LangChain-Study

A simplified re-implementation of LangChain in ~200 lines of TypeScript. This is a study project to understand the LangChain framework and its components. The goal is to create a minimalistic version of LangChain that retains the core functionality while being easy to understand and modify.

This work is based on the blog post Re-implementing LangChain in 100 lines of code by Colin Eberhardt.

This project uses the following libraries:

• SerpAPI: For web search.
• OllamaJS: For interacting with an LLM.

Features

• Chainable calls to an LLM for reasoning and tool invocation (currently includes search and calculator tools).
• Conversational interface that retains context across multiple questions.
• Uses TypeScript for a more robust codebase.

Differences from Original Implementation

Language

• The original implementation was in JavaScript. This version is written in TypeScript, providing better type safety and maintainability.

Libraries

• SerpAPI is used instead of the fetch API to perform web searches and provide type definitions.
• OllamaJS is used for interacting with an LLM, replacing the OpenAI API calls and allowing testing with different local models.

Setup

Prerequisites

• Node.js (v22 or later)
• Ollama installed and running locally. You can install it from Ollama's website.
• An API key for SerpAPI. You can sign up for a free account at SerpAPI.

Steps to Run the Project

1. Clone the repository:

   git clone git@github.com:danst3in/langchain-study.git
   cd langchain-study

2. Install dependencies:

   npm install

3. Start the application:

   npm start

   or alternatively, if you want to run in watch mode while adjusting the prompt template or selected ollama model:

   npm run dev

4. Exit the application by pressing Ctrl+c to exit gracefully when you are done. (May require pressing Ctrl+c twice on some systems due to pending I/O operations.)