DeepAgent Template

A comprehensive template for building Deep Agents using the DeepAgents library and LangGraph Studio.

This template showcases all key DeepAgent capabilities:

• 🔧 Custom Tools - Internet search with Tavily
• 🤖 Subagents - Specialized agents for different tasks
• 📁 File System - Built-in file operations
• 📋 Planning Tool - Task organization and tracking
• 🎨 LangGraph Studio - Visual debugging and development

Features

Core DeepAgent Capabilities

• Planning Tool: Automatically breaks down complex tasks into manageable steps
• File System: Create, read, edit, and manage files within the agent's context
• Subagents: Delegate specialized tasks to focused sub-agents
• Custom Tools: Extensible tool system for domain-specific functionality
• LangGraph Integration: Full compatibility with LangGraph Studio for visual debugging

Template Structure

src/
├── agent/
│   └── graph.py          # Main agent definition
├── tools/
│   └── tool.py           # Custom tools (internet search)
└── prompts/
    └── prompt.py         # Agent prompts and instructions

Quick Start

1. Install dependencies

uv add deepagents tavily-python
pip install -e . "langgraph-cli[inmem]"

2. Set up environment

cp .env.example .env
# Add your API keys to .env

3. Start LangGraph Studio

langgraph dev

4. Open Studio UI: https://smith.langchain.com/studio/?baseUrl=http://127.0.0.1:2024

Configuration

Environment Variables

Create a .env file with:

# Required for internet search
TAVILY_API_KEY=your_tavily_key

# Optional: LangSmith tracing
LANGSMITH_API_KEY=lsv2_your_key
LANGCHAIN_TRACING_V2=true

# Optional: Other LLM providers
ANTHROPIC_API_KEY=your_anthropic_key
OPENAI_API_KEY=your_openai_key

Custom Model

To use a different model, uncomment and modify in graph.py:

from langchain_anthropic import ChatAnthropic

agent = create_deep_agent(
    tools=[internet_search],
    instructions=research_instructions,
    subagents=[analysis_subagent],
    model=ChatAnthropic(model="claude-3-haiku-20240307")
)

Customization

Adding Tools

1. Create your tool function in src/tools/tool.py
2. Add it to the tools list in graph.py

Adding Subagents

1. Define the subagent prompt in src/prompts/prompt.py
2. Create the subagent config in graph.py:

my_subagent = {
    "name": "my-agent",
    "description": "What this agent does",
    "prompt": my_agent_prompt,
    "tools": ["tool_name"]  # Optional: restrict tools
}

Modifying Instructions

Edit src/prompts/prompt.py to customize the main agent's behavior and capabilities.

Development

File System Usage

The agent has access to a virtual file system. Example usage:

# In your agent interactions:
# - write_file("research.md", "# My Research\n...")
# - read_file("data.json")
# - edit_file("report.txt", old_content, new_content)

Subagent Usage

Call specialized subagents for focused tasks:

# The agent can call: "analysis-agent" for detailed technical analysis

LangGraph Studio

• Visual Debugging: See your agent's decision process step-by-step
• Hot Reload: Changes to your code are automatically applied
• State Management: Edit past states and rerun from any point
• Thread Management: Create new conversations with the + button

Example Usage

The template includes a research agent that demonstrates:

1. Planning: Breaking down research tasks
2. Search: Using Tavily for web research
3. Analysis: Calling subagents for detailed analysis
4. File Management: Organizing findings in files
5. Reporting: Compiling results into structured reports

Try asking: "Research the latest developments in AI agents and create a comprehensive report"

Resources

• DeepAgents Documentation
• LangGraph Documentation
• LangGraph Studio Guide
• Tavily Search API

License

MIT