deep-agents-core

1<overview>
2Deep Agents are an opinionated agent framework built on LangChain/LangGraph with built-in middleware:
3 
4- **Task Planning**: TodoListMiddleware for breaking down complex tasks
5- **Context Management**: Filesystem tools with pluggable backends
6- **Task Delegation**: SubAgent middleware for spawning specialized agents
7- **Long-term Memory**: Persistent storage across threads via Store
8- **Human-in-the-loop**: Approval workflows for sensitive operations
9- **Skills**: On-demand loading of specialized capabilities
10 
11The agent harness provides these capabilities automatically - you configure, not implement.
12</overview>
13 
14<when-to-use>
15 
16| Use Deep Agents When | Use LangChain's create_agent When |
17|---------------------|-----------------------------------|
18| Multi-step tasks requiring planning | Simple, single-purpose tasks |
19| Large context requiring file management | Context fits in a single prompt |
20| Need for specialized subagents | Single agent is sufficient |
21| Persistent memory across sessions | Ephemeral, single-session work |
22 
23</when-to-use>
24 
25<middleware-selection>
26 
27| If you need to... | Middleware | Notes |
28|------------------|------------|-------|
29| Track complex tasks | TodoListMiddleware | Default enabled |
30| Manage file context | FilesystemMiddleware | Configure backend |
31| Delegate work | SubAgentMiddleware | Add custom subagents |
32| Add human approval | HumanInTheLoopMiddleware | Requires checkpointer |
33…