langagent

LangAgent 2.1.6

LangAgent is a versatile multi-agent system designed to automate and streamline a wide range of complex tasks, including research, code generation, logical reasoning, data analysis, and dynamic reporting. Powered by advanced language models, LangAgent integrates seamlessly with APIs, databases, and various data formats, making it an essential tool for developers, data analysts, and business professionals to optimize workflows and extract valuable insights.

Key Features

• Research Automation: Automatically gather, analyze, and summarize information from multiple sources such as web content, databases, and documents.
• Code Generation & Debugging: Generate, debug, and optimize code for various programming languages, supporting complex multi-step tasks.
• Mathematical Reasoning: Solve logical and mathematical queries with clear step-by-step breakdowns.
• Data Analysis & Topic Generation: Perform data clustering, trend analysis, and generate actionable insights from structured and unstructured data.
• Professional Reporting: Create detailed, high-quality reports with dynamic data visualizations and clear summaries.
• Supervisor Chain: Manage and coordinate workflows between multiple agents for complex tasks.

Installation

You can install the langagent package via pip:

pip install langagent==2.1.6

Getting Started

To start using the langagent library, you need to import the agents from their respective teams. Below is a detailed guide explaining each agent, its arguments, and how to effectively use them in your projects.

Research Team Agents

The Research Team consists of three agents: Researcher, Coder, and Weather. Each agent is designed to assist with gathering research, generating code, or fetching weather data.

1. Researcher Agent

   • The researcher agent gathers and synthesizes information based on user queries. It automates the research process by providing structured and concise insights.

   Arguments:

   • llm: The language model to be used (e.g., ChatOpenAI).
   • messages: A list of HumanMessage objects representing the research query.

   Example:

   from langagent.research_team.agents import researcher
   from langchain_openai import ChatOpenAI
   from langchain_core.messages import HumanMessage
   import yaml
   
   llm = ChatOpenAI()
   
   os.environ['TAVILY_API_KEY'] = yaml.safe_load(open('credentials.yml'))['online']
   
   search = {
       'api_key': os.environ.get("TAVILY_API_KEY"),
       'max_results': 5,
       'search_depth': "advanced"
   }
   
   researcher_agent = researcher.researcher(llm=llm, tavily_search=search)
   
   # Invoke the agent with a query
   result = researcher_agent.invoke({"messages": [HumanMessage(content="Who is Messi?")]})
   print(result['output'])
   

Example of Resarcher Workflow

1. Coder Agent

   • The coder agent generates, debugs, and optimizes code based on user input. It handles various programming challenges, from basic functions to advanced tasks.

   Arguments:

   • llm: The language model you want to use.
   • messages: A list of HumanMessage objects representing the code generation task.

   Example:

   from langagent.research_team.agents import coder
   from langchain_openai import ChatOpenAI
   from langchain_core.messages import HumanMessage
   
   llm = ChatOpenAI()
   coder_agent = coder(llm)
   
   # Ask the agent to generate code
   result = coder_agent.invoke({"messages": [HumanMessage(content="Write a Python function to calculate the factorial of a number.")]} )
   print(result['output'])
   

Example of Coder Workflow

1. Weather Agent

   • The weather agent fetches and interprets weather data for a specified location using API integrations.

   Arguments:

   • llm: The language model to be used.
   • OPENWEATHERMAP_API_KEY: Your OpenWeatherMap API key.
   • messages: A list of HumanMessage objects representing the weather query.

   Example:

   from langagent.research_team.agents import weather
   from langchain_openai import ChatOpenAI
   from langchain_core.messages import HumanMessage
   
   llm = ChatOpenAI()
   weather_agent = weather(llm, OPENWEATHERMAP_API_KEY="your-api-key-here")
   
   # Ask for weather details
   result = weather_agent.invoke({"messages": [HumanMessage(content="What is the weather today in New York?")]})
   print(result['output'])
   

Example of Weather Workflow

Logic Team Agents

The Logic Team consists of two agents: Reasoner and Calculator, both designed to solve logical and mathematical problems efficiently.

1. Reasoner Agent

   • The reasoner agent handles logical reasoning tasks and provides clear, step-by-step explanations for complex problems.

   Arguments:

   • llm: The language model to be used.
   • messages: A list of HumanMessage objects describing the logic problem.

   Example:

   from langagent.logic_team.agents import reasoner
   from langchain_openai import ChatOpenAI
   from langchain_core.messages import HumanMessage
   
   llm = ChatOpenAI()
   reasoner_agent = reasoner(llm)
   
   # Solve a logic problem
   result = reasoner_agent.invoke({"messages": [HumanMessage(content="I have a 7 in the tens place. I have an even number in the ones place. I am lower than 74. What number am I?")]})
   print(result['output'])
   

Example of Reasoner Workflow

1. Calculator Agent

   • The calculator agent solves mathematical queries and calculations based on natural language inputs.

   Arguments:

   • llm: The language model to be used.
   • messages: A list of HumanMessage objects containing the mathematical query.

   Example:

   from langagent.logic_team.agents import calculator
   from langchain_openai import ChatOpenAI
   from langchain_core.messages import HumanMessage
   
   llm = ChatOpenAI()
   calculator_agent = calculator(llm)
   
   # Solve a mathematical problem
   result = calculator_agent.invoke({"messages": [HumanMessage(content="Calculate the square root of 16")]})
   print(result['output'])
   

Example of Calculator Workflow

Analysis Team Agents

The Analysis Team consists of two agents: Topic Generator and SQL Databaser, both designed to provide insights from data, either through text clustering or SQL-based analysis.

