gemini-fullstack-langgraph-quickstart

This repository is a quickstart template showing how to build a full-stack web application where an AI agent researches topics on your behalf. The frontend is a browser-based chat interface; the backend is a Python server powered by a framework called LangGraph, which controls how the AI thinks step by step.

When you type a question, the agent does not simply reply from memory. Instead it automatically generates a set of search queries, looks up web pages using Google Search, then reflects on what it found to check for gaps in its knowledge. If gaps exist, it generates follow-up searches and repeats the loop. Once satisfied, it writes a final answer that includes citations linking back to its sources. This cycle of search, reflect, and refine is configurable — you can set a maximum number of loops.

To run it locally you need a Google Gemini API key (for the AI model) and Node.js plus Python 3.11 or later. A single make command starts both the frontend and backend servers at once. For production deployment, the app is packaged as a Docker image; it requires a Redis instance (used for streaming real-time output) and a PostgreSQL database (used to store conversation history and manage background tasks).

The project is officially a learning example demonstrating how to combine a React frontend with a LangGraph research agent powered by Google's Gemini models.