ChatGPT-style UI for document RAG with topics and threads

What it does

doc-chatbot is a Next.js app that ingests .pdf, .docx, and .txt files, vectorizes them through LangChain, and parks the embeddings in Pinecone. You query the results through a ChatGPT-inspired interface. The organizational hook is that files live inside topics, and every topic supports multiple independent chat windows with their own local-storage history.

The interesting bit

Most RAG tutorials stop at a single script; this adds a full React frontend with topic-based Pinecone namespaces and browser-side index management. It is essentially a polished glue layer around OpenAI, LangChain, and Pinecone, forked from a simpler PDF chatbot but rebuilt with a multi-chat, multi-topic architecture.

Key highlights

• Browser-based upload and deletion of Pinecone namespaces.
• Topics act as folders for files; each topic holds multiple isolated chat threads.
• Chat history persists in local storage by default; a separate mongodb-and-auth branch offers server-side persistence.
• Supports .pdf, .docx, and .txt inputs.
• Built on the standard retrieval stack: Next.js, React, Tailwind, LangChain, and Pinecone.

Caveats

• The mongodb-and-auth branch is several commits behind main and lacks certain features.
• Pinecone Starter-plan indexes are deleted after seven days of inactivity.
• File parsing can fail on corrupted, scanned, or OCR-dependent documents.

Verdict

A handy reference for developers who want a ready-made Next.js RAG interface with user-friendly organization primitives. Look elsewhere if you need a production-grade auth layer or a vector backend that isn’t tied to Pinecone’s free-tier lifecycle.