Seamless Integration, Instant Intelligence[1]: MaxKB Knowledge Base Q&A System, Zero Coding Embedded in Third-Party Business System, Customized Exclusive Intelligent Solutions, Soaring User Satisfaction

1. Introduction

MaxKB (Max Knowledge Base) is an open source knowledge base quiz system based on the LLM Large Language Model.

FuzzyCloud is a leading open source software company in China. Its open source products include 1Panel, JumpServer, DataEase, MeterSphere, Halo, MaxKB, etc., covering various fields such as operation and maintenance panel, operation and maintenance audit, BI analysis, software testing, CMS, knowledge base and Q&A. MaxKB open source knowledge base Q&A system, covering operation and maintenance panel, operation and maintenance audit, BI analysis, software testing, CMS building, knowledge base Q&A and other fields. FizzCloud's open source projects have shown excellent growth in the open source community, and have received more than 120,000 stars on GitHub, the code hosting platform.

Official Website:/
GitHub：/1Panel-dev/MaxKB

1.1 Product Advantages

Out-of-the-box: support for direct uploading of documents, automatic crawling of online documents, support for automatic text splitting, vectorization, RAG (Retrieval Augmented Generation), intelligent Q&A interactive experience is good;

Seamless Embedding: Support zero-code rapid embedding into third-party business systems, so that existing systems quickly have intelligent Q&A capabilities to improve user satisfaction;

Flexible scheduling: Built-in powerful workflow engine supports scheduling AI workflow to meet the needs of complex business scenarios;

Model Neutrality: Support docking various big language models, including local private big models (Llama 3 / Qwen 2, etc.), domestic public big models (Tongyi Qianqian / Wisdom Spectrum AI / Baidu Qianfan / Kimi / DeepSeek, etc.) and foreign public big models (OpenAI / Azure OpenAI / Gemini, etc.).

Remember to use the latest version!

new feature
- [Knowledge Base] Supports the creation of an empty knowledge base.
- [Knowledge Base] Supports the selection of different vector models for vectorization when creating and setting up a knowledge base.
- Model Management] Support adding OpenAI, Ollama and local vector models.
- [Model Management] Supports setting model permissions as public or private.
- [Application] The specified response node of advanced orchestration supports shortcut question output: shortcut question.
- 【Application】Support to set the entry icon of dialog box in floating window mode. (X-Pack)
- [Application] Supports customizing the avatar for AI replies to dialog boxes. (X-Pack)
- 【Application】Support to set whether the dialog box in floating window mode can be dragged and dropped position. (X-Pack)
- [Application] Support to set whether or not the dialog box displays the history. (X-Pack)
- [System Settings] LDAP single sign-on is supported. (X-Pack)
- 【System Settings】Support to customize the theme appearance, set the system's website logo, login logo, theme color, login background image, website name, welcome message and so on. (X-Pack)
- [System Settings] Open System API.(X-Pack)
Function Optimization
- [Knowledge Base] Optimize the document indexing process to improve knowledge recall.
- [Knowledge Base] Adjust the maximum length of segmented content to 100,000 characters.
- [Application] Only knowledge bases that use the same vector model can be selected when adjusting the associated knowledge base.
- [Application] Display knowledge sources adjusted to the display settings.
- [Model Management] Optimize the model list style.
- [About] Optimize the information displayed on the About page.

1.2 Version Features VS Version Comparison

Knowledge base management

application management

Large Language Model Docking

System administration and setup

Installation deployment/service specifications/product pricing

1.3 MaxKB System Architecture

Overall structure

Implementation Principle

technology stack
- Front-end:, logicflow
- Backend: Python / Django
- Langchain：Langchain
- Vector Databases: PostgreSQL / pgvector
- Big Models: Ollama, Azure OpenAI, OpenAI, Tongyi Thousand Questions, Kimi, Baidu Thousand Sails Big Models, Xunfei Starfire, Gemini, DeepSeek, and more.

2. Installation and deployment

2.1 Offline installation

Environmental requirements

Make sure your system meets the installation conditions before installing:

OS: Ubuntu 22.04 / CentOS 7 64-bit;
CPU / RAM: 2C/4GB or more recommended;
Disk space: 100GB;
Browser Requirements: Please use modern browsers such as Chrome, FireFox, Edge and others;
Offline deployment is recommended for production environments。
Offline deployment [✔]

Note: The offline package only supports x86 servers.

