Unvios is an AI-powered personal memory assistant built with Next.js 16. It stores, retrieves, and intelligently recalls user memories using LLM integration and vector embeddings, creating a personalized knowledge base that grows with you.

• AI Memory System: Store personal facts, preferences, and experiences with automatic semantic search
• Vector Embeddings: Optional pgvector integration for intelligent memory retrieval based on context
• LLM Chat Interface: Conversational AI that references your stored memories in responses
• Memory Management: Full CRUD operations with tagging, editing, and duplicate detection
• Activity Logging: Track user interactions and system events

• Cookie-based JWT authentication with email/password sign-up
• Stripe integration for subscription management
• Protected dashboard routes with role-based access

• Interactive onboarding tour for new users
• Real-time chat interface with memory annotations
• Dashboard with memory browser and analytics
• Marketing pages (features, pricing, about, blog, research)

• Framework: Next.js 16 (canary) with App Router
• Database: PostgreSQL with pgvector
• ORM: Drizzle
• LLM Integration: OpenAI-compatible API endpoints
• Embeddings: Configurable embedding service for vector search
• Payments: Stripe
• UI Library: shadcn/ui
• Styling: Tailwind CSS

• Node.js 18+ and pnpm
• PostgreSQL database with pgvector extension (or use USE_LOCAL_MEMORIES=1 for development)
• Optional: OpenAI API key or compatible LLM endpoint

git clone https://github.com/olserra/unvios
cd unvios
pnpm install

Use the included setup script to create your .env file:

pnpm db:setup

Or manually create a .env file with the following variables:

# Database
POSTGRES_URL=postgresql://user:password@localhost:5432/unvios

# Auth (auto-generated in dev if not provided)
AUTH_SECRET=your-secret-key-here

# Base URL
BASE_URL=http://localhost:3000

# LLM API (OpenAI or compatible endpoint)
LLM_API_URL=https://api.openai.com/v1
LLM_API_KEY=your-llm-api-key

# Embedding API (for vector search)
EMBEDDING_API_URL=https://api.openai.com/v1
EMBEDDING_API_KEY=your-embedding-api-key

# Stripe (for payments)
STRIPE_SECRET_KEY=your-stripe-secret-key
STRIPE_WEBHOOK_SECRET=your-webhook-secret

# Dev mode (skip DB, use local JSON file)
USE_LOCAL_MEMORIES=1

Run migrations to create the database schema:

pnpm db:migrate

Seed the database with a default user:

pnpm db:seed

This creates a test user:

• Email: [email protected]
• Password: admin123

pnpm dev

Open http://localhost:3000 to see the app.

• pnpm dev - Start development server with Turbopack
• pnpm build - Build for production
• pnpm start - Start production server
• pnpm lint - Run ESLint
• pnpm db:setup - Initialize database and create .env
• pnpm db:migrate - Run Drizzle migrations
• pnpm db:seed - Seed database with test data
• pnpm db:reset-seed - Reset and re-seed database
• pnpm test - Run Vitest unit tests
• pnpm test:e2e - Run Playwright e2e tests

The memory system uses a special annotation format in LLM responses:

[MEMORY: User prefers dark mode | preferences, ui]
[MEMORY: Lives in San Francisco | location, personal]

The chat API (app/api/llm/chat/route.ts) automatically:

1. Parses memory annotations from LLM responses
2. Checks for duplicates using vector similarity
3. Saves new memories with optional embeddings
4. Retrieves relevant memories for context in future conversations

When embeddings are enabled:

1. User message is embedded
2. Nearest memories are retrieved using cosine distance (<=>)
3. Memories are included in LLM prompt for context
4. New memories are embedded and stored with vector column

Set USE_LOCAL_MEMORIES=1 to use lib/devMemories.ts instead of the database. Useful for:

• Quick prototyping without DB setup
• Testing memory logic in isolation
• Offline development

app/
  ├── (dashboard)/        # Protected dashboard pages
  │   ├── chat/          # AI chat interface
  │   └── dashboard/     # Memory browser, settings
  ├── (login)/           # Auth pages (sign-in, sign-up)
  ├── api/               # Next.js API routes
  │   ├── llm/chat/      # LLM chat endpoint with memory integration
  │   ├── memories/      # Memory CRUD + embeddings
  │   ├── auth/          # Authentication endpoints
  │   └── stripe/        # Payment webhooks
  └── [marketing pages]  # Public pages (about, pricing, etc.)

lib/
  ├── auth/              # Session management, JWT handling
  ├── db/                # Drizzle schema, queries, migrations
  ├── memoryParser.ts    # Memory annotation parsing logic
  └── devMemories.ts     # Local development fallback store

components/
  ├── dashboard/         # Chat, memory panels, onboarding
  └── ui/                # shadcn/ui primitives

scripts/
  ├── backfill-embeddings.ts  # Generate embeddings for existing memories
  ├── dev-seed.ts             # Seed development data
  └── reset-and-seed.ts       # Full reset workflow

pnpm test

Tests are in tests/ and use Vitest. Current coverage includes:

• Memory parser logic (tests/memoryParser.test.ts)
• Utility functions (tests/utils.test.ts)

pnpm test:e2e

Playwright tests verify critical user flows end-to-end.

Set all required variables in your production environment:

POSTGRES_URL=your-production-db-url
AUTH_SECRET=$(openssl rand -base64 32)
BASE_URL=https://yourdomain.com
LLM_API_URL=your-llm-endpoint
LLM_API_KEY=your-llm-key
EMBEDDING_API_URL=your-embedding-endpoint
EMBEDDING_API_KEY=your-embedding-key
STRIPE_SECRET_KEY=your-stripe-secret-key
STRIPE_WEBHOOK_SECRET=your-webhook-secret

Ensure your production database has the pgvector extension:

CREATE EXTENSION IF NOT EXISTS vector;

Run migrations:

pnpm db:migrate

1. Create a production webhook in Stripe Dashboard
2. Point it to https://yourdomain.com/api/stripe/webhook
3. Select events: checkout.session.completed, customer.subscription.updated, etc.
4. Copy the webhook secret to STRIPE_WEBHOOK_SECRET

Deploy to Vercel or your preferred platform:

vercel --prod

See docs/adr/ for detailed Architecture Decision Records covering:

• CI/CD workflows and caching strategies
• Security audit and fixes
• Cookie consent implementation
• Onboarding tour implementation
• Test infrastructure

When making significant architectural changes, please create an ADR in docs/adr/ using the template at docs/adr/adr-template.md.

See SECURITY_FIXES_APPLIED.md for recent security improvements.

Report security issues to: [security contact info]

See LICENSE file for details.