PDFTalk

PDFTalk is a modern Retrieval-Augmented Generation (RAG) web application that lets you upload, search, and chat with your PDFs. It combines semantic search with powerful LLMs to deliver accurate, context-aware answers directly from your documents.

Screenshots

Example UI previews of the app

Upload PDF	Chat with Document

Homepage	Auth

---

Tech stack

• Framework: Next.js
• RPC / API: Hono / hono-rpc
• Styling: Tailwind CSS + shadcn/ui
• Data fetching: TanStack Query
• Vector DB: Pinecone
• LLM / embeddings: LangChain + Gemini
• Auth: Better Auth + Better Auth UI
• DB: PostgreSQL + Prisma

---

Features

• Upload & index PDF documents (extract text + embeddings)
• Search/query PDF content using vector similarity search (Pinecone)
• RAG answers using LangChain orchestration + Gemini LLM
• Authenticated users (Better Auth)
• Responsive UI using Tailwind + shadcn components
• Client caching & background queries with TanStack Query

---

Quickstart

Requirements

• Node.js (v18+)
• PostgreSQL
• Pinecone account and index
• Google/Anthropic/Vertex (Gemini) API key (or any LLM provider supported by LangChain)

Install

# install dependencies
bun install
# or
npm install

Environment

Create a .env file with the following (example):

DATABASE_URL="postgresql://<DB_USERNAME>:<DB_PASSWORD>@<DB_HOSTNAME>:<DB_PORT>/<DB_NAME>"

BETTER_AUTH_SECRET=t0h3BmZShSgBTtQe2BOun8J5AtBiWtPU
BETTER_AUTH_URL=http://localhost:3000

NEXT_PUBLIC_API_URL=http://localhost:3000
GOOGLE_API_KEY=

PINECONE_API_KEY=
PINECONE_ENVIRONMENT=us-east-1
PINECONE_INDEX_NAME=

Database

bunx prisma generate
bunx prisma migrate dev --name init

Local Dev

bun run dev
# or
npm run dev