Code Execution Agent

The project generates and executes code based on user-defined problems. It utilizes the Composio and connects your AI Agent to E2B’s Code Interpreter to facilitate code execution, allowing users to input a problem statement and receive executable code as output. The agent is designed to operate in a sandbox environment, ensuring safe execution and accurate results. Key functionalities include code generation, execution, and result interpretation, making it an invaluable resource for developers and data scientists alike.

JavaScript

1. Import Required Packages

   Import necessary packages for the Code Execution Agent:

   Import statements

   import dotenv from 'dotenv';
   import { ChatOpenAI } from "@langchain/openai";
   import { createOpenAIFunctionsAgent, AgentExecutor } from "langchain/agents";
   import { pull } from "langchain/hub";
   import { LangchainToolSet } from "composio-core";
   
   dotenv.config();
   
2. Initialize Composio Toolset

   Set up the Composio toolset and get the required tools:

   Connect to CodeInterpreter

   const toolset = new LangchainToolSet({ 
       apiKey: process.env.COMPOSIO_API_KEY
   });
   
   const tools = await toolset.getTools({ 
       actions: ["codeinterpreter_create_sandbox", "codeinterpreter_execute_code"] 
   });
   
3. Set up the AI Model

   Initialize the OpenAI ChatGPT model:

   Initialise Model

   const llm = new ChatOpenAI({ 
       model: "gpt-4o",
       apiKey: process.env.OPEN_AI_API_KEY
   });
   
4. Create the AI Agent

   Set up the agent’s prompt and create the OpenAI Functions Agent:

   Setup Agent

   const prompt = await pull("hwchase17/openai-functions-agent");
   const agent = await createOpenAIFunctionsAgent({ llm, tools, prompt });
   
5. Set up the Agent Executor

   Create the AgentExecutor to run the agent:

   Creating Agent Executor

   const agentExecutor = new AgentExecutor({ 
       agent, 
       tools, 
       verbose: true,
   });
   
6. Define the Code Execution Function

   Create the main function to generate and execute code:

   Main Function

   async function executeCodeAgent(userProblem) {
       // Generate code
       console.log("Generating code for the problem...");
       const codeGenerationResult = await agentExecutor.invoke({ 
           input: `Generate Python code to solve the following problem: ${userProblem}. 
                   Only provide the code, no explanations.`
       });
       const generatedCode = codeGenerationResult.output;
       console.log("Generated Code:", generatedCode);
   
       // Execute code
       console.log("\nExecuting the generated code...");
       const executionResult = await agentExecutor.invoke({ 
           input: `Execute the following Python code:\n${generatedCode}`
       });
       console.log("\nExecution Result:", executionResult.output);
   }
   
7. Run the Code Execution Agent

   Execute the agent with a sample problem:

   Run the Agent

   const userProblem = "Create a list of prime numbers up to 50";
   executeCodeAgent(userProblem).catch(error => console.error("An error occurred:", error));
   

Complete Code

Here’s the complete JavaScript code for the Code Execution Agent:

JavaScript Final Code

import dotenv from 'dotenv';
import { ChatOpenAI } from "@langchain/openai";
import { createOpenAIFunctionsAgent, AgentExecutor } from "langchain/agents";
import { pull } from "langchain/hub";
import { LangchainToolSet } from "composio-core";

dotenv.config();

async function executeCodeAgent(userProblem) {
    const toolset = new LangchainToolSet({ 
        apiKey: process.env.COMPOSIO_API_KEY
    });

    const tools = await toolset.getTools({ 
        actions: ["codeinterpreter_create_sandbox", "codeinterpreter_execute_code"] 
    });

    const llm = new ChatOpenAI({ 
        model: "gpt-4o",
        apiKey: process.env.OPEN_AI_API_KEY
    });

    const prompt = await pull("hwchase17/openai-functions-agent");
    const agent = await createOpenAIFunctionsAgent({ llm, tools, prompt });

    const agentExecutor = new AgentExecutor({ 
        agent, 
        tools, 
        verbose: true,
    });

    console.log("Generating code for the problem...");
    const codeGenerationResult = await agentExecutor.invoke({ 
        input: `Generate Python code to solve the following problem: ${userProblem}. 
                Only provide the code, no explanations.`
    });
    const generatedCode = codeGenerationResult.output;
    console.log("Generated Code:", generatedCode);

    console.log("\nExecuting the generated code...");
    const executionResult = await agentExecutor.invoke({ 
        input: `Execute the following Python code:\n${generatedCode}`
    });
    console.log("\nExecution Result:", executionResult.output);
}

const userProblem = "Create a list of prime numbers up to 50";
executeCodeAgent(userProblem).catch(error => console.error("An error occurred:", error));