Ollama ph3dGPT FrontEnd and Ollama WhatsApp Bot

A powerful, agentic WhatsApp bot integrating local LLMs via Ollama, along with experimental browser-based Claude.ai features. It features a dual-architecture with a Node.js frontend (WhatsApp Web JS) and a Python Flask backend for executing agentic tools like web search, file parsing, and system command execution.

• Local AI via Ollama: Connects to your local Ollama models with a dynamically adjustable backend setup.
• The "ph3dGPT" Persona:
  • Configured with a sharp, no-nonsense personality with a bit of an edge.
  • Intelligent handling of Coding (gives functional code with brief comments).
  • High-tier Creative Writing (always respects rhyming constraints for raps and poems).
  • Math & Logic solving (shows full working and equations).
  • Real-world Date & Time awareness (handled automatically via tool calling).
  • Automatic language detection mimicking the user.
• Agentic Capabilities (ph3dMode): The bot can decide to use tools autonomously to fulfill requests:
  • web_search: Uses the Tavily API for highly accurate, AI-distilled real-time web searches.
  • scrape_url: Fetches and extracts text from external webpages for RAG (Retrieval-Augmented Generation).
• Advanced Utilities (Backend API):
  • Local command execution via the backend (/command endpoint running PowerShell).
  • File generation and reading (/read_file, /write_file).
  • File uploading with automatic PDF-to-Text extraction (/upload).
  • Hardware Monitoring: Built-in /stats endpoint to track local GPU and VRAM utilization.
• Claude.ai Integration (claudeMode): Forward prompts directly to the Claude.ai web interface via an automated Puppeteer browser.
• Dynamic Configuration: Change models, tweak "think" levels, and toggle features directly from WhatsApp chat commands.

• Node.js (v14 or higher)
• Python 3.8+
• Ollama installed and running. Pull your desired models (e.g., ollama run dolphin-mixtral:8x7b).
• Google Chrome (Required for WhatsApp Web auth and Claude Puppeteer navigation).
• Tavily API Key: Required for web search capabilities.

1. Clone the repository (or download the source files).
2. Setup Python Backend: Configure your Tavily API key inside server2.py if not already set. Install the required Python packages:

   pip install flask flask-cors werkzeug tavily-python PyMuPDF psutil GPUtil requests beautifulsoup4

3. Setup Node Frontend: Navigate to the wabot directory and install the Node dependencies:

   cd wabot
   npm install

The project includes convenient startup scripts to launch all services sequentially.

On Windows:

start_services.bat

On Linux/macOS:

chmod +x start_services.sh
./start_services.sh

What this does:

1. Verifies that Ollama is currently running on localhost:11434.
2. Starts server2.py (the Tools Backend Server on port 5000).
3. Starts bot.js (the WhatsApp Bot).

On the very first run, you will need to scan the QR code generated in the terminal with your WhatsApp app (Linked Devices) to authenticate.

Once the bot is running, you can send the following commands in WhatsApp to interact with and configure it:

• /mode - Display the current active model, think level, and active modes.
• /model <model_name> - Switch the current Ollama model dynamically.
• /ais - View a list of all installed Ollama models on your machine.
• /ph3dgpt - Toggle tool-calling mode ON/OFF. When ON, it enables autonomous Web Search and Scraping.
• /think <0|1|2> - Adjust the internal reasoning verbosity of the bot:
  • 0: Instant answers, minimal internal reasoning.
  • 1: Moderate internal reasoning.
  • 2: Full step-by-step deep thinking before answering.
• /claude - Toggle Claude Mode ON/OFF. Routes questions through the opened Claude.ai tab in Chrome.
• /reset - Clear the conversation history (up to 20 previous messages) for the current chat context.

Note on Usage: The bot is configured to natively reply in Direct Messages or specific whitelisted groups. If a prefix is configured (default: !ai), prefix your requests with !ai for the bot to respond.

In addition to WhatsApp, the project includes a local Web UI (ph3dGPT10.html) with a built-in command palette. Type / in the chat input to access these commands:

• /personality - Set a custom personality for this session.
• /reset - Reset personality to the default ph3dGPT persona.
• /model - Switch the active local Ollama model.
• /clear - Clear the current conversation to start fresh.
• /compact - Compress the conversation context to save VRAM and tokens.
• /search - Force a web search.
• /scrape - Scrape content from a URL directly.
• /temp <0.0-2.0> - Set the model's temperature (default 0.7). Lower = more focused/deterministic, higher = more creative/random.
• /run - Run a PowerShell command securely through the backend.
• /read - Read the contents of a local file.
• /write | - Write text content to a local file.

• start_services.bat - Windows launch script.
• start_services.sh - Linux/macOS launch script.
• server2.py - Flask API backend serving tools and utilities.
• agent.py - Core agent logic/testing utility.
• wabot/bot.js - Main WhatsApp Web JS application and prompting logic.
• wabot/package.json - Node dependencies.