deep-agents-orchestration

1<overview>
2Deep Agents include three orchestration capabilities:
3 
41. **SubAgentMiddleware**: Delegate work via `task` tool to specialized agents
52. **TodoListMiddleware**: Plan and track tasks via `write_todos` tool
63. **HumanInTheLoopMiddleware**: Require approval before sensitive operations
7 
8All three are automatically included in `create_deep_agent()`.
9</overview>
10 
11---
12 
13## Subagents (Task Delegation)
14 
15<when-to-use-subagents>
16 
17| Use Subagents When | Use Main Agent When |
18|-------------------|-------------------|
19| Task needs specialized tools | General-purpose tools sufficient |
20| Want to isolate complex work | Single-step operation |
21| Need clean context for main agent | Context bloat acceptable |
22 
23</when-to-use-subagents>
24 
25<how-subagents-work>
26Main agent has `task` tool -> creates fresh subagent -> subagent executes autonomously -> returns final report.
27 
28**Default subagent**: "general-purpose" - automatically available with same tools/config as main agent.
29</how-subagents-work>
30 
31<ex-custom-subagents>
32<python>
33Create a custom "researcher" subagent with specialized tools for academic paper search.
34 
35```python
36from deepagents import create_deep_agent
37from langchain.tools import tool
38 
39@tool
40def search_papers(query: str) -> str:
41    """Search academic papers."""
42    return f"Found 10 papers about {query}"
43 
44agent = create_deep_agent(
45    subagents=[
46        {
47            "name": "researcher",
48            "description": "Conduct web research and compile findings",
49…