MindSearch with Tavily

English |简体中文

✨ MindSearch with Tavily: A Powerful Deep AI Searcher

• MindSearch is an open-source AI search engine framework that can use open-source LLMs (InternLM2.5 series models), specially optimized to provide excellent performance within the MindSearch framework.
• It can use search engine APIs such as DuckDuckGo, Bing, Brave, and Google.
• This project adds the TavilySearch engine, allowing the use of the Tavily API for searches.
• Tavily has a high reputation in academia and can provide more accurate search results.

⚽️ Build Your Own MindSearch with Tavily

🐞 Local Debugging of Tavily API

• First, register a new account at https://app.tavily.com/sign-in and set up a free Tavily API Key.
• Enter the mindsearch/agent folder, open the test_tavily.py file, insert your Tavily API Key, and run the following command to test if the Tavily API is working correctly:

python test_tavily.py

🐞 Get SiliconFlow API Key

• With SiliconFlow providing free InternLM2.5-7B-Chat services, you can use the API to get services, reducing the deployment MindSearch threshold.
• To use SiliconFlow's API Key, first open https://account.siliconflow.cn/login to register a SiliconFlow account (if you have already registered, log in directly).
• After completing registration, open https://cloud.siliconflow.cn/account/ak to prepare the API Key. First, create a new API key, then click the key to copy it for future use.
• After completing registration, open https://cloud.siliconflow.cn/account/ak to prepare the API Key. First, create a new API key, then click the key to copy it for future use.

Step 1: Install Dependencies

git clone https://githubfast.com/chenzubing/MindSearch.git
cd MindSearch
pip install -r requirements.txt

Step 2: Start MindSearch API

Start the FastAPI server:

python -m mindsearch.app --lang en --model_format internlm_silicon --search_engine TavilySearch

• --lang: The language of the model, en for English.
• --model_format: The format of the model.
• internlm_silicon for the InternLM2.5-7B-Chat silicon server.
• --search_engine: The search engine.
• TavilySearch for the TavilySearch search engine.

Step 3: Start MindSearch Frontend

Several frontend interfaces are provided:

• Gradio

python frontend/mindsearch_gradio.py

• Streamlit (recommended for more stable results)

streamlit run frontend/mindsearch_streamlit.py

📝 License

This project is released under the Apache 2.0 License.

Academic Citation

If this project has been helpful for your research, please consider citing it as follows:

@article{chen2024mindsearch,
  title={MindSearch: Mimicking Human Minds Elicits Deep AI Searcher},
  author={Chen, Zehui and Liu, Kuikun and Wang, Qiuchen and Liu, Jiangning and Zhang, Wenwei and Chen, Kai and Zhao, Feng},
  journal={arXiv preprint arXiv:2407.20183},
  year={2024}
}