RMODZ AI — Self-Learning AI Chat & Curhat

Partner ngobrol AI yang asik, santai, dan supportive. Privacy-first, semua data tersimpan di device lu.

Quick Start

1. Install Dependencies

# From project root
npm install

# Install server dependencies
cd server && npm install

# Install client dependencies
cd ../client && npm install

# Or all at once from root:
npm run install:all

2. Configure Environment

cp .env.example .env

Edit .env:

# For testing (echo mode - no API key needed)
AI_PROVIDER=echo

# For real AI (OpenAI)
# AI_PROVIDER=openai
# AI_API_KEY=sk-your-key-here
# AI_MODEL=gpt-3.5-turbo

# For local AI (Ollama)
# AI_PROVIDER=ollama
# AI_MODEL=llama3
# AI_BASE_URL=http://localhost:11434

# Search (disabled by default)
# SEARCH_PROVIDER=duckduckgo

3. Development

npm run dev

• Frontend: http://localhost:5173
• Backend: http://localhost:3001

4. Production Build

npm run build
npm run start

How It Works

Memory System

All data stored locally in IndexedDB:

• Conversation Memory — recent chat context
• User Preferences — language, style, interests
• Explicit Memory — things you ask AI to remember
• Learning Memory — feedback patterns

No data leaves your device unless sent to AI provider for current response.

Feedback-Based Learning

1. Rate responses with 👍/👎
2. Negative feedback → optional reason selection
3. Patterns stored locally
4. AI adapts style based on recurring feedback

Web Search

• Disabled by default (free, private)
• Enable in settings for current info queries
• Uses DuckDuckGo (no API key needed)
• Sources shown with responses

AI Providers

| Provider | Cost | Setup | |----------|------|-------| | echo | Free | No config needed | | ollama | Free | Install Ollama locally | | openai | Paid | Get API key |

Architecture

rmodz-ai/
├── client/          # React + Vite + Tailwind
├── server/          # Express + TypeScript
└── shared/          # Shared types

Server Services

• AI Provider — abstraction for multiple AI backends
• Emotion Analysis — detects conversation mood
• Safety System — handles crisis situations
• Search Provider — web information retrieval
• Reasoning — orchestrates the response pipeline

Features

• Mobile-first responsive design
• Dark modern interface
• Markdown rendering with code blocks
• Conversation history
• Memory management
• Feedback system
• PWA installable
• Privacy-first (no accounts, local data)

Development from Android

Works with Acode + Termux:

# In Termux
pkg install nodejs
cd /path/to/rmodz-ai
npm run install:all
npm run dev

Known Limitations

• Echo mode only for testing without API key
• No streaming responses yet
• No voice input/output
• No image understanding
• Limited emotion detection (keyword-based)

Future

• Firebase auth & cloud sync
• Multiple AI providers
• Advanced RAG
• Voice input/output
• Image understanding
• Model fine-tuning

License

MIT