Building a multi-agent researcher with llms.txt

A common pattern with Deep Research agents is the ability to ask clarifying questions in a planing phase followed by iterative research in a research phase. Here, we show that a multi-agent architecture (via Swarm) is well-suited for this talk, with a planner agent asking clarifying questions and a research agent performing iterative research. We build a simple multi-agent system that can answer questions about LangGraph documentation using our new llms.txt file (https://langchain-ai.github.io/langgraph/llms.txt) along with a simple URL loader tool. Code: https://github.com/langchain-ai/langgraph-swarm-py/tree/main/examples/research Video notes: https://mirror-feeling-d80.notion.site/Multi-Agent-Research-1c4808527b17808b936ff1ad24125ee6?pvs=4 Chapters: 0:00 - Introduction to Deep Research Agents 0:27 - Demo of Planner-Researcher System 1:00 - Multi-agent System for Research Tasks 1:34 - Introduction to Swarm Architecture 2:06 - Exploring the Code Implementation 2:30 - Defining Agent Handoff Tools 3:00 - The Planner Agent's Prompt and Role 3:34 - The Researcher Agent's Responsibilities 4:00 - Setting Up the Swarm Framework 4:17 - Testing the System in a Notebook 4:40 - Tracing Through the Execution Process 5:22 - Verifying the Generated Implementation 5:40 - Key Benefits of Multi-agent Research 6:00 - Conclusion and Final Thoughts