basic-llm-agent

A simple example of building an LLM agent from scratch in python.

Inspired by Aaron Dunn's great tutorial: https://youtu.be/mYo7UFwnW1k?si=uSZ_fhA8rJPsApkz

Overview

This project illustrates the construction of a simple Large Language Model (LLM) agent using Python. It serves as an educational example for developers who want to understand the fundamental concepts and implementation details of an LLM agent without relying on external agent frameworks.

Features

• Built from scratch in Python with minimal dependencies
• Demonstrates the core concepts of LLM-based agents
• Includes various tools like web search, calculator, and market data analysis
• Web search capabilities through Tavily API

Requirements

• Python 3.8+
• OpenAI API key (or other LLM provider)
• Tavily API key for web search functionality
• Required Python packages:
  • openai>=1.0.0
  • tavily-python>=0.5.1
  • python-dotenv>=1.0.0
  • yfinance>=0.2.54
  • pandas>=2.0.3

Installation

1. Clone this repository:

   git clone https://github.com/yourusername/basic-llm-agent.git
   cd basic-llm-agent

2. Install the required packages:

   pip install -r requirements.txt

3. Set up your API keys:

   Create a .env file in the root directory with the following content:

   OPENAI_API_KEY=your_api_key_here
   TAVILY_API_KEY=your_tavily_api_key_here
   

   Note: Please check that the .env file is included in the .gitignore file to ensure your API keys are never accidentally committed to version control.

   Alternatively, you can set these as environment variables:

   export OPENAI_API_KEY=your_api_key_here
   export TAVILY_API_KEY=your_tavily_api_key_here

   You can obtain a Tavily API key by signing up at https://tavily.com/

   You can obtain an OpenAI API key by signing up at https://platform.openai.com/

Usage

Use the agent_query function in llm.py:

from llm import agent_query

# Ask the agent to perform a task
response = agent_query("What is the sum of today's high temperature in Paris and Boston?")
print(response)

The repository includes both simple and advanced usage examples at the bottom of llm.py.

LLM Providers

This project implements a flexible architecture that allows you to use different language model providers:

• OpenAIProvider: Uses OpenAI's API (GPT models) for generating responses
• OllamaProvider: Uses locally run Ollama models for generating responses

You can easily specify which provider to use when creating an agent:

from llm import Agent

# Use Ollama (default)
agent = Agent()

# Or use OpenAI
agent = Agent(provider="openai")

# Generate a response
response = agent.generate_response("Your prompt here")

Both providers implement the same interface, making it easy to switch between them or add new providers.

License

MIT License