4. Perform DeepSearch Loop
The key to the Deep Research workflow lies in its powerful data collection capability. In this implementation, that capability is realized through a recursive workflow loop that starts with the original query and is extended with AI-generated sub-queries.
The number of sub-queries to generate is determined by the specified depth and breadth parameters.
Each sub-query produces a “learning,” which accumulates in each iteration of the loop.When the loop reaches the depth limit, it ends, and all the learnings are gathered to generate the final research report.
Deep Research 流程的关键在于其强大的数据收集能力。在本实现中，该能力通过递归式的工作流程循环来体现，该循环从原始查询开始，并由 AI 生成的子查询不断扩展。
生成多少个子查询由设定的深度和广度参数决定。 每个子查询都会产生“学习点（Learnings）”，并在每轮循环中累积。当循环达到深度限制时结束，此时所有的学习点将被收集，用于生成最终的研究报告。