What is chatbot-with-langgraph?

This project is a chatbot that lets you upload your own PDF documents and ask questions about their content, with the answers grounded in what those documents actually say. It combines a FastAPI backend, a Next.js frontend, a PostgreSQL database, and LangGraph, a library for building multi-step AI agent workflows. When a user uploads a PDF, it is stored in Amazon S3, and a background process extracts the text, breaks it into chunks, and indexes those chunks in Pinecone, a vector search database, so relevant passages can be found later. When a user sends a chat message with documents attached, LangGraph routes the request through a decision graph: it first searches the indexed document chunks for relevant context, falls back to a live DuckDuckGo web search if nothing relevant is found, and otherwise lets the model answer directly. The final response is generated using whatever context was gathered along the way. Conversations are saved turn by turn in Postgres, so chat history survives across sessions and server restarts. The app also has a long-term memory feature that can pull out useful facts or preferences from a conversation and store them for future sessions. Authentication uses JWT tokens stored in a secure cookie, with signup, login, and rate limiting on API routes, plus basic tracking of how many tokens each user has used. The frontend is built with Next.js and styled using Tailwind CSS and shadcn UI components, with Zustand handling client-side state and React Hook Form plus Zod handling form validation. The backend uses SQLAlchemy and Alembic for database models and migrations, and relies on Groq for running the underlying language model. Getting it running locally requires Python 3.13 or newer, Node.js 18 or newer, PostgreSQL, and API keys for Pinecone, Groq, and AWS S3.