Usage Guide#

This document introduces the recommended workflow for DeepRead, helping you get started quickly from article downloading to AI Q&A.

Quick Start#

1. Open DeepRead and go to the Download Center. DeepRead starts up not fast, usually taking 10 to 30 seconds. Please be patient.
2. Enter the WordPress category URL.
3. Set a Knowledge Base Name (recommend using the site name or topic).
4. Set the number of concurrent downloads, then click download.
5. Once the download is complete, go to AI Chat, select the newly created knowledge base, and start asking questions.

Typical Workflow#

(1) Batch Downloading Articles#

• Use a category URL like https://example.com/category/your-category/ for easier batch collection.
• For first-time use, it's recommended to perform a small batch test to ensure fields and content meet expectations.
• If the target site is slow, decreasing concurrency can lead to a higher success rate.

(2) Managing Local Knowledge Bases#

• It's recommended to use unique knowledge base aliases for each topic to facilitate future retrieval.
• In Article Management, you can search by title, preview Markdown content, and locate image files.
• Periodically clean up unused knowledge bases to keep search results focused.

(3) AI Q&A and Deep Reading#

• Select a knowledge base for Q&A.
• Questions should ideally have a clear scope, e.g., "Based on articles in the XX category, summarize 3 core insights."
• For complex questions, ask step-by-step: summarize first, then follow up with specific details and evidence sources.

(4) Parameters and Environment Configuration#

Configure the following in System Settings:

• Download Folder: Downloaded articles and images will be saved in this folder. Usually, you can keep the default value.
• Skip Downloaded: If an article is detected to already exist locally, the download will be skipped. This can reduce the target server pressure and save network traffic.
• Use Mainland China AI Model Download Mirror: If you are located in Mainland China, the app will download the AI large model locally. It is recommended to enable this to increase the download speed. If you are outside China, please do not enable it.
• Concurrent Downloads: In theory, a larger value means higher download efficiency. However, since target websites often have rate limits, a concurrency that is too high might lead to download failures. It is recommended to keep it between 1 and 3 (1 is the safest).
• Minimum and Maximum Request Delay: The interval time between each article download. To simulate random download behavior as much as possible, the interval time will be randomly selected between the minimum and maximum values each time.
• Articles per Page: The number of articles downloaded per request. You need to strike a balance between efficiency and target website pressure. A smaller value makes downloading easier, but lowers efficiency. It is recommended to keep the default value.
• Skip Threshold: During the download process, when N consecutive articles are detected to have already been downloaded locally, the downloading will stop to help reduce target server pressure.
• LLM API Key: The key for the large language model server (currently only DeepSeek is supported).
• Document Retrieval Count: When conducting AI Q&A, the AI will search for the top N local articles most relevant to the question as a reference. A larger value helps the AI find relevant content more easily, but increases its thinking time.
• Proxy Server: When you are in an extreme network environment, such as being unable to access the AI model or the target website, you can set this value to proxy the traffic.

Recommendations:

• After initial configuration, perform 1-2 short Q&A sessions to confirm connectivity and response speed.
• If responses are slow, first check the model service status and your network connection.

(5) License Activation#

• Enter your license information on the License Management page to activate.
• You can view the list of bound devices.
• Before switching devices, it's recommended to unbind on the old device.

Tips for Efficient Use#

• Process "downloads" and "Q&A" in separate batches to avoid loading too much data at once.
• For key conclusions, it's recommended to cross-check with the original article.