Build a Research Agent with Deep Agents

Description

Deepagents is a simple, open source agent harness built by LangChain. It uses some common principle seen in popular agents such as Claude Code and Manus, including planning (prior to task execution), computer access (giving the able access to a shell and a filesystem), and sub-agent delegation (isolated task execution). We're introducing a new repo with a collection of quickstarts that demonstrate different agents can be easily configured on top of the deepagents harness. Quickstarts repo: https://github.com/langchain-ai/deepagents-quickstarts Deepagents repo: https://github.com/langchain-ai/deepagents Deepagents docs: https://bit.ly/480icl1 Deepagents UI: https://github.com/langchain-ai/deep-agents-ui Chapters -- 0:00 Introduction to DeepAgents 1:00 Agent Trajectory Overview 2:00 Built-in Tools in DeepAgents 3:00 Quick Start Setup Options 4:00 Task-Specific Tools: Search & Think 5:00 Task-Specific Instructions 6:00 Preventing Agent Spin Out 7:00 Custom Prompts & Instructions 8:00 Custom Subagents for Context Isolation 9:00 Workflow Instructions & Delegation Strategy 10:00 Initializing DeepAgent 11:00 Middleware Overview 12:00 Running the Agent in Notebook 13:00 File System Backends 14:00 Subagent Research Output 15:00 LangSmith Tracing 16:00 Deploying with LangGraph Server 17:00 Summary & Key Takeaways