Open the community website to download the MaxKB offline packageCommunity Edition Offline Package

After uploading to the server for unzipping, execute the following command:

#maxkb-v1.2. replace with the name of the downloaded package
tar -zxvf maxkb-v1.2.

To install MaxKB, execute the following command:

# Go to the directory where the installation package was extracted
cd maxkb-v1.2.0-offline

#Execute the installation command
bash

If you encounter problems related to docker installation you can refer to the following link:

An article to get you started with the vector database milvus: with docker installation, milvus installation and use, attu visualization, a complete guide to start Milvus performed a vector similarity search

After successful installation, MaxKB can be accessed through your browser:

http://target server IP Address:8080

Default Login Information
Username: admin
Default Password:MaxKB@123...

Offline Upgrade

together with2 Offline deployment The steps are the same.

2.2 Online installation

Environmental requirements

Make sure your system meets the installation conditions before installing:

OS: Ubuntu 22.04 / CentOS 7 64-bit;
CPU / RAM: 2C/4GB or more recommended;
Disk space: 100GB;
Browser Requirements: Please use modern browsers such as Chrome, FireFox, Edge and others;
Internet access。
Online Rapid Deployment

MaxKB supports one-click startup by simply executing the following command:

docker run -d --name=maxkb -p 8080:8080 -v ~/.maxkb:/var/lib/postgresql/data cr2./1panel/maxkb

Friendly note: The Community Edition is limited to 1 team member, 5 apps, and 50 knowledge bases.

After successful installation, MaxKB can be accessed through your browser:

http://target server IP Address:Destination Port

Default login information
User name: admin
Default Password: MaxKB@123...

If you are using a cloud server, go to the security group to open the target port.

Online Upgrade

Note: Confirm the data persistence directory (the directory after -v) before upgrading, and keep the same data persistence directory as the previous one when creating a new container, otherwise the data will be empty after startup.
Execute the following command:

1 Download the latest image

docker pull cr2./1panel/maxkb

2 Confirm the last data persistence directory, copy and save it, and use it in step 4.

3 Deleting a Running MaxKB Container

4 Create and start the MaxKB container

docker run -d --name=maxkb -p 8080:8080 -v /opt/maxkb/pgsql/data:/var/lib/postgresql/data cr2./1panel/maxkb

#Note: Make sure the data persistence directory (the directory after -v) is the same as the directory in [Step 2], otherwise the data will be empty after startup.

2.3 1Panel Installation

For more information about 1Panel installation and deployment, please refer to 1Panel official documentation. After you have finished installing and deploying 1Panel, open the browser to enter 1Panel according to the prompted URL, the following interface will be displayed.

Installation of MaxKB

Go to the list of apps in the App Store and find the MaxKB app under the [AI / Big Model] category and install it.

Select the latest MaxKB version on the application details page to install it and set the relevant parameters.

Name: The name of the MaxKB application to be created.
Version: Select the latest version of MaxKB.
Port: The service port for the MaxKB application.
Container name: MaxKB application container name.
CPU Limit: The number of CPU cores that MaxKB application can use.
Memory Limit: MaxKB The amount of memory the application can use.
Port External Access: MaxKB applications can be accessed using IP:PORT (MaxKB applications must have external port access open).

After clicking Start Installation, the page automatically jumps to the list of installed apps, and the status of MaxKB apps waiting to be installed changes to Started.

Access to MaxKB

After successful installation, log in to MaxKB by accessing the following page from your browser:

Address: http://target serverIPaddress:service runtime port (default 8080)
Username: admin
Password: MaxKB@123...

The first time you log in, you need to change the password of the admin user. After changing the password, log in the system again to use MaxKB.

2.4 kbctl operation and maintenance tools

kbctl command

The MaxKB offline installer comes with a built-in command line maintenance tool, kbctl, which can be accessed by executing kbctl help.

Usage.
kbctl [COMMAND] [ARGS...]
kbctl --help

Commands: kbctl [COMMAND] [ARGS...] kbctl --help
status Check the status of MaxKB service
start Starts the MaxKB service.
stop Stop the MaxKB service.
restart Restart the MaxKB service
reload Reload the MaxKB service
uninstall Uninstall the MaxKB service
upgrade Upgrade MaxKB service
version View MaxKB version information

3. MaxKB Quick Practice

3.1 Logging in to the MaxKB System

Use your browser to open the service at http://target serverIPaddress:target port.
Default login information:
Username: admin
Default password: MaxKB@123...

