OllamaSharp

OllamaSharp 🦙

OllamaSharp provides .NET bindings for the Ollama API, simplifying interactions with Ollama both locally and remotely.

🏆 Recommended by Microsoft

Features

• Ease of use: Interact with Ollama in just a few lines of code.
• Reliability: Powering Microsoft Semantic Kernel, .NET Aspire and Microsoft.Extensions.AI
• API coverage: Covers every single Ollama API endpoint, including chats, embeddings, listing models, pulling and creating new models, and more.
• Real-time streaming: Stream responses directly to your application.
• Progress reporting: Real-time progress feedback on tasks like model pulling.
• Tools engine: Sophisticated tool support with source generators.
• Multi modality: Support for vision models.
• Native AOT support: Opt-in support for Native AOT for improved performance.

📖 Documentation

Topic	Description
Getting started	Installation, client setup and first steps
Chat and Generate	Chat class, GenerateAsync, streaming, images, structured output, thinking models
Model Management	List, pull, push, copy, delete, show, embed
Tool support	Function calling with source generators
Advanced Configuration	Configuration options that go beyond the basics
Native AOT support	Ahead-of-Time compilation guidance
API reference	Full auto-generated API docs

Usage

OllamaSharp wraps each Ollama API endpoint in awaitable methods that fully support response streaming.

The following list shows a few simple code examples.

ℹ Try our full featured demo application that's included in this repository

Initializing

// set up the client
var uri = new Uri("http://localhost:11434");
var ollama = new OllamaApiClient(uri);

// select a model which should be used for further operations
ollama.SelectedModel = "qwen3.5:35b-a3b";

Native AOT Support

For .NET Native AOT scenarios, create a custom JsonSerializerContext with your types and pass it into the constructor.

[JsonSerializable(typeof(MyCustomType))]
public partial class MyJsonContext : JsonSerializerContext { }

// Use the static factory method for NativeAOT
var ollama = new OllamaApiClient(uri, "qwen3.5:35b-a3b", MyJsonContext.Default);

See the Native AOT documentation for detailed