Skip to content

README.md

Latest commit

 

History

History
170 lines (126 loc) · 4.38 KB

README.md

File metadata and controls

170 lines (126 loc) · 4.38 KB

Sample Agent

A sample LangGraph agent demonstrating the supervisor pattern with custom handoff tools and message forwarding.

Overview

This sample agent showcases:

  • Supervisor Pattern: Central coordination between specialized sub-agents
  • Custom Handoff Tools: Seamless agent-to-agent communication
  • Message Forwarding: Efficient information flow between agents
  • Specialized Sub-agents: Math expert and research expert agents

Architecture

┌─────────────────┐
│   Supervisor    │
│     Agent       │
└─────────┬───────┘
          │
    ┌─────┴─────┐
    │           │
┌───▼───┐   ┌───▼───┐
│ Math  │   │Research│
│Expert │   │Expert │
└───────┘   └───────┘

Sub-agents

  • Math Expert: Handles mathematical calculations using add and multiply tools
  • Research Expert: Performs web searches and research tasks

Quick Start

Development Server

# Start development server
make dev sample-agent

# Start without opening browser
make dev sample-agent -- --no-browser

# With custom host/port
make dev sample-agent -- --host 0.0.0.0 --port 3000

Testing

# Run unit tests
make unit sample-agent

# Run integration tests
make integration sample-agent

# Run all tests
make test_apps

Linting & Formatting

# Lint code
make lint_apps

# Format code
make format_apps

Configuration

Environment Variables

Copy .env.example to .env and set:

# OpenRouter API Key
OPENROUTER_API_KEY=your_key_here

# LangSmith (optional)
LANGCHAIN_TRACING_V2=true
LANGCHAIN_PROJECT=your_project_name
LANGCHAIN_API_KEY=your_langsmith_key

Agent Configuration

The agent uses:

  • Model: Anthropic Claude Sonnet 4 via OpenRouter
  • Dependencies: Shared libs (shared, common)
  • Graph Entry Point: src/sample_agent/graph.py:make_graph

Project Structure

apps/sample-agent/
├── README.md                   # This file
├── pyproject.toml             # Package configuration
├── langgraph.json             # LangGraph configuration
├── .env                       # Environment variables
├── .env.example               # Environment template
├── src/sample_agent/          # Main package
│   ├── __init__.py
│   ├── graph.py              # Main graph definition
│   ├── state.py              # Agent state schema
│   ├── prompts.py            # Agent prompts
│   ├── context.py            # Context utilities
│   ├── subagents/            # Sub-agent definitions
│   │   ├── math.py           # Math expert agent
│   │   └── research.py       # Research expert agent
│   └── tools/                # Tool definitions
│       ├── basic.py          # Basic tools (add, multiply, web_search)
│       └── handoff.py        # Custom handoff tools
└── tests/                    # Test suite
    ├── conftest.py           # Test configuration
    ├── unit/                 # Unit tests
    └── integration/          # Integration tests

Usage Examples

Using the API

from sample_agent.graph import make_graph

# Create the graph
graph = make_graph()

# Use the agent
result = await graph.ainvoke({
    "messages": [{"role": "user", "content": "Calculate 15 * 24"}]
})

Development

# Import components
from sample_agent.state import AgentState
from sample_agent.subagents import create_math_agent, create_research_agent
from sample_agent.tools.handoff import create_custom_handoff_tool

Extending the Agent

  1. Add new sub-agents in src/sample_agent/subagents/
  2. Add new tools in src/sample_agent/tools/
  3. Update prompts in src/sample_agent/prompts.py
  4. Modify the graph in src/sample_agent/graph.py

API Endpoints

When running the dev server:

Dependencies

  • langgraph-supervisor>=0.0.28 - Supervisor pattern implementation
  • langgraph-up-devkits - Development utilities (workspace dependency)
  • Shared libraries from libs/shared and libs/common

License

MIT