3.2 Adding models

3.2.1 Adding the Chifan Big Model

Click [Add Model], select the supplier [Chifan Big Model], and directly enter the next step to fill in the Chifan Big Model form. Or the left side of the supplier first select [Chifan Big Model], then click [Add Model], then directly into the Chifan Big Model form.

Model Name: The name of the customized model in MaxKB.
Model Type: Large Language Model.
Base Model: The LLM model of the supplier, support customized input official model, the options listed in some of the commonly used large language models. Note that the name of the customized base model needs to be consistent with the model name of the vendor's platform in order to pass the validation.
API Key and Secret Key: API Key and Secret Key of the application in Chifan's big model (you need to create the application first)

3.2.2 Adding an Ollama Model

API Domain Name and API Key: Connectivity information for the provider (Ollama service address, e.g.:http://42.92.198.53:11434 ). If you do not have an API Key, you can enter any character.
After clicking [Add], if the verification passes, the model will be added successfully, and you can select the model in the AI model list of the application.

3.2.3 Adding an Azure OpenAI Model

Base Model: The LLM model for the vendor, supports customized inputs and drop-down options. Currently, large language models such as gpt-3.5-turbo-0613, gpt-3.5-turbo-0301, gpt-3.5-turbo-16k-0613, etc. can be selected from the drop-down list.

API Domain Name, API Key, Deployment Name: are provided in the model deployment of Azure OpenAI and need to be filled out consistently, refer to the figure below:

3.2.4 Adding an OpenAI Large Model

Base Model: LLM model for the vendor, support customized inputs, drop-down options are some big language models commonly used by OpenAI such as: gpt-3.5-turbo-0613, gpt-3.5-turbo, gpt-4, etc.
API domain name (foreign):/v1 API Domain Name: the reverse proxy address of the domestic server / v1. API Domain Name: Domestic server reverse proxy address / v1. API Key: Key to access OpenAI.

3.2.5 Adding the Cyberdyne Starfire Big Model

Base model: support custom input, corresponding to the domain in the interface document, the drop-down option is Xunfei Starfire commonly used in some of the big language model.
API domain name: Each base model corresponds to a different API domain name, please enter the corresponding API domain name according to the selected base model.CyberStarfire Official Documentation。

APP ID, API Key, Secret Key need to create the application now in Xunfei Starfire Open Platform and the remaining tokens>0 of the application to be added successfully.

3.2.6 Adding the Smart Spectrum AI Big Model

API Key: needs to be created now on the Smart Spectrum AI Open Platform.

3.2.7 Adding the Tongyi Thousand Questions Big Model

API Key: You need to get it in AliCloud ->DashScope Reed Product Modeling Service ->API Key Management.

3.2.8 Adding the Kimi Large Model

API domain name:/v1
API Key: Get the key in the API Key Manager in Kimi Account Center.

3.2.9 Adding a DeepSeek Large Model

API Key: InDeepSeek Official Get the API Key.

3.2.10 Adding a Gemini Large Model

Note: To use the Gemini API, you need to make sure that the server on which the program resides is located in theRegions supported by the Gemini API , otherwise the API cannot be called and there is no access to Google AI Studio.

3.3 Creating a generic knowledge base

Open the [Knowledge Base] page and click [Create Knowledge Base] to enter the Create Knowledge Base page. Input the knowledge base name, knowledge base description, and select the general-purpose knowledge base type. Then upload the offline documents by dragging and dropping or selecting the file upload method.

Upload a document
- Upload document requirements:
  - Support text file formats: Markdown, TXT, PDF, DOCX, HTML; support QA question and answer pairs format: Excel, CSV;
  - Upload up to 50 files at a time;
  - Each file does not exceed 100 MB; supports selecting a folder and uploading the files under the folder that meet the requirements.
Documentation specification recommendations:
1. Standardized segmentation marking: The segmentation marking of offline documents should be somewhat standardized, otherwise the split paragraphs will be irregular.
2. Paragraphs should be complete: it is desirable to describe a complete function point or problem in a single subparagraph.

