Overview
Deep Agents is a LangChain library designed to build “deep” agents capable of long-running, multi-step reasoning. It combines planning tools, subagents, filesystem utilities and persistent memory to decompose complex tasks into manageable subtasks and coordinate their execution reliably.
Key features
- Built-in planning and todo-list tools to break down problems and track progress.
- Subagent and middleware support for responsibility isolation and composability.
- Filesystem tools and memory primitives to manage long contexts and external data.
Use cases
- Deep research assistants that gather, synthesize and produce structured reports.
- Automated code workflows that decompose large engineering tasks into tool-driven steps.
- Multi-stage business automation requiring cross-step state and memory.
Technical highlights
- Modular middleware architecture (PlanningMiddleware, FilesystemMiddleware, SubAgentMiddleware) for extensibility.
- Native Python support and packaging (pip/poetry) and integration with LangGraph for model/tool interoperability.
- MIT-licensed for broad reuse in both community and commercial projects.