deepagents-langchain

# deepagents-langchain USE FOR: - "production-ready agent with LangGraph" - "batteries-included coding/research agent" - "sub-agents with isolated context windows" - "LangChain agent framework" - "agent with planning + filesystem + shell" - "MCP tools in LangGraph agent" tags: [LangGraph, LangChain, agent, production, sub-agents, MCP, planning, filesystem, shell, open-source] kind: framework category: ai-agent-builder --- ## What Is DeepAgents? Production-ready, batteries-included LangGraph agent by LangChain. No manual setup — `create_deep_agent()` returns a fully functional agent. - Repo: https://github.com/langchain-ai/deepagents - Install: `pip install deepagents` - Framework: **LangGraph** (compiled graph, streaming, persistence, checkpointing) - LangGraph Studio compatible --- ## Quick Start ```python pip install deepagents ``` ```python from deepagents import create_deep_agent agent = create_deep_agent() result = agent.invoke({ "messages": [{"role": "user", "content": "Research the latest AI agent frameworks and summarize"}] }) print(result["messages"][-1].content) ``` --- ## Built-in Capabilities | Capability | Tools Included | |-----------|---------------| | **Planning** | `write_todos` — task decomposition + progress tracking | | **Filesystem** | read, write, edit, search files | | **Shell** | execute commands (with sandboxing) | | **Sub-agents** | delegate tasks with isolated context windows | | **Context** | auto-summarization, large output → file handling | --- ## Architecture (LangGraph) ```python # Returns a compiled LangGraph graph agent = create_deep_agent() # Supports all LangGraph features: # - Streaming for chunk in agent.stream({"messages": [("user", "task")]}): print(chunk) # - Persistence / checkpointing from langgraph.checkpoint.memory import MemorySaver agent = create_deep_agent(checkpointer=MemorySaver()) # - LangGraph Studio compatibility (visual debug) ``` --- ## Customization ```python from deepagents import create_deep_agent from langchain_anthropic import ChatAnthropic # Custom model agent = create_deep_agent( model=ChatAnthropic(model="claude-opus-4-6") ) # Add custom tools from langchain_core.tools import tool @tool def my_tool(query: str) -> str: """Custom tool description""" return do_something(query) agent = create_deep_agent(tools=[my_tool]) # Custom system prompt agent = create_deep_agent( system_prompt="You are an expert financial analyst..." ) ``` --- ## MCP Integration ```python from langchain_mcp_adapters import MCPToolkit # Connect any MCP server toolkit = MCPToolkit(server_command=["npx", "gitnexus", "mcp"]) mcp_tools = toolkit.get_tools() agent = create_deep_agent(tools=mcp_tools) ``` --- ## Sub-Agents Pattern ```python # Main agent delegates to sub-agents with isolated contexts # Sub-agents don't share main agent's conversation history # Useful for: parallel research, isolated code execution agent = create_deep_agent( enable_subagents=True, subagent_model=ChatAnthropic(model="claude-haiku-4-5-20251001") # cheaper for subtasks ) ``` --- ## Use Cases - **Research pipeline**: fetch + summarize + synthesize across sources - **Code automation**: read codebase → plan changes → edit files → run tests - **Data processing**: ingest files → transform → write outputs - **Multi-step workflows**: plan → delegate subtasks → aggregate results --- ## vs. Other Agent Frameworks | Feature | DeepAgents | OpenAlice | AutoHedge | |---------|-----------|-----------|-----------| | Framework | LangGraph | Custom TS | Swarms | | Built-in tools | ✓ (full) | ✓ (trading) | ✓ (trading) | | Sub-agents | ✓ | ✗ | ✗ | | MCP support | ✓ | ✓ (planned) | ✗ | | Domain | General | Trading | Trading | | Studio UI | ✓ LangGraph | ✓ Web | ✗ | ---