LangGraph Agent Template

A template project demonstrating how to build a sequential agent workflow using LangGraph and LangChain.

Overview

This project implements a multi-step agent workflow using LangGraph's StateGraph. The agent processes input through three distinct stages:

1. Initial processing (initial_step)
2. LLM-based processing (model_call)
3. Final output generation (final_step)

Project Structure

• _agent_graph.py: Main implementation file containing the agent workflow
• README.md: This documentation file

Key Components

State Management

The project uses TypedDict classes for strict type checking across different stages:

• InputState: Handles initial input parameters

  • agent_input_a: First input parameter
  • agent_input_b: Second input parameter

• OverallState: Manages intermediate state during processing

  • temp_value_a: Temporary string value
  • temp_value_b: List of temporary string values

• OutputState: Defines the final output format

  • agent_output_value: Final processed result

Processing Steps

1. Initial Step: Performs preliminary processing on input data
2. Model Call: Integrates with GPT-4 for advanced processing
   • Uses a custom prompt from LangChain Hub
   • Implements structured output parsing using Pydantic
3. Final Step: Transforms processed data into the required output format

Dependencies

• LangChain
• LangGraph
• OpenAI
• Pydantic

Client & SDK usage

main.py demonstrates the use of LangGraph python SDK to call the Agent hosted in LangGraph cloud.