A Web Research chain turns one user question into a research report. It generates multiple searches, processes each search in parallel, summarizes retrieved pages, and combines the summaries into a final answer. section 4.7 section 4.7.4
Build the workflow from focused chains:
- Use an Assistant Instructions chain to select a research assistant and define its system prompt. section 4.7
- Use a Web Searches chain to generate several queries that provide different perspectives or divide a complex question. section 4.7
- Run a Search and Summarization chain for every query; each instance searches, retrieves URLs, scrapes pages, and summarizes content. section 4.7
- Join the summaries with a RunnableLambda, then pass them and the original question to a Research Report chain. section 4.7.4
- Use LCEL's map() operator to run the search-and-summarization sub-chain concurrently for each generated search. section 4.7.4
Travel research about Astorga
Invoking the chain with “What can I see and do in the Spanish town of Astorga?” produces a Markdown report covering attractions, activities, history, and trip-planning information. section 4.7.4
Figure 4.7 Chain architectural diagram showing how parallelization is applied in the research summarization engine, highlighting the execution of separate chain instances in parallel for efficiency
The architecture diagram makes the concurrency pattern visible: separate Search and Summarization chain instances execute in parallel before report synthesis. section 4.7
