MaxKB

1. Project/Product Overview

2. What does it mostly do?

MaxKB is a smart body platform for enterprise complete AI landing, not a single-function toolbox. Core competency stratification:

2.1 Knowledge Base Management (Building Enterprise Exclusive Brain)

2.2 RAG Retrieval Enhancement Generation Pipeline

MaxKB's RAG pipeline is divided into four stages: "ingestion → processing → retrieval → generation:

-Dual-mode search: full-text keyword search fusion with vector semantic matching. Reordering algorithm is supported

-Precise Control: You can configure the similarity threshold (0.75 is recommended), the TOP number of referenced segments, and the maximum number of characters.

-Generation Policy: Select "Optimize model generation" or "Return the original text directly" when the knowledge base hits. You can specify a default response when the knowledge base misses.

-Multi-way recall node:v2.10 LTS adds the output parameter "List of results that meet direct response"

2.3 Workflow Workflow Orchestration

MaxKB's built-in, powerful drag-and-drop workflow engine allows non-technical people to build automated AI processes:

-Node Type:AI dialogue, knowledge base retrieval, document label retrieval, multi-channel recall, problem optimization, judge, specified reply, document content extraction, text-to-speech, voice-to-text, picture understanding, video understanding, picture generation, text-to-video, picture-to-video, form collection, loop, variable splitting, variable aggregation, parameter extraction, variable assignment, MCP call, etc.

-Variable System: Supports global variables, session variables, and external parameters (displayed in tiles in the input box)

-Trigger Mode: supports timing trigger and event trigger

-Streaming Execution: displays the name of the executing node when the AI answers.

2.4 MCP Tools and Skills

-MCP service tool: interface with third-party service API to realize database query, mail sending and ERP/CRM system interaction. For example, Zhengzhou Customs builds a AI review intelligent body through MCP.

-Custom Script Tool: Use Python to write custom functions for data processing and logical judgment.

-Workflow Type Tools: new in v2.8.0, encapsulate the entire workflow as a reusable tool

-Skills: Introduced in v2.7.0, smart bodies call Skills autonomously, such as allowing OpenClaw to call the enterprise knowledge base, supporting multi-smart body collaboration, expert routing and permission isolation.

-Built-in 200 functions: common operations such as data cleaning and API calls

2.5 Multimodal Interaction

2.6 Zero Encoding Integration

- front-end embedding : copy the generated code and embed it into websites, public numbers, enterprise wechat, DingTalk, flying books, Slack, etc. in floating window/full screen/mobile mode

- Back-end docking : open RESTful API Skills interface, ERP/CRM and other systems can be directly called

- Third Party Application Integration (Professional Edition): Enterprise WeChat Application/WeChat Customer Service, DingTalk Application, Flying Book Application, WeChat Public Number

3. Applicable Scenario

4. Not quite the scene

5. Core Competence List

5.1 knowledge base ability

5.2 agent capability

Simple smart body based on the user's intention to ask questions, independent analysis of requirements and matching adaptation skills, MCP, smart body and other tools, automatically trigger calls.

Advanced intelligences (workflow orchestration), drag-and-drop creation, supporting 30 nodes, covering all processes such as AI dialogue, retrieval, judgment, extraction, generation, etc.

| Multimodal Dialog | Since v2.10 LTS, the AI dialog node fully supports multimodal input such as text and pictures. |

| External Parameters | User input parameters are displayed in tiles in the input box, which supports explicit/implicit switches. |

| Timed/Event Triggering | Supports timed triggering and event-triggered smart body execution |

| Embed third party | Full screen mode, mobile mode, floating window mode, custom portal icon, AI avatar, disclaimer |

| Long-term memory | Since v2.9.0, long-term memory is supported, and user context is maintained across sessions |

| Conversation log operation analysis | Operation monitoring statistics, user satisfaction trend, token consumption tracking |

5.3 model support

5.4 enterprise governance

6. Architecture/deployment/integration approach

Technical Architecture

Deployment Mode

Model integration mode

-Public cloud model: Direct connection through API Key (OpenAI, Claude, Tongyi Qiwen, DeepSeek, etc.)

