research_agent

Description

research_agent leverages AI to automate the research process. It can: Analyze websites and extract relevant information. Generate summaries of findings. Create detailed reports in markdown format. The library supports multiple AI models

checkout frontend app build with the library: https://genie-ai.youssefhany.dev/app

• OpenAI
• Mistral AI (partial)

Getting Started

Install

pnpm add research_agent

Usage

import { writeFileSync } from 'fs';
import { ResearchManager } from 'research_agent';

const writeToFile = async (fileName, data) => {
  writeFileSync(fileName, data);
};

const main = async () => {
  const manager = new ResearchManager(); // You can customize the LLM model here
  const result = await manager.search('Who is this person in that website "www.youssefhany.dev"?');

  // Write the report to a markdown file
  await writeToFile('output.md', result);
};

void main();

Example with Mistral AI

import { WebPageLoaderWithSummary, QuestionGeneratorAgent, ResearchManager, NotificationManager } from 'research_agent';
import { ChatMistralAI } from '@langchain/mistralai';

const main = async (): Promise<void> => {
  const chatgpt = createOpenAIChatModel({
    modelName: 'gpt-3.5-turbo',
  });
  const mistralai = new ChatMistralAI({
    apiKey: process.env.MISTRAL_API_KEY,
    modelName: 'mistral-large-latest',
  });

  NotificationManager.addListener(console.log);

  const questionGeneratorAgent = new QuestionGeneratorAgent(mistralai); // still not stable with mistral ai
  const researchManager = new ResearchManager(mistralai, questionGeneratorAgent, {
    maxIterations: 2,
    researchWorkerConfig: {
      llmModel: chatgpt,
      tools: {
        webLoader: new WebPageLoaderWithSummary(mistralai),
      },
    },
  });
  const query = 'Write a report about ramadan and its importance';
  await researchManager.search(query);
};

await main();

This code snippet creates a ResearchManager instance and uses it to search for information about a person on a specific website. The search results are then written to a markdown file.

Features

• Multiple AI model support: Choose from OpenAI, Mistral AI, and potentially others.
• Question generation: Automatically generates questions to guide the research process.
• Web search and analysis: Utilizes tools like Tavily Search and Puppeteer to gather information from the web.
• Report generation: Creates comprehensive reports in markdown format.

Contributing

Contributions are welcome!

License

This project is licensed under the MIT license.

API Reference

For detailed information about the library's API, please refer to the exported classes and functions: ResearchManager: Manages the research process and generates reports. QuestionGeneratorAgent: Generates questions to guide research. SearchWorker: Performs web searches and analyzes websites.