🚀 VeloraAI is a modern and flexible C# library designed to simplify local LLM integration. It allows developers to interact with quantized AI models directly from .NET Standard 2/.NET 8.0 applications — with a single line of code. Whether you're building chatbots, creative tools, or AI companions, VeloraAI is optimized for speed, reliability, and customization.

---

✨ Features

• ⚡ Quick-Start Model Loading — Choose from pre-integrated models or load your own via TestingMode.
• 🧠 Support for Multiple Models — CrystalThink, Qwen, Mistral, DeepSeek, Llama and more.
• 🔁 Event-driven Response System — React to TextGenerated, ResponseStarted, and ResponseEnded in real-time.
• 🔐 Customizable System Prompts — Use friendly or aggressive instruction styles (e.g., NoBSMode).
• 📦 Model Downloader — Automatically fetches models from Hugging Face if not already available.
• 📷 Experimental Vision Mode — Send image + prompt for visual reasoning (WIP).

---

⚠️ WARNING: VeloraAI does NOT work with the Legacy .NET Framework. You must use .NET Standard 2. or .NET 8.0. The following exception occurs when trying VeloraAI on .NET Framework:

Exception thrown: 'System.TypeLoadException' in LLamaSharp.dll
Authentication failed: Could not load type 'LLama.Native.NativeApi' from assembly 'LLamaSharp, Version=0.0.0.0, Culture=neutral, PublicKeyToken=null' because the method 'llama_backend_free' has no implementation (no RVA).

The fix for this problem is currently unfamiliar, however there is a chance that finding this fix would need LLamaSharp's source code integrated into VeloraAI's project and modified to make it work properly with .NET Framework. This is not promised. You are free to fork the project and find a fix for this issue and push it.

---

🧱 Built With

• LLamaSharp — Backbone inference engine.
• .NET 8.0 — Modern C# support.
• WinForms & Console — Sample UI and CLI clients included.

🎥 Raw Usage Example

Example.mp4

---

📂 Models Available

Model	Size	Strengths
Crystal_Think_V2_Q4	2.32 GB	🥇 Fast, tiny, math-heavy reasoning, Chain-of-Thought format
Qwen_V3_4B_Chat	2.70 GB	🥈 Fast general model with good code and reasoning
Mistral_7B_Chat	2.87 GB	🥉 Informative and precise longer-form chat
Llama_7B_Chat	3.07 GB	Reliable general conversations
DeepSeek_6B_Coder	3.07 GB	Code generation, math-only
DeepSeek_7B_Chat	5.28 GB	Slower general chat, strong context retention

---

🔧 Usage

1. Authenticate and Start Chatting

var result = await VeloraAI.AuthenticateAsync(VeloraAI.Models.Crystal_Think_V2_Q4);
if (result == VeloraAI.AuthState.Authenticated)
{
    await VeloraAI.AskAsync("What is the capital of France?");
}

2. Hook Into Events

VeloraAI.TextGenerated += (_, text) => Console.Write(text);
VeloraAI.ResponseStarted += (_, __) => Console.WriteLine("\n[VELORA is typing...]");
VeloraAI.ResponseEnded += (_, __) => Console.WriteLine("\n\n[Done]");

3. Use Custom Models

VeloraAI.TestingMode = true;
VeloraAI.TestingModelPath = @"C:\path\to\your_model.gguf";
await VeloraAI.AuthenticateAsync(VeloraAI.Models.TestingModel);

---

⚙️ Advanced Prompt Modes

Friendly Assistant (Default)

Follows a natural conversational tone with emojis and personality.

NoBS Mode

Blunt, hyper-logical response style with no emotional overhead or filler.

await VeloraAI.AuthenticateAsync(VeloraAI.Models.Crystal_Think_V2_Q4, NoBSMode: true);

---

📥 Model Auto-Download

Models are downloaded on first use to:

%APPDATA%/VeloraAI

Progress can be tracked using:

VeloraAI.CurrentDownloadProgress;

---

🔄 Reset History

VeloraAI.ResetHistory(); // or use custom system prompt

---

🎯 Custom Inference Parameters

You can fine-tune Velora's behavior using the following optional parameters in AskAsync:

Parameter	Description	Recommended for Speed
Temperature	Controls randomness (lower = more deterministic)	0.2 - 0.3
TopP	Nucleus sampling threshold	0.0 - 0.3
TopK	Limits token pool to top-K options	0 for fastest
RepeatPenalty	Penalizes repetition	1.05 - 1.2
MaxTokens	Maximum tokens to generate	80 - 128

await VeloraAI.AskAsync(
    userInput: "Summarize this paragraph.",
    temperature: 0.25f,
    TopP: 0.2f,
    TopK: 0,
    RepeatPenalty: 1.1f,
    maxTokens: 80
);

---

🛠️ Contributing

Pull requests are welcome! Please submit improvements, optimizations, or new model integrations.

---

📄 License

MIT

---

💬 Example Console Output

Authenticating model...
Authentication result: Authenticated

> What is 21 * 2?
[VELORA is typing...]

42

[Done]

---

🧪 Credits

• Developed by voidZiAD
• Powered by LLamaSharp, GGUF models, and the C#/.NET 8.0 ecosystem

---

🌐 Links

• LLamaSharp GitHub
• HuggingFace Models (Hosted by Velora)
• VeloraAI Console/WinForms Example

---

🧠 "VELORA" Personality

"I'm VELORA — not just another chatbot. I'm here to help you code, reason, and think clearer. No nonsense, just clarity."