🦾 War-Machine AI

A high-performance, local AI character integration using Node.js (ES Modules) and Ollama. This project is specifically tuned for the Intel i5-1235U with 16GB RAM, focusing on bypassing Windows DNS latency and providing real-time streaming.

---

🚀 Features

• Custom Personality: A "War-Machine" persona defined via a dedicated Modelfile.
• Zero-Lag Networking: Direct IPv4 (127.0.0.1) binding to skip the 2-second Windows localhost lookup delay.
• Real-Time Streaming: Uses HTTP chunked encoding to deliver words the millisecond they are generated.
• RAM Optimized: Includes keep_alive logic to ensure the model stays in your 16GB RAM.

---

🛠️ Hardware Context

• Processor: 12th Gen Intel Core i5-1235U (2 P-Cores, 8 E-Cores).
• RAM: 16GB DDR4/DDR5.
• Performance Note: On this CPU, expect roughly 5-8 tokens per second. Streaming is enabled to ensure "First Token" delivery in < 1.5s.

---

📥 Setup Instructions

1. Install Ollama

Download the engine at ollama.com. Ensure the Ollama icon is visible in your system tray.

2. Project Initialization

# Initialize and install dependencies
npm init -y
npm install express ollama
npm pkg set type="module"

3. Build the Character

Ensure you have a file named Modelfile in your root directory. Then, register the character:

# Run via terminal
ollama create war-machine -f Modelfile

4. Ignite the Server

node server.js

---

📡 API Reference

POST /ask

The primary endpoint for interacting with War-Machine.

Headers: Content-Type: application/json

Request Body:

{
  "prompt": "War-Machine, what is your current status?"
}

Testing via PowerShell (cURL):

curl.exe -X POST [http://127.0.0.1:3000/ask](http://127.0.0.1:3000/ask) `
-H "Content-Type: application/json" `
-d '{"prompt": "Give me a status report on the CPU cores."}'

---

⚙️ Key Optimizations Applied

1. Direct IP: Changed localhost to 127.0.0.1 in the client to stop DNS lag.
2. Streaming Loop: Implemented for await (const part of stream) to pipe output directly.
3. Keep-Alive: Added keep_alive: '30m' to prevent the i5 from reloading from SSD.

---

📝 Example Response

Below is a live look at War-Machine in action. Notice the low-latency streaming and the character-driven persona.

Note: On an i5-1235U, the first token is delivered in ~1.2s, with a full response completed in under 9s thanks to streaming optimizations.

---

📜 License

MIT - Created for the War-Machine Project. 🤖🦾