🐾 Mochi: Autonomous Local AI Desktop Agent

An asynchronous, event-driven AI desktop assistant powered by local LLMs and Live2D WebSocket integration.

Unlike standard text-based chatbots, this project bridges a local language model (Llama 3.2) with a graphical Live2D frontend (VTube Studio) to create an autonomous, emotive, and system-aware virtual companion.

🚀 Architecture & Tech Stack

• LLM Backend: Ollama running llama3.2 locally for zero-latency, privacy-first inference.
• Middleware: Python asyncio and WebSockets for non-blocking execution.
• Frontend: VTube Studio API for rendering Live2D models and triggering state-based animations via JSON payloads.

🧠 How It Works

1. Prompt & Generation: The user interacts via the terminal. The Python script queries the local Llama model, which is contextually prompted to act as "Mochi," an energetic chibi mascot.
2. Context Parsing: The LLM injects specific emotion tags (e.g., [LOVE], [STAR], [CRY]) into its response based on the conversation's context.
3. Event-Driven Execution: The Python script strips the tags from the user-facing text and asynchronously sends a Hotkey Trigger request via WebSockets to VTube Studio on Port 8001.
4. Seamless Rendering: The Live2D avatar reacts instantly on the user's desktop without requiring window focus or physical keyboard macros.

🛠️ Prerequisites

• Python 3.10+
• Ollama installed with llama3.2 pulled locally.
• VTube Studio installed via Steam (with API started on Port 8001).

📦 Installation

1. Clone the repository:

   git clone [https://github.com/YourUsername/mochi-ai-agent.git](https://github.com/YourUsername/mochi-ai-agent.git)
   cd mochi-ai-agent