Subconscious — Next.js + Groq + Gemini

Knowledge vault built with Next.js 15 App Router, NextAuth v5, Prisma, PostgreSQL + pgvector, Groq (Llama 3.3), and Gemini embeddings.

Entirely free AI tier — Groq free tier + Gemini free tier = zero cost.

Quick Start

# 1. Install
npm install

# 2. Configure
cp .env.example .env
# Edit .env — add NEXTAUTH_SECRET, GROQ_API_KEY, GEMINI_API_KEY

# 3. Start Postgres + pgAdmin
docker compose up postgres pgadmin -d

# 4. Run migrations
npx prisma migrate dev --name init

# 5. Start dev server
npm run dev

App → http://localhost:3000
pgAdmin → http://localhost:5050

Production (full Docker)

docker compose up -d

Free API Keys

Service	Get key	Cost
Groq (chat)	console.groq.com	Free tier: 14,400 req/day
Gemini (embeddings)	aistudio.google.com	Free: 1M tokens/month
Google OAuth	console.cloud.google.com	Free

Stack

Layer	Tech
Framework	Next.js 15 App Router
Auth	NextAuth v5 (JWT strategy, Google OAuth + credentials)
Database	PostgreSQL 16 + pgvector
ORM	Prisma
AI Chat	Groq SDK (llama-3.3-70b-versatile)
Embeddings	Google Gemini (text-embedding-004, 768 dims)
State	Redux Toolkit
UI	Tailwind CSS
Deploy	Docker Compose

Auth Flow

NextAuth v5 uses JWT session strategy — no database sessions table needed.

• Credentials: Register via /api/auth/register → sign in via NextAuth credentials provider
• Google OAuth: Single button, handled entirely by NextAuth → auto-creates user on first login
• Session: JWT stored in cookie (HTTP-only), 30-day expiry
• Server auth: auth() in Server Components and Route Handlers
• Client auth: useSession() from next-auth/react

RAG Chat

1. On save: content is embedded with text-embedding-004 (768 dims) → stored in contents.embedding as pgvector
2. On chat: question is embedded → cosine similarity search (<=> operator) finds top-6 matches
3. Groq Llama 3.3-70b generates answer using matched content as context
4. Embedding runs in setImmediate so saves are never blocked

Ports

Service	Port
Next.js	3000
PostgreSQL	5432
pgAdmin	5050