WhatsApp LLM Bot (سوسي)

A WhatsApp bot using Baileys (multi-device). It replies in Arabic as an anime expert “سوسي”, triggered by hotwords (e.g., سوسي). Uses OpenAI if configured, otherwise falls back to a local Ollama model.

Features

• QR login (multi-file auth state)
• Hotword-triggered replies in groups/private
• Very short answers (3–4 words), off-topic => "هممم"
• OpenAI first; Ollama fallback

Prerequisites

• Node.js 18+
• WhatsApp on your phone
• Optional: Ollama running on http://localhost:11434

Setup

1. Install

npm install

2. Create .env next to bot.js:

# Bot
GROUP_HOTWORDS=سوسي,يا سوسي

# OpenAI (recommended)
OPENAI_API_KEY=your_openai_key
OPENAI_MODEL=gpt-4.1

# Ollama fallback
OLLAMA_MODEL=command-r7b-arabic

Important: Do not commit .env. Rotate any keys that were exposed.

3. Run

npm start

Scan the QR in your terminal.

Usage

Mention a hotword (e.g., “سوسي …”) in any chat. The bot will reply briefly in Arabic.

Docker (optional)

A simple compose example:

services:
  wa-llm-bot:
    build: .
    container_name: wa-llm-bot
    restart: unless-stopped
    environment:
      OPENAI_API_KEY: ${OPENAI_API_KEY}
      OPENAI_MODEL: ${OPENAI_MODEL:-gpt-4.1}
      OLLAMA_MODEL: ${OLLAMA_MODEL:-command-r7b-arabic}
      GROUP_HOTWORDS: ${GROUP_HOTWORDS:-سوسي,يا سوسي}
    volumes:
      - ./auth_info:/app/auth_info
      - ./.env:/app/.env:ro

Run:

docker compose up -d --build
docker compose logs -f wa-llm-bot

Troubleshooting

• No reply: ensure message contains one of GROUP_HOTWORDS.
• OpenAI errors: verify key/model and internet access.
• Ollama: ensure service is running and model is pulled.
• .env must be in the directory you run node bot.js from.

Security

• Never commit .env or auth_info/.
• Rotate any keys that were shared publicly.