-Local Private Model: Access through inference gateways such as Ollama, vLLM, and Xinference

-Multimodal model: Voice, visual and video models are connected through standard interfaces

-Embedding model:bge-m3, text2vec, etc., support local and cloud

System Requirements

How to use #7.

7.1 installation and deployment (3 minutes to start)

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

# 访问：http://你的服务器IP:8080
# 默认账号：admin
# 默认密码：MaxKB@123..

Chinese users can refer to offline installation document if Docker image Pull fails.

7.2 Three-step Opening AI Landing

Step 1: Access the big model

Log in to WebUI → System Settings → Model Management → Add Model → Select Supplier (such as DeepSeek, Qwen, OpenAI)→ Fill in API Key or configure local model address.

Step 2: Build a Knowledge Base

Knowledge Base → New Knowledge Base → Drag and drop uploaded documents (or enter URL to crawl) → System automatic segmentation, vectorization → Wait for "Ready" status.

Step 3: Create a smart body.

Smart body, new, select simple smart body, associate the newly created knowledge base, publish, share links or embed third-party systems.

7.3 Advance: The Progressive Route from RAG to Agent

基础 RAG 问答（知识库关联即用）
    ↓
高级检索优化（混合检索 + 重排序 + 术语库）
    ↓
工作流自动化（拖拽编排：检索→判断→工具调用→生成）
    ↓
MCP 工具集成（连接数据库、ERP、CRM 等外部系统）
    ↓
Skills 技能 + 多智能体协同（专家路由、任务分发）

7.4 Data Backup

# 备份数据目录
tar -czf maxkb_backup_$(date +%Y%m%d).tar.gz ~/.maxkb

# 恢复数据
tar -xzf maxkb_backup_20250101.tar.gz -C ~/

8. What can I say before sales

8.1 a sentence positioning

• * "MaxKB is the most easy-to-use enterprise-level intelligent body platform in China-in a word, Docker deployment, full-featured WebUI operation, gradual upgrade from RAG to Agent, has served 1000 enterprise customers. "**

8.2 customer pain points → solutions

8.3 Differentiated Selling Points

vs RAGFlow (most often compared)

Official Comparison Page: https://maxkb.cn/maxkb-vs-ragflow

Selection suggestion: If the enterprise goal is to use AI for customer service, knowledge service, and business collaboration as soon as possible, MaxKB is preferred. If the core goal is to understand in-depth documents and optimize RAG links, RAGFlow is more appropriate. The two are non-zero sum-MaxKB can be used as a platform, RAGFlow as an engine, and complementary use.

vs Dify

8.4 Customer Value Story Line

1. Cut in:"Do you have internal knowledge management or customer service system now? How about the efficiency of manual processing?"
2. Pain Point Resonance:"Knowledge is scattered everywhere, new employees cannot be found, old employees have no time to teach, and customer service repeatedly answers the same question......"
3. Quick Verification : Live Demonstration-Docker Deployment in One Minute → Drag in Several Documents → Question and Answer on the Spot, Customers See the Effect
4. Progressive route:"You can start with a department's knowledge base, spread it to the entire company, and then upgrade to an automated workflow."
5. Security centering :"The data is completely on your own server, supports offline installation, and seamlessly interfaces with the intranet environment."
6. Long-term guarantee :"v2.10 LTS version provides long-term maintenance, with new version iterations every month, as well as 400 customer service calls and original technical support."
7. Heavy Case :"The customer service of Wallace's thousands of stores is already in use, and your colleague XX Company is also our customer."

