🌿 kuki research

kuki search is a fast and concurent reserch engine for deep web analysis and synthesis. it uses duckduckgo for broad search and crawl4ai for markdown content. 🌿

🌿 features

• concurent crawling: fast fetching and parsing of multiple results using crawl4ai
• smart synthesis: auto-generates subqueries to make sure everythng is covered
• universal llm support: built on litellm so it works with openai, anthropic, gemini, kimi and more
• clean resuts: saves summarized insights
• perfomance optimized: kinda fast

🌿 project structure

├── config.py           # llm config and things 
├── main.py             # entry point and crawler warmup 
├── src/
│   ├── engine.py       # core logic and research pipeline 
│   ├── ddgs.py         # duckduckgo search stuff 
│   ├── queries.py      # llm sub-query logic 
│   └── prompts/        # prompt stings for research 
│       ├── subqueries.md
│       └── research.md
└── results/        # logs and results 

🌿 setup

1. instalation: get the dependencies ready:

python -m pip install requirements.txt

2. configuation: change config.py or set env vars for ur llm.

LLM_MODEL = "gpt-4"
LLM_API_KEY = "your-api-key"

usage

just run it with ur query as an argument:

python main.py "How do the current battery technologies compare for EV range?"

🌿 the engine will:

1. initialize and pre-warm the cralwer
2. generate 5 subqueries
3. search and crawl top 10 results at once
4. synthesize final answer with citations
5. save it all in a folder named after ur query

🌿 architecture

its a fairly simple project, and will implement more features in the future!