MindSearch is an LLM-powered multi-agent framework designed to mimic human thought processes for web search, aiming to provide deeper and more comprehensive results than traditional search engines. It targets researchers and power users seeking advanced AI-driven information retrieval, offering a sophisticated alternative to platforms like Perplexity.ai Pro and SearchGPT.

How It Works

MindSearch employs a multi-agent architecture, where individual agents concurrently query various search engines and process information. This approach, inspired by mimicking human minds, allows for parallel exploration and synthesis of data, leading to more nuanced and potentially more accurate search outcomes. The framework supports asynchronous operations for enhanced concurrency and allows customization of search engines and LLM models.

Quick Start & Requirements

• Installation: Clone the repository, install dependencies via pip install -r requirements.txt.
• Environment Variables: Rename .env.example to .env and configure API keys (e.g., WEB_SEARCH_API_KEY, TENCENT_SEARCH_SECRET_ID, TENCENT_SEARCH_SECRET_KEY).
• Backend Setup: Run python -m mindsearch.app --lang en --model_format internlm_server --search_engine DuckDuckGoSearch --asy. Supports DuckDuckGo, Bing, Brave, Google (Serper), and Tencent search APIs.
• Frontend Setup: Navigate to frontend/React, configure vite.config.ts with the backend URL, run npm install and npm start. Gradio and Streamlit frontends are also available.
• Prerequisites: Python, Node.js/npm, specific API keys for search engines (except DuckDuckGo and Tencent).
• Docs: Paper, Demo

Highlighted Details

• Mimics human thought processes for deep AI search.
• Supports multiple LLM formats (InternLM2.5-7b-chat, GPT4) and search engines.
• Offers concurrent multi-query search with asynchronous agent deployment.
• Provides React, Gradio, and Streamlit frontend options.

Maintenance & Community

The project is actively maintained, with recent updates refactoring the agent module for better performance and improving the UI. It is part of the InternLM ecosystem, which includes related projects like Lagent, AgentFLAN, and T-Eval.

Licensing & Compatibility

Released under the Apache 2.0 license, permitting commercial use and integration with closed-source projects.

Limitations & Caveats

The framework requires careful setup of environment variables and API keys. While supporting multiple models, optimal performance may depend on specific LLM configurations and hardware.