ExamplesJavaScript Examples

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.
1

Import Required Packages

Import necessary packages for the Code Execution Agent:

Import statements
1import dotenv from 'dotenv';
2import { ChatOpenAI } from "@langchain/openai";
3import { createOpenAIFunctionsAgent, AgentExecutor } from "langchain/agents";
4import { pull } from "langchain/hub";
5import { LangchainToolSet } from "composio-core";
6
7dotenv.config();
2

Initialize Composio Toolset

Set up the Composio toolset and get the required tools:

Connect to CodeInterpreter
1const toolset = new LangchainToolSet({ 
2    apiKey: process.env.COMPOSIO_API_KEY
3});
4
5const tools = await toolset.getTools({ 
6    actions: ["codeinterpreter_create_sandbox", "codeinterpreter_execute_code"] 
7});
3

Set up the AI Model

Initialize the OpenAI ChatGPT model:

Initialise Model
1const llm = new ChatOpenAI({ 
2    model: "gpt-4o",
3    apiKey: process.env.OPEN_AI_API_KEY
4});
4

Create the AI Agent

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

Setup Agent
1const prompt = await pull("hwchase17/openai-functions-agent");
2const agent = await createOpenAIFunctionsAgent({ llm, tools, prompt });
5

Set up the Agent Executor

Create the AgentExecutor to run the agent:

Creating Agent Executor
1const agentExecutor = new AgentExecutor({ 
2    agent, 
3    tools, 
4    verbose: true,
5});
6

Define the Code Execution Function

Create the main function to generate and execute code:

Main Function
1async function executeCodeAgent(userProblem) {
2    // Generate code
3    console.log("Generating code for the problem...");
4    const codeGenerationResult = await agentExecutor.invoke({ 
5        input: `Generate Python code to solve the following problem: ${userProblem}. 
6                Only provide the code, no explanations.`
7    });
8    const generatedCode = codeGenerationResult.output;
9    console.log("Generated Code:", generatedCode);
10
11    // Execute code
12    console.log("\nExecuting the generated code...");
13    const executionResult = await agentExecutor.invoke({ 
14        input: `Execute the following Python code:\n${generatedCode}`
15    });
16    console.log("\nExecution Result:", executionResult.output);
17}
7

Run the Code Execution Agent

Execute the agent with a sample problem:

Run the Agent
1const userProblem = "Create a list of prime numbers up to 50";
2executeCodeAgent(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
1import dotenv from 'dotenv';
2import { ChatOpenAI } from "@langchain/openai";
3import { createOpenAIFunctionsAgent, AgentExecutor } from "langchain/agents";
4import { pull } from "langchain/hub";
5import { LangchainToolSet } from "composio-core";
6
7dotenv.config();
8
9async function executeCodeAgent(userProblem) {
10    const toolset = new LangchainToolSet({ 
11        apiKey: process.env.COMPOSIO_API_KEY
12    });
13
14    const tools = await toolset.getTools({ 
15        actions: ["codeinterpreter_create_sandbox", "codeinterpreter_execute_code"] 
16    });
17
18    const llm = new ChatOpenAI({ 
19        model: "gpt-4o",
20        apiKey: process.env.OPEN_AI_API_KEY
21    });
22
23    const prompt = await pull("hwchase17/openai-functions-agent");
24    const agent = await createOpenAIFunctionsAgent({ llm, tools, prompt });
25
26    const agentExecutor = new AgentExecutor({ 
27        agent, 
28        tools, 
29        verbose: true,
30    });
31
32    console.log("Generating code for the problem...");
33    const codeGenerationResult = await agentExecutor.invoke({ 
34        input: `Generate Python code to solve the following problem: ${userProblem}. 
35                Only provide the code, no explanations.`
36    });
37    const generatedCode = codeGenerationResult.output;
38    console.log("Generated Code:", generatedCode);
39
40    console.log("\nExecuting the generated code...");
41    const executionResult = await agentExecutor.invoke({ 
42        input: `Execute the following Python code:\n${generatedCode}`
43    });
44    console.log("\nExecution Result:", executionResult.output);
45}
46
47const userProblem = "Create a list of prime numbers up to 50";
48executeCodeAgent(userProblem).catch(error => console.error("An error occurred:", error));