Deep Agents

Deep Agents

Deep Agents is the easiest way to start building agents powered by LLMs—with built-in capabilities for task planning, file systems for context management, subagent delegation, and long-term memory. It is an "agent harness" built on LangChain core building blocks and the LangGraph runtime.

When to use

Use Deep Agents when you need to:

• Build agents fast with sensible defaults and minimal configuration
• Handle complex, multi-step tasks that benefit from automatic planning
• Manage context with a built-in virtual filesystem for large inputs
• Delegate subtasks to specialized subagents
• Run code safely in sandboxed execution environments
• Use a terminal agent via Deep Agents Code

When NOT to use

• For simple tool-calling agents without planning or subagents, use LangChain agents instead—lighter weight
• For custom graph-based orchestration with explicit control flow, use LangGraph directly
• Deep Agents is the highest-level abstraction—it trades flexibility for convenience

Install

# Python
pip install deepagents

# JavaScript/TypeScript
npm install deepagents langchain @langchain/core

Quick reference

Create a deep agent

# pip install deepagents langchain-anthropic
from deepagents import create_deep_agent

def get_weather(city: str) -> str:
    """Get weather for a given city."""
    return f"It's always sunny in {city}!"

agent = create_deep_agent(
    model="anthropic:claude-sonnet-4-6",
    tools=[get_weather],
    system_prompt="You are a helpful assistant",
)

result = agent.invoke(
    {"messages": [{"role": "user", "content": "What is the weather in SF?"}]}
)

Use Deep Agents Code

# Install Deep Agents Code
pip install deepagents-code

# Run an interactive terminal agent
deepagents

Built-in capabilities

| Capability | Description |

|-----------|-------------|

| Planning | Automatic task decomposition for complex requests |

| File system | Virtual filesystem for reading, writing, and managing context |

| Subagents | Spawn child agents for parallel subtask execution |

| Context management | Automatic context compression for long conversations |

| Sandboxed execution | Run code in isolated environments (Modal, Runloop, Daytona) |

| Protocols | ACP, MCP, and A2A support for interoperability |

Key documentation

• Overview—What Deep Agents is and how it compares to LangChain and LangGraph
• Quickstart—Build your first deep agent
• Customization—Configure models, tools, and behavior
• Context engineering—Manage context for complex tasks
• Subagents—Delegate work to child agents
• Sandboxes—Run code in isolated environments
• Code—Deep Agents Code, the terminal agent interface
• Deploy—Deploy to production

API reference

For SDK class and method details, use the LangChain API Reference site:

• MCP server: https://reference.langchain.com/mcp

Related skills

• langchain—Core building blocks that Deep Agents is built on
• langgraph—Runtime that powers Deep Agents' durable execution
• langsmith—Trace, evaluate, and deploy your deep agents