deepsearch-service

Features

An LLM-based iterative deep search service capable of:

1. Decomposing complex questions into multiple sub-queries
2. Iteratively searching to collect information from multiple sources
3. Evaluating information sufficiency and dynamically adjusting the search strategy
4. Comprehensively generating a structured analysis report

Use Cases

• Complex questions requiring multi-perspective analysis
• Needing to synthesize information from multiple sources
• Requiring the generation of a complete research report

How to Call

from services.deepsearch_service.client import DeepSearchClient

client = DeepSearchClient()

Health Check

status = client.health_check()

Deep Search (default parameters)

result = client.search("What are the best practices for Python asynchronous programming?")

Custom Parameters

result = client.search( query="How do you design a highly available microservice architecture?", max_iterations=3, # Maximum number of iterations (1-5) queries_per_iteration=3, # Number of queries per round (1-5) depth_level="deep" # Search depth: quick/normal/deep )

Get the Report and Sources

print(result["report"]) for source in result["sources"]: print(f"- {source['title']}: {source['url']}")

## 返回格式
```json
{
  "query": "Python异步编程的最佳实践有哪些？",
  "report": "# Python异步编程最佳实践\n\n## 1. 核心概念...",
  "sources": [
    {
      "title": "Python官方asyncio文档",
      "url": "https://docs.python.org/...",
      "relevance": 0.95,
      "snippet": "asyncio是Python标准库中的异步I/O框架..."
    }
  ],
  "iterations": [
    {
      "iteration": 1,
      "queries": ["Python asyncio 教程", "async await 用法"],
      "results_count": 6,
      "key_findings": ["asyncio是标准库组件"]
    }
  ],
  "total_iterations": 2,
  "total_sources": 8,
  "elapsed_seconds": 45.32,
  "search_timestamp": "2025-12-28T18:30:00"
}

Differences from websearch_service

Feature	WebSearch	DeepSearch
Search Mode	Single-round search	Iterative multi-round search
Query Strategy	Directly use the input	LLM decomposition/optimization
Output Format	Structured list	Comprehensive report + sources
Response Time	10-30s	30-120s