Tool Calling Agent with More LLM Models

Open in ColabOpen in GitHub

Overview

LangChain is not limited to OpenAI models. It also supports implementations from diverse LLM providers such as Anthropic, Google Gemini, Together.ai, Ollama, and Mistral. This flexibility allows developers to leverage the unique characteristics and strengths of each model to create agents optimized specific requirements for their applications.

Key Topics

In this chapter, we will delve into the process of creating and executing tool-calling agents using various LLMs. Here are the key topics covered, we'll explore:

  • Tool Selection: How agents choose the most suitable tools for specific tasks.

  • LLM Integration: Integrating LLMs from OpenAI and other providers into LangChain to enable agent functionality.

  • Agent Creation: Creating agents using LangChain's agent classes.

  • Agent Execution: Executing agents to perform tasks.

Objectives By the end of this chapter, you will be able to:

  • How to create and execute tool-calling agents using various LLMs.

  • Create automated workflows that call various tools using LangChain's agent classes.

  • Combine multiple LLMs to implement agents with optimized performance.

Now, let’s explore how to maximize productivity using LangChain’s flexible agent framework. 🚀

Table of Contents

References

Environment Setup

Set up the environment. You may refer to Environment Setup for more details.

[Note]

  • langchain-opentutorial is a package that provides a set of easy-to-use environment setup, useful functions and utilities for tutorials.

  • You can checkout the langchain-opentutorial for more details.

  • langchain-ollama is a package that provides a set of easy-to-use environment setup, useful functions and utilities for tutorials.

List of LLMs Supporting Tool Calling

Before we proceed with the hands-on tutorial, you'll need to complete the following setup steps for each LLM you want to use:

  1. Obtain an API Kye: Follow the provided link to requiest an API key for each LLM call.

  2. Add the issued key to the .env file.

Anthropic

Gemini

Ollama

  • List of Ollama Tool Calling Supported Models

  • Ollama uses a different approach. Instead of API keys, you'll need to install Ollama itself. Follow the instructions here to install Ollama: Ollama installation

  • This tutorial will use the lama3.1 model

  • After installing Ollama, you can download it using the following commands: ollama pull llama3.1

  • You can also download the qwen2.5 model using the following command: ollama pull qwen2.5

Together AI

Working with Multiple LLM Integrations in LangChain

This section guides you through integrating and configuring various LLMs in LangChain, allowing you to do experiments with different models from providers like OpenAI, Anthropic, Google, and others.

Creating tools

Before creating tools, we will build some functions to fetch news from websites based on user's input keywords.

_fetch_news(url: str, k: int = 3) -> List[Dict[str, str]]: This funtion takes a URL as input and retrieves news articles from that source. The function returns a list of dictionaries.

  • Args: url: str is for fetching news articles. The k: int = 3 (default: 3) is a number of news to fetch.

  • Return: List[Dict[str, str]] is a list of dictionaries that contains news title and link.

_collect_news(news_list: List[Dict[str, str]] -> List[Dict[str, str]]: This function return a sorted list of the same news items.

  • Args: news_list: List[Dict[str, str]] is a list of dictionaries that contains news information.

  • Return: List[Dict[str, str]] is a list of dictionaries containing the URL and the full contents.

search_by_keyword(keyword: str, k: int = 3) -> List[Dict[str, str]]: This funtion is the main entry point for searching news. It accepts a keyword and returns a list of dictionaries.

  • Args: keyword: str is a keyword to search. k: int = 3(default: 3) is a number of news to fetch.

  • Return: List[Dict[str, str]] is a list of dictionaries that contains the URL and contents.

This set of functions enables a tool that can fetch relevant news from Google News Website based on user-provided input keywords.

Let's create tools.

Generating Prompts for Agents

A prompt is text that describes the task the model will perform whose input is the tool name and its role.

  • chat_history: A variable that stores previous conversation history (can be omitted if multi-turn support is not required).

  • agent_scratchpad: A variable for temporary storage used by the agent.

  • input: The user's input.

Let's generate agents per each LLM basis.

Generating an AgentExecutor

Now, let's import AgentExecutor, run agents, and review the outputs.

The following function generates and runs an agent using the provided LLM and outputs the results.

Generate and run agents for each LLM and outputs the results.

PreviousTool Calling AgentNextIteration-human-in-the-loop

Last updated