Deep Agents is LangChain’s batteries-included harness for building capable AI agents. Where a basic agent is just a model calling tools in a loop, Deep Agents ships the pieces production agents actually need out of the box: planning, sub-agents, a virtual file system, persistent memory and detailed system prompts.
Deep Agents has earned more than 30,400 stars on GitHub since its release and is updated continuously, with releases in October 2026. It is built on LangGraph, LangChain’s agent runtime, and packages the patterns that make agents reliable on long, multi-step tasks. An agent can write out a plan as a to-do list, spawn focused sub-agents for subtasks so the main context stays clean, read and write files in a sandboxed workspace, and remember information across sessions. The library is available for both Python and TypeScript, is model-agnostic, and integrates with LangSmith for tracing and evaluation when you want observability. It is the same architecture lineage as the deep research and coding agents that popularized the pattern.
Deep Agents is free and open source under the MIT license. You pay only for the model API usage of whichever provider you plug in; optional LangSmith observability has its own plans.
Deep Agents is for developers building serious agent applications, research assistants, coding agents and workflow automation, who want proven agent infrastructure instead of assembling planning, memory and delegation from scratch.