Setting up segmentation rules

Intelligent Segmentation

MarkDown type of file segmentation rules are: according to the title of the drill-down level by level segmentation (supports up to 6 levels of title), the maximum number of characters in each paragraph is 4096 characters;
                        When the number of characters in the last level of the text paragraph exceeds the set segment length, it will find the carriage return within the segment length for interception.
HTML, DOCX type of segmentation rules are: to identify the title format converted to markdown title style, drilling down level by level for segmentation (supports up to 6 levels of title) the maximum number of characters in each paragraph is 4096 characters;
TXT and PDF type document segmentation rules are: according to the title # for segmentation, if there is no # title in accordance with the number of characters 4096 characters for segmentation, will look for segmentation within the length of the carriage return for interception.

Advanced Segmentation

Users can customize the setting of segment identifier, segment length and automatic cleaning according to the document specification.
Segment identifier support: #, ###, ####, ####, #####, ######, -, blank line, carriage return, space, semicolon, comma, period, if there is no option can also be customized to enter the segment identifier.
Segment length: Support minimum 50 characters and maximum 4096 characters.
Auto-cleaning: when turned on, the system will automatically remove repetitive and redundant symbols such as spaces, blank lines, tabs, etc.

Add Segment Headings as Associated Issues on Import

Checking the box will set the title of all segments as the associated issue for the segment.

previews

After the segmentation rules are set up, you need to click [Generate Preview] to view the segmentation effect of the latest rules.

Users can edit and delete unreasonable segments in the segment preview.

After clicking [Create and Import], the system will perform automatic segmentation -> storage -> vectorization processing operations on the document. You can see the number of imported documents, the number of segments, and the number of characters in the creation completion page.

4. Creating applications

Click [Create Application], enter the application name, select [Easy Configuration Application], and click [Create].

After the application creation is completed, you will enter the setting page of the easy configuration application, with the application information on the left side and the debugging preview screen on the right side.
Application name: the title and name of the dialog box when the user asks a question.
Application Description: A description of the application scenario and usage.
AI Model: A large language model added in [System Settings] - [Model Management].
Prompt words: The system has intelligent knowledge base prompt words by default, users can customize the content of the prompt words by adjusting the content, you can guide the direction of the big model chat. Multiple rounds of dialog: When it is on, when the user asks a question, it carries the last 3 questions of the user in the current session; if it is not on, it only submits the current question to the big model.
Associated Knowledge Base: Segments are retrieved in the associated knowledge base after a user asks a question.
Opener: the system popup greeting when user opens a dialog. Support Markdown format; the content after [-] is a quick question, one line at a time.
Problem Optimization: The problem posed by the user is first subjected to an LLM optimization process, and the optimized problem is vectorized for retrieval in the knowledge base;
Turning it on improves the accuracy of retrieving the knowledge base, but increases the length of time it takes to answer the question due to asking the larger model more than once.

After the application information is set up, you can preview the question in the debug preview on the right side, and the content of the question in the debug preview is not counted in the dialog log.

Detailed description of parameter settings
1. Retrieval mode:
  - vector search: Use the vector model to compute the text segments that are most similar to the user's question by vector distance;
  - full text search: Search by keyword and return the text segments that contain the most keywords;
  - hybrid search: Perform both full-text and vector searches, and then perform reordering to select the best result that matches the user's problem from the results of the two types of queries.
2. Similarity:
  - Higher similarity represents a stronger correlation between the question and the segmentation.
3. Cite the number of segments:
  - The questioning carries N segments by similarity to generate cue words to interrogate the LLM model.
4. Quotes the maximum number of characters:
  - Set the maximum number of characters for quoting segmented content, and truncate it if it exceeds the maximum number of characters.
5. When there is no reference to the knowledge base, there are 2 types of processing that can be set:
  - Keep asking questions.: You can customize the prompts, and there needs to be a placeholder for {question} the user's question before it will be sent to the model.
  - Specify the content of the response: You can specify the content of the reply when no knowledge base segments are hit.

Once you have completed the above 4 steps, you are ready for the Q&A dialog.

5. Demonstration of results

On the application list page, click [Demo] or click the application panel, and then click Demo on the overview page or copy the public access link to your browser to go to the Q&A page to ask a question.

6. Dialogue log

The Conversation Log clock records details of all questions and answers in a user session, including user feedback on AI responses. Maintainers can correct answers by reviewing the dialog logs and user feedback.

More quality content please pay attention to the public number: Ting, artificial intelligence; will provide some related resources and quality articles, free access to read.

More quality content please follow CSDN: Ting, Artificial Intelligence; will provide some related resources and quality articles, free access to read.