How adamplatin123/open-deep-research-workflow-on-dify works
Project type and purpose
• This is an AI‐driven research‐report pipeline implemented as a Dify workflow.
• Purpose: ingest a user query, break it into hierarchical topics, retrieve relevant content from local and online sources, orchestrate multiple LLMs in a cascaded/branched fashion, and emit a structured Markdown report.
File structure
• Top level contains:
• “Deep Researcher On Dify .yml” – the Dify workflow definition, listing nodes, data flows and conditional branches.
• “README.md” – describes core features, architecture and external dependencies.
• There is no monolithic codebase: the entire logic lives as modules (nodes) in the .yml file, orchestrated by the Dify engine.
README for insights
• Core components: Theme parsing, question generation, conditional user feedback loops, secondary topic extraction, hybrid retrieval engine, multi‐model generation, final report assembly.
• External systems: Local knowledge base, Wikipedia/Google/Bing APIs.
• Models: Gemini 2.0 Flash, deepseek-r1-distill (and placeholders for custom LLMs).