🎯 Pain Points and Solutions#
Encountering a busy DeepSeek server? 😫
Insufficient local hardware computing power? 🤔
Dissatisfied with the distilled model's performance? 😑
Why not try the "Local Cherry Studio Client + DeepSeek API" method, because:
- It avoids server congestion
- Solves local hardware computing power issues
- Uses the full version of DeepSeek - R1
- Highly customizable and extensible
- Data stored locally, privacy assured
📊 Comparison of Solutions#
| Solution | Advantages | Disadvantages | Applicable Scenarios |
|---|---|---|---|
| Official DeepSeek Web Version | Simple operation, ready to use | Server congestion, limited features | Light users |
| Other platform web versions (e.g., Tencent Yuanbao) | Rich features, no local deployment required | Data upload, privacy risks | Regular users |
| Local Deployment | Data security, controllable performance | High hardware requirements, complex configuration | Professional developers |
| Cherry Studio + API | Localized data, high performance, flexible customization | Requires simple configuration | Deep users & developers |
⚙️ Configuration Method: Done in Three Steps!#
Divided into Cherry Studio Client Download and API Access two parts, you can get started in three minutes!
The client provides a local chat window, and the API provides access to DeepSeek services.
Cherry Studio Download#
Official website: https://cherry-ai.com/
Download and install the client.
Note 1: Official usage documentation: https://docs.cherry-ai.com/
Note 2: This project is open source on GitHub: https://github.com/CherryHQ/cherry-studio, currently has over 19k stars, free for personal use.
Silicon Flow to Access DeepSeek API#
First, it is important to clarify that calling the API incurs charges; however, the cost is very low. I have used it for a week and consumed about 10,000 tokens, costing around 0.15.
Why not use the official DeepSeek API? On one hand, the official server has limited capacity; on the other hand, even though the official API is already very cheap, other big companies offer lower prices and faster speeds, making them a better choice.
Moreover, the most important thing is that now Silicon Flow has a new user reward program. Click the Read the original text link to register and use the invitation code 29KuvDgU to receive 14 yuan in account funds, equivalent to 20 million tokens, calculated by usage without a deadline. This means you can use the API for free for a long time (based on my usage this week, that's 2000 weeks).
Registration link (click Read the original text to jump): https://cloud.siliconflow.cn/i/29KuvDgU
Or scan the QR code to open the registration interface:
Note 3: The invitation code will be automatically filled in after clicking this link.
After logging in, find "Account Management - API Keys" in the left menu, select "Create New API Key," and you can fill in the description as "cherry studio." After completing, click to copy the API.
Client Configuration#
Back to the Cherry Studio client, click the settings in the lower left corner, select "Model Service - Silicon Flow," paste the copied API in the "API Key" field, and enter the following "API Address": https://api.siliconflow.cn
Finally, click check, select any model, and if it shows "Connection Successful," you can start enjoying the basic features!
🛠️ Additional Features#
Internet Connection#
The above-mentioned Silicon Flow models are not connected to the internet by default. To enable internet functionality, please find Tavily on the settings page and visit the official website for setup.
Intelligent Agents#
Cherry Studio itself is equipped with various intelligent agents to choose from. Additionally, it supports customization and allows separate prompt words and models for different personalized intelligent agents (assistants).
Knowledge Base#
Cherry Studio supports customizing a personal knowledge base. First, you need to select "Manage" in the model settings interface, then find "Embed" in the top menu bar, select the first "BAAI/bge-m3," and then click on the knowledge base in the left menu to configure your personal knowledge base. You can upload local files, or links to websites or individual web pages.
Usage Example#
For example, I set up a knowledge base for "High School Mathematics Textbooks" and uploaded five textbooks from the People's Education Press A version for high school mathematics.
Then, in the chat interface, I added the assistant "High School Mathematics Textbook Assistant," with the prompt:
"When I ask about mathematics knowledge, please answer me based on the content in the high school mathematics textbooks."
Choosing the knowledge base I just created, when asking this assistant, I will get the following effect:
Model Temperature#
DeepSeek's official model temperature settings:
| Scenario | Temperature |
|---|---|
| Code generation/Math problem solving | 0.0 |
| Data extraction/analysis | 1.0 |
| General conversation | 1.3 |
| Translation | 1.3 |
| Creative writing/Poetry | 1.5 |
Drawing#
I used it once, and it turned out quite nice:
Mini Programs#
You can access the official websites of major mainstream models within the client:
🔗 Links#
- Cherry Studio: https://cherry-ai.com/
- Silicon Flow: https://cloud.siliconflow.cn/i/29KuvDgU