Project Mello 🚀

Local-First AI Operating Layer & Desktop Assistant

Mello is a persistent local intelligence system designed to sit between you and your operating system. Unlike traditional chatbots, Mello focuses on persistent memory, desktop automation, and modular skill orchestration—all running 100% locally for maximum privacy and low latency.

---

🌟 Core Vision

Mello is not just a chat interface; it's a contextual automation engine. It remembers what you do, understands your file patterns, and learns your preferences to proactively help you manage your digital workspace.

---

🛠 Features (Phase 1–5 Complete)

🧠 Phase 1 & 2: Foundation & Skill Engine

• Intelligent Folder Watching: Real-time monitoring of your filesystem (e.g., Downloads).
• AI Classification: Uses local qwen3.5:0.8b to categorize new files (Documents, Images, etc.).
• Modular Skill Architecture: A "superpower" system where Mello can use specific tools:
  • mkdir / move: Manage your folder structure.
  • read_file: Analyze text, CSVs, and code.
  • open_app: Launch installed software.
  • screenshot: Take snapshots to "see" your desktop state.

🔍 Phase 3: Context Intelligence

• Semantic Memory: Integrated ChromaDB for vector-based search. Mello "remembers" events by meaning, not just keywords.
• Episodic Logging: Every action and detection is logged in a searchable SQLite database.
• Memory Compression: Automatic summarization of past events to keep the AI's context lean and fast.
• Preference Learning: Mello detects your habits (e.g., "User prefers snake_case for invoices") and stores them permanently.

⚡ Phase 4 & 5: Automation & Advanced AI

• Desktop Control: Open any Windows application or run PowerShell commands.
• Self-Healing Workflows: If a command fails, Mello automatically analyzes the error and tries a different approach.
• Windows-Smart: Built-in translation for Linux commands (touch, ls) to their Windows equivalents.

---

🖥 The Web Dashboard

Mello features a modern, dark-themed web interface (FastAPI + Vanilla CSS) accessible at http://localhost:8000.

• Intelligence Layer: Real-time streaming chat with Voice Input (Mic).

---

🚀 Quick Start

1. Prerequisites

• Ollama installed and running.
• Models downloaded: qwen3.5:0.8b and qwen3.5:4b.
• Python 3.10+ installed.

2. Setup

# Clone the repository
# Install dependencies
pip install -r requirements.txt

3. Configure

Edit the .env file to set your watch path:

WATCH_PATH=C:/Users/YourName/Downloads
THINK_ENABLED=true

4. Run

python main.py

Open your browser to http://localhost:8000 to start using Mello!

---

🔒 Security & Safety

Mello is built on the principle of Deterministic AI.

• Sandbox: Mello can only access folders and apps explicitly listed in security_policy.json.
• Command Filter: Dangerous shell keywords (like rm or format) are blocked by the Security Layer.
• Local Only: No data ever leaves your machine. Your files and memories stay private.

---

📂 Project Structure

mello/
├── main.py            # Entry point (Server + Watcher)
├── src/
│   ├── chat.py        # Streaming chat logic
│   ├── skills/        # Modular skill definitions
│   ├── memory.py      # SQLite & Vector DB management
│   ├── security.py    # Permission enforcement
│   └── server.py      # FastAPI REST API
├── static/            # Web Dashboard (HTML/CSS/JS)
└── .env               # Configuration

---

Created with ❤️ as a local-first AI infrastructure.