Deep Researcher
A multi-model collaborative workflow that generates structured research reports in minutes
⬇️ Try it out before reading
📖 Overview
Deep Researcher is a workflow built on the Dify platform that reproduces the core functionality of Deep Research. By integrating multi-source retrieval (local knowledge base + web search) with multi-model collaboration, it can generate structured research reports of tens of thousands of words in 5 minutes. The system adopts a modular design and supports flexible replacement of underlying models and data sources.
✨ Core Features
Intelligent Topic Analysis
Using Gemini 2.0 Flash model for multi-level topic decomposition, supporting 4-dimensional in-depth analysis
Hybrid Retrieval Engine
Local Knowledge Base + Wikipedia/Google/Bing API multi-channel retrieval
Dynamic Rhythm Control
Using a 2>1 model cascade architecture, implementing processing rhythm optimization through conditional branches and dialogue round markers
Efficient Report Generation
Integrating models like deepseek-r1-distill for paragraph-level content generation, supporting Markdown structured output
🛠️ Technical Architecture
The workflow adopts the following architecture:
⚠️ Notes
Performance Optimization Suggestions
In principle, the workflow supports any model. When the request pressure of using local models is high, it may trigger the Timepouterror of the LLM node. You can consider switching to an online API service or modifying the timeout time in the Dify configuration file.
If you are a Google free API user, you can insert a local model node to limit the RPM (Google's default rate limit is 15 RPM, and errors will be reported if there are too many requests in a short time)
