Deep Search Lighting

What is Deep Search Lighting?

Deep Search Lighting is a lightweight, pure web search solution for large language models that supports multi-engine aggregated search, deep reflection, and result evaluation. It is designed for developers needing a framework-free, easy-to-integrate search capability across models of any size.

How to use Deep Search Lighting?

Install dependencies via pip install -r requirements.txt, rename .env.examples to .env, and fill in your model API information (OpenAI‑style APIs). Run the test case with python test_demo.py, launch the Streamlit demo with streamlit run streamlit_app.py, or start the built‑in MCP server with python mcp_server.py and connect the client via python langgraph_mcp_client.py.

Key features of Deep Search Lighting

• Multi-engine aggregated search (Baidu, DuckDuckGo, Bocha, Tavily)
• Reflection strategies for self-evaluation by the LLM
• Customizable pipelines for any LLM model
• OpenAI-style API compatibility out of the box
• Pure model source code for easy integration
• Built-in MCP server support

Use cases of Deep Search Lighting

• Enhancing an LLM with real‑time web search without a heavy framework
• Rapid prototyping of search‑augmented applications using free APIs
• Offering adjustable depth for balancing speed and result quality
• Running web search tasks on smaller LLMs that struggle with tool‑calling patterns

FAQ from Deep Search Lighting

What makes Deep Search Lighting different from traditional web search solutions for LLMs?

It is framework‑free, works with free APIs, supports models of any size, and includes a reflection mechanism for self‑evaluation, all while keeping implementation simple without web parsing or text chunking.

What runtime dependencies are required?

Python 3.11 and packages listed in requirements.txt; optionally install requirements_langchain.txt for LangChain support.

Which search engines are supported and what are their requirements?

Baidu (free, no key), DuckDuckGo (free but needs VPN), Bocha (requires API key), and Tavily (requires registration key). Baidu is enabled by default.

How do I configure and run the MCP server?

Set up your .env file with model API details, then run python mcp_server.py and python langgraph_mcp_client.py.

Is there any cost or performance limit mentioned?

No explicit cost or performance limit is stated, but the project emphasizes using free APIs for most engines and supporting adjustable depth to trade speed versus result quality.