StreamAI — AI Chat

StreamAI is a Next.js 15 app that blends fast, streaming AI chat with personal long‑term memory and Retrieval‑Augmented Generation (RAG). Bring your own data from PDFs, websites, text, and YouTube; index it into Qdrant; and chat with it using Google Generative AI embeddings. It features Google login, persistent chat history, code/maths rendering, and an ergonomic UI.

Key Features

• Streaming chat UX: Low‑latency, incremental responses with a virtualized message list for smooth long threads.
• BYO data (RAG mode): Upload PDFs and text, crawl websites, or pull YouTube transcripts. Indexed to Qdrant for semantic retrieval.
• Personal memory: Optional Mem0 memory integration to remember preferences, facts, and context across chats.
• Rich messages: Code blocks with syntax highlighting, KaTeX math, inline images, and attachments.
• Auth & security: Better Auth with Google provider; per‑user collections, chats, and ACL checks.
• Modern stack: Next.js 15, React 19, Tailwind v4, Prisma 6, Postgres, Qdrant, AI SDK.

Why StreamAI over generic AI chat apps?

• Your data, first‑class: Native RAG across PDFs, websites, text, and YouTube—not just file uploads.
• Search that cites: Retrieval metadata is preserved for transparency and debugging.
• Long‑term memory: Mem0 integration augments assistant behavior beyond a single session.
• Developer‑friendly: Clear server actions, typed schemas, and modular libs for rapid extension.
• Production‑ready plumbing: Prisma models, auth, vector store wiring, and clean UI out of the box.

Tech Stack

• Web: Next.js 15, React 19, Tailwind v4, Radix UI, Zustand
• AI: AI SDK with Google Generative AI, @langchain/*, Google embeddings
• RAG: Qdrant (JS client), chunking/indexing pipelines
• DB: Prisma 6 with PostgreSQL
• Auth: Better Auth (Google OAuth)
• Storage/Media: Cloudinary (optional)

Quick Start

Prerequisites

• Node.js 20+
• PostgreSQL database
• Qdrant (managed or self‑hosted)

1) Clone and install

git clone https://github.com/your-org/streamai.git
cd streamai
bun install  # or npm install / bun install / yarn

2) Environment variables

Create .env with the following (set values as applicable):

# Database
DATABASE_URL="postgresql://user:password@localhost:5432/streamai"

# Auth (Better Auth)
GOOGLE_CLIENT_ID="..."
GOOGLE_CLIENT_SECRET="..."
BETTER_AUTH_URL="http://localhost:3000"  # public URL of the app (prod: https://yourdomain)

# AI / Embeddings (Google Generative AI)
GOOGLE_API_KEY="..."  # any of these keys are supported in code paths
# GOOGLE_AI_API_KEY=...
# GOOGLE_GENERATIVE_AI_API_KEY=...
# GEMINI_API_KEY=...

# Vector DB (Qdrant)
QDRANT_URL="http://localhost:6333"
QDRANT_API_KEY=""  # set if your Qdrant requires auth

# Optional: Cloudinary for media
CLOUDINARY_CLOUD_NAME="..."
CLOUDINARY_API_KEY="..."
CLOUDINARY_API_SECRET="..."

# Optional: Weather tool
WEATHER_API_KEY="..."

# Optional: Mem0 memory
MEM0_API_KEY="..."

Notes:

• The app checks multiple Google key env names; providing any one is sufficient.
• Cloudinary, Weather, and Mem0 are optional; affected features gracefully no‑op if unset.

3) Database setup

bun run prisma:generate
bun run prisma:migrate  # or: bun db:push

4) Run the app

bun dev
# visit http://localhost:3000

Core Concepts

Chats

• Stored in Postgres via Prisma (User, Chat, Message).
• Streaming responses with a virtualized list for performance.

RAG Mode

• Create per‑user RagCollections from PDFs, websites, text, or YouTube.
• Content is chunked, embedded with Google embeddings, and indexed into Qdrant.
• Chat with a collection via RagChat and RagMessage records; sources can be attached to responses.

Memory (optional)

• If MEM0_API_KEY is set, the app writes durable facts/preferences to Mem0 and reads them to personalize replies.

Project Structure

/src
  app/               # Next.js routes and server actions
  components/        # UI components (chat, sidebar, RAG modal, etc.)
  lib/               # AI, RAG, Prisma, auth, utilities
  stores/            # Zustand stores
  types/             # Shared types
/prisma              # Prisma schema and migrations

Scripts

bun dev            # start dev server (Next.js turbopack)
bun build          # prisma generate + next build
bun start          # start production server
bun prisma:migrate # run Prisma migrations
bun prisma:studio  # open Prisma Studio

Deployment

• Works great on Vercel or any Node host.
• Set all required environment variables in your hosting provider.
• Ensure your BETTER_AUTH_URL (or public URL) matches the deployed domain.

Security & Privacy

• Per‑user isolation for chats and RAG collections.
• Do not commit .env. Rotate keys regularly. Use least‑privilege keys for Qdrant and Cloudinary.

Troubleshooting

• Missing keys: features log warnings and skip optional integrations.
• Qdrant connection: verify QDRANT_URL and QDRANT_API_KEY and that collections are created.
• DB migrations: run bun prisma:migrate after schema changes.

License

MIT