Building a Local Private Knowledge Base Q&A System with MaxKB + Ollama

Building a Local Private Knowledge Base Q&A System with MaxKB + Ollama

MaxKB is an open-source RAG system focused on knowledge base construction that, when combined with locally deployed Ollama and the DeepSeek R1 7B model, creates a completely offline, highly secure intelligent Q&A platform.

Technical Components:

  • Knowledge Base Engine: MaxKB (includes vectorization, retrieval, and Q&A chain)
  • Model Service: Ollama (lightweight LLM deployment tool)
  • Large Language Model: DeepSeek R1 7B (high-performance Chinese-English bilingual model)
  • Deployment Method: Docker containerized deployment, supporting CPU/GPU inference

The powerful capabilities of MaxKB combined with Ollama enable you to quickly build a high-performance, privately deployed knowledge base Q&A system, ensuring enterprise data security while enabling intelligent document retrieval and question answering.

šŸ” Project Advantages

Why Choose a Locally Deployed Private Knowledge Base?

Feature Local Private Deployment Public Cloud Models
Data Security āœ… Completely offline, data never leaves local network āŒ Risk of sensitive information leakage
Control āœ… Complete control over model parameters/vector database āŒ Limited by platform policies
Response Speed āœ… Local inference, low latency āŒ Heavily affected by network conditions
Customization āœ… Deep customization of models and UI interface āŒ Highly standardized, difficult to customize
Usage Cost āœ… One-time deployment, long-term use āš ļø Token-based billing, high long-term costs
Compliance āœ… Meets data localization regulatory requirements āŒ Potential compliance risks

The combination of MaxKB + Ollama + DeepSeek provides enterprises with a complete localized RAG (Retrieval-Augmented Generation) solution, especially suitable for scenarios with strict data security and privacy requirements, such as:

  • Enterprise internal knowledge base retrieval
  • Intelligent Q&A for technical documentation
  • Self-service product manual queries
  • Customer service knowledge base assistance

šŸ“ Environment Preparation

System Requirements

  • Operating System: Linux (Ubuntu 20.04+), macOS, or Windows 11
  • Memory Requirements: Minimum 8GB, 16GB+ recommended
  • Storage Space: At least 10GB available space
  • GPU: Optional, NVIDIA graphics card with CUDA support can significantly improve performance

Dependencies

  • Docker & Docker Compose
  • Git (optional, for obtaining the latest source code)

🧰 Installation Steps

Step 1: Install MaxKB

Clone the project code from the GitHub repository:

git clone https://github.com/1Panel-dev/MaxKB.git
cd MaxKB

Start the MaxKB service using Docker Compose:

docker-compose up -d

Note: If you need custom configuration, you can edit the docker-compose.yml file

After installation, access the MaxKB management interface at: http://localhost:3000

Step 2: Install Ollama

Ollama is a cross-platform LLM management client (supporting MacOS, Windows, and Linux) that enables seamless deployment of large language models like DeepSeek, Llama, and Mistral. Ollama provides a one-click model deployment solution, ensuring all data is stored locally for complete security and privacy.

Install Ollama in your local environment:

MacOS / Linux:

curl -fsSL https://ollama.com/install.sh | sh

Windows:

After installation, the Ollama service will run in the background with the default port 11434

Step 3: Pull the DeepSeek R1 7B Model

Pull the DeepSeek R1 7B model from the Ollama library:

ollama pull deepseek:7b

The initial model download requires approximately 4-5GB of space, please ensure your network connection is stable

Verify that the model was successfully installed:

ollama run deepseek:7b

If everything is normal, you will see a prompt indicating that the model has loaded and is ready for use.

Step 4: Configure MaxKB to Use the Local Ollama Model

  1. Log in to the MaxKB admin panel
  2. Go to the "Model Settings" page
  3. Click "Add Model"
  4. Select Ollama:
    • Name: Ollama-DeepSeek
    • Type: Custom Model (OpenAI compatible)
    • Inference Address: http://localhost:11434 (if MaxKB and Ollama are on different servers, use the actual IP)
    • API KEY: ollama-local (any value can be entered)
    • Model: deepseek:7b
  5. Click "Test Connection" to confirm model connectivity, then save the configuration

šŸ“š Usage Guide

Creating a Knowledge Base

  1. In the MaxKB management interface, click APP
  2. Click Create APP
  3. Enter the application name, description, and select the type as SIMPLE
  4. Click Settings to configure the application
  5. Configure AI Model and select the DeepSeek model you just configured
  6. Click Parameter Settings in the Related Knowledge section to set retrieval parameters (default values are usually suitable for most scenarios)

Importing Documents

MaxKB supports multiple document import methods:

  • File Upload: Supports PDF, Markdown, Word, TXT, and other formats
  • Web Scraping: Enter a URL to automatically scrape web content
  • API Import: Automate document import via API

After importing, MaxKB automatically performs document chunking and vectorization to complete the knowledge base construction.

Starting Q&A

  1. Go to the "Q&A Testing" page
  2. Enter a question, and the system will retrieve relevant document fragments and generate an answer
  3. You can view the original source of the retrieved information to ensure the reliability of the answer

šŸ’” Frequently Asked Questions

Q1: What are the characteristics of the DeepSeek R1 7B model?

A: DeepSeek R1 7B is a high-performance Chinese-English bilingual large model that provides excellent understanding and generation capabilities at the 7B parameter scale. It's suitable for knowledge base Q&A scenarios and can run smoothly on ordinary hardware.

Q2: How can I improve Q&A quality?

A: You can optimize by adjusting the following parameters:

  • Increase the number of retrievals (typically 3-5 fragments work best)
  • Adjust document chunking size (choose based on document characteristics)
  • Optimize prompt templates to guide the model to answer more accurately

Q3: Can it run on servers without a GPU?

A: Absolutely. Ollama supports running DeepSeek R1 7B in CPU mode. Although the response speed will be slower than with a GPU, it's still sufficient for internal enterprise use.

Q4: How can I expand to more knowledge bases?

A: MaxKB supports creating multiple independent knowledge bases, each using different document sets, different retrieval parameters, and even different models to meet the requirements of multiple business scenarios.

Q5: How can I ensure deployment security?

A: The following measures are recommended:

  • Deploy on an internal network, limiting external access
  • Enable MaxKB's user authentication features
  • Regularly update Ollama and MaxKB to the latest versions
  • After the initial model download, consider running in a disconnected production environment

āœ… Conclusion

MaxKB + Ollama + DeepSeek R1 7B provides a complete, secure, and efficient knowledge base solution, allowing enterprises to enjoy advanced AI Q&A capabilities without worrying about data security issues or paying expensive API fees.

For enterprise users who need deeper customization and adaptation to more complex scenarios, MaxKB also provides enterprise-level support services to help you build an intelligent knowledge base system that better meets your business needs.

Back to blog