Replicating Deep Research in Jan (www.jan.ai)

In a significant move for the AI research community, a new hybrid deep research system has been introduced in Jan, combining local and cloud-based models to compete with established platforms like OpenAI and Claude. This system leverages a systematic methodology that incorporates exhaustive web search and report generation, adapting techniques pioneered by top players in the field. The approach outlines a structured pipeline with clear phases—planning, searching, analyzing, and synthesizing—which enhances the efficiency of information gathering and report creation, while maintaining user data privacy. The initial testing of this hybrid model revealed promising results, although it has not yet reached the comprehensiveness of leading commercial deep research tools. While models like Jan-Nano and GPT-4o delivered solid report outputs, they lacked the depth and breadth of content produced by their competitors. Importantly, the integration of Model Context Protocol (MCP) allows for effective data interaction, potentially paving the way for future enhancements and broader accessibility in AI-assisted research. This development signifies a step forward in democratizing deep research capabilities, catering to both privacy-conscious users and those seeking scalable solutions for comprehensive data analysis.