RMLLamaSharp 1.0.0

RMLLamaSharp

A lightweight .NET 8 client library for the Ollama REST API.
Run local LLMs (Llama 3, Mistral, Phi, Gemma and more) from any .NET application with a clean, async-first API.

Installation

dotnet add package RMLLamaSharp

Requirements

• Ollama running locally (default: http://localhost:11434)
• .NET 8 or later

Quick start

using LLamaSharp;

await using var client = new Client();

// Single prompt
string reply = await client.GenerateAsync("llama3", "Explain quantum entanglement in one sentence.");
Console.WriteLine(reply);

// Chat
var messages = new[]
{
    new ChatMessage("user", "What is the capital of France?")
};
string answer = await client.ChatAsync("llama3", messages);
Console.WriteLine(answer);

Features

Streaming

await foreach (var token in client.GenerateStreamAsync("llama3", "Tell me a short story."))
{
    Console.Write(token);
}

Multi-turn conversation

Use ChatSession to automatically maintain conversation history:

using var client = new Client();
var session = new ChatSession(client, "llama3", systemPrompt: "You are a helpful assistant.");

string r1 = await session.SendAsync("My name is Alice.");
string r2 = await session.SendAsync("What is my name?"); // remembers context

Streaming is also supported:

await foreach (var token in session.SendStreamAsync("Summarize our conversation."))
    Console.Write(token);

Generation options

Fine-tune sampling behaviour with GenerationOptions:

var options = new GenerationOptions(
    Temperature: 0.7f,
    TopP: 0.9f,
    Seed: 42,
    NumPredict: 200,
    Stop: [".", "\n"]
);

string reply = await client.GenerateAsync("llama3", "Once upon a time", options);

Embeddings

// Single input
List<float> vector = await client.GetEmbeddingsAsync("nomic-embed-text", "Hello world");

// Batch
List<List<float>> vectors = await client.GetEmbeddingsBatchAsync(
    "nomic-embed-text",
    ["First sentence.", "Second sentence."]);

Model management

// List all available models
List<string> models = await client.GetModelsAsync();

// Download a model
await client.PullModelAsync("mistral", status => Console.WriteLine(status));

// Delete a model
await client.DeleteModelAsync("mistral");

// Show metadata
ModelInfo info = await client.ShowModelAsync("llama3");
Console.WriteLine(info.Template);

Custom server URL

using var client = new Client("http://192.168.1.100:11434", timeout: TimeSpan.FromMinutes(5));

Error handling

All methods throw LLamaSharpException on server errors or unexpected responses, and ArgumentException for invalid inputs.

try
{
    string reply = await client.GenerateAsync("llama3", "Hello");
}
catch (LLamaSharpException ex)
{
    Console.WriteLine($"Ollama error: {ex.Message}");
}

License

MIT