9. Frequently Asked Customer Questions

10. PoC Recommendations

Recommended PoC Direction: Intelligent Question Answering of Enterprise Internal Knowledge Base

1. Basic Question and Answer Verification... Create a simple intelligent body, associate the knowledge base, test typical questions... 0.5 days.........................................................................................

| 4. Effect Tuning | Configure Term Database, Adjust Retrieval Parameters, Optimize Prompt Words | 0.5 Days | Meet Accuracy Requirements |

| 5. System integration demonstration | Embedded into internal portal or enterprise WeChat to show actual use | 0.5 days | Complete system that can be demonstrated |

| 6. Evaluation Report | Statistics Retrieval Recall Rate, Answer Accuracy Rate, User Satisfaction | 0.5 Days | PoC Evaluation Report |

Validation Metrics:

-Knowledge base document upload success rate> 95%

-Retrieval recall> 85% (after optimization by term base)

-End-to-end answer accuracy> 80%

-Average response time <5 seconds

-Total time taken to deploy to demo state <3 days

PoC Advanced Direction: Intelligent Customer Service Workflow

For customer service scenario customers, an additional 2 days can be added:

-Build customer service workflow (intention judgment → knowledge base retrieval → conditional routing → multi-round dialogue)

-Docking enterprise WeChat/WeChat public number

-Statistics on self-service resolution and customer satisfaction

11. Risks and Considerations

12. My Pre-Sales Judgment

Recommendation: Highly recommended (domestic preferred for domestic enterprise knowledge bases and smart body scenarios, deploying the simplest enterprise-class AI platform)

Reason:

1. Extreme deployment experience: A 'docker run' command is used to complete the deployment. The built-in PostgreSQL pgvector does not need to install any additional components. This is one of the easiest enterprise AI platforms to deploy on the market.

1. Full-featured WebUI :RAG Q & A, workflow orchestration, MCP tool management, Skills skill configuration, knowledge base management, multi-modal configuration-all operations are completed on the web side, and business personnel can use them independently.

1. Progressive AI upgrade :RAG → Workflow → Agent clear path, enterprises can pilot a department first, verify the value and then gradually expand, risk control.

1. Optimal Chinese Ecology : Native Chinese UI, Chinese Documents, Chinese Community, Chinese Cases, Chinese Technical Support (400 Phone), Almost Zero Threshold for Domestic Developers.

1. Enterprise-level verification is sufficient :1000 paying customers cover 30 industries, and real cases from Wallace, Shenzhen Tong, many universities and government agencies are available for reference.

1. Long-term maintenance guarantee :LTS version is iterated on a monthly basis 500 contributors fly to cloud enterprise support without worrying about giving up the project.

1. The price is transparent and reasonable : The community version is free and has complete functions. The professional version 48000 permanent authorization (far lower than similar commercial products). The all-in-one machine 99000 includes a complete set of software and hardware solutions.

Recommended Customer Persona:

-Domestic enterprises (government, education, medical, manufacturing, retail, transportation and other industries)

-Small and medium-sized enterprises that want to land AI quickly but the technical team is not strong

-Compliance scenarios that require privatization deployment and data not out of the domain

-Take knowledge base question and answer or intelligent customer service as the main entry scene

-Enterprises with office ecology such as flying books/DingTalk/WeChat

-Requires original technical support and LTS long term maintenance

Not recommended situations:

-The core requirement is in-depth document analysis research (recommended RAGFlow)

-Requires Apache-2.0/MIT protocol for secondary commercial distribution (RAGFlow or Dify recommended)

-Pure English/overseas markets (recommended Haystack)

-Requires highly customized LLM application development (recommended LangChain)

-Multi-Agent complex academic research is a core requirement (Agno/AgentScope recommended)

13. REFERENCE

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

-Official website: https://maxkb.cn

-Documentation: https://maxkb.cn/docs/v2/

-Update log: https://maxkb.cn/docs/v2/changelog/

-Version comparison: https://maxkb.cn/price

-All-in-one: https://maxkb.cn/appliance

-MaxKB vs RAGFlow:https://maxkb.cn/maxkb-vs-ragflow

-Community Forum: https://bbs.fit2cloud.com/c/mk/11

-Fly to Cloud Blog (MaxKB Classification):https://blog.fit2cloud.com/categories/maxkb

-B station demo video: https://www.bilibili.com/video/BV1vpamzREeS/

-Training certification: https://edu.fit2cloud.com/

-Technical Support Email: support@fit2cloud.com

-Customer service telephone number: 400-052-0755

• analysis date: 2026-07-02 | data aging: GitHub information is pulled in real time, product functions are based on official document v2.10.2-lts *