1. Topic Generator Agent

   • The topic generator agent clusters and analyzes text data, producing topics and visualizations based on user data.

   Arguments:

   • llm: The language model to be used.
   • path: The path to the CSV file containing text data.
   • text_column: The column in the CSV that contains the text data.
   • user_topics: Optional, user-provided topics for clustering.

   Example:

   from LangAgent.analysis_team.agents import topic_generator
   from sentence_transformers import SentenceTransformer
   from langchain_openai import ChatOpenAI
   
   llm = ChatOpenAI()
   topic_generator_agent = topic_generator(llm)
   
   inputs = {
       'path': '../Comments.csv',
       'text_column': 'Comments',
       'user_topics': None
   }
   
   result = topic_generator_agent.invoke(inputs)
   result['summary_df']  # View the summarized data
   result['fig'].show()  # Show the visualization
   

Example of Topic Generator Workflow

1. SQL Databaser Agent

   • The SQL databaser agent executes SQL queries and provides insights from the results, including generating charts.

   Arguments:

   • db_path: The path to your SQLite or SQL database.
   • llm: The language model.
   • user_question: The question that drives the SQL query and chart generation.

   Example:

   from langagent.analysis_team.agents import sql_databaser
   from langchain_openai import ChatOpenAI
   from langchain_experimental.utilities import PythonREPL
   
   llm = ChatOpenAI()
   PATH_DB = "sqlite:///database/leads_scored.db"
   sql_databaser_agent = sql_databaser(db_path=PATH_DB, llm=llm)
   
   question = """
   What are the total sales by month-year? 
   Use suggested price as a proxy for revenue for each transaction and a quantity of 1. 
   Make a line chart of sales over time.
   """
   
   initial_input = {
       "user_question": question
   }
   
   # Invoke the agent with the input
   result = sql_databaser_agent.invoke(initial_input)
   print(result['sql_query'])  # SQL Query
   print(result['summary'])  # Summary of results
   
   # Execute the Python code for chart generation
   repl = PythonREPL()
   repl.run(result['chart_plotly_code'])
   

Example of SQL Databaser Workflow

Reporting Team Agents

The Reporting Team consists of two agents: Interpreter and Summarizer, both designed to provide insights and summaries of outputs from various sources such as data, charts, or documents.

1. Interpreter Agent

   • The interpreter agent provides clear, insightful interpretations of various types of outputs, including visual plots, tables, and data.

   Arguments:

   • llm: The language model.
   • code_output: The result or output that needs to be interpreted.

   Example:

   from langagent.reporting_team.agents import interpreter
   from langchain_openai import ChatOpenAI
   
   llm = ChatOpenAI()
   interpreter_agent = interpreter(llm)
   
   result = interpreter_agent.invoke({"code_output": "Bar chart showing sales data."})
   print(result['output'])
   

Example of Interpreter Workflow

1. Summarizer Agent

   • The summarizer agent summarizes documents like PDFs, PPT, DOCX, TXT files or Plain Text into concise, actionable insights.

   Arguments:

   • llm: The language model.
   • input_data: The path to the document or text file that needs to be summarized.

   Example:

   from langagent.reporting_team.agents import summarizer
   from langchain_openai import ChatOpenAI
   
   llm = ChatOpenAI()
   summarizer_agent = summarizer(llm)
   
   inputs = {
       'input_data': '../Comments.csv'
   }
   
   result = summarizer_agent.invoke(inputs)
   print(result['summary'])
   

Example of Summarizer Workflow

Supervisor Chain

The Supervisor Chain manages the workflow between multiple agents. It selects the next agent based on predefined criteria or completion conditions. This is useful when you want to automate multi-step tasks and route tasks to different agents based on the flow of the conversation or task requirements.

• The Supervisor Chain allows you to manage complex workflows by selecting different agents based on their capabilities or status.

Arguments:

• subagent_names: A list of subagents (workers) that will perform tasks.
• llm: The language model that powers decision-making.
• subagent_descriptions: A dictionary of subagent names with descriptions of their roles (optional).
• completion_criteria: A string representing the criteria that signals completion of the workflow.
• history_depth: The number of previous messages to consider when making routing decisions (optional).

Example:

from langagent.supervisor import supervisor_chain
from langchain_openai import ChatOpenAI

llm = ChatOpenAI()

# Define the subagents and their descriptions
subagent_names = ["researcher", "coder", "calculator"]
subagent_descriptions = {
    "researcher": "Finds relevant research and gathers data.",
    "coder": "Generates, debugs, and optimizes code.",
    "calculator": "Solves mathematical problems and queries."
}

# Create a supervisor chain to manage the workflow
supervisor = supervisor_chain(
    subagent_names=subagent_names, 
    llm=llm, 
    subagent_descriptions=subagent_descriptions, 
    completion_criteria="FINISH"
)

Example of Supervision Workflow

Dependencies

LangAgent relies on several core libraries, including:

• LangChain for agent creation and task management.
• Sentence Transformers for text embedding.
• SQLAlchemy for database interactions.
• Pandas for data manipulation.
• Plotly and Matplotlib for data visualization.

Install the required dependencies using:

pip install -r requirements.txt

Contributing

We welcome contributions to LangAgent! If you'd like to contribute:

1. Fork the repository.
2. Create a branch for your feature (git checkout -b feature/new-feature).
3. Commit your changes (git commit -m "Add new feature").
4. Push your branch (git push origin feature/new-feature).
5. Create a Pull Request.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Contact

For any inquiries or issues, please contact:

• Author: Kwadwo Daddy Nyame Owusu - Boakye
• Email: [e-mail]
• GitHub: https://github.com/knowusuboaky/LangAgent