确率95%+ · 企业级可靠性

MaxKB正在重新定义企业智能体开发——通过提供强大易用的开源平台,它让企业能够快速构建和部署AI代理系统。正如用户反馈:

"从概念到生产部署只需几天而非数月,MaxKB让AI代理开发变得简单而高效"

该平台已被电商、金融、教育、研究机构广泛采用,日均处理 ​超过100万次​ 知识查询,成为企业智能化的核心基础设施。

Logo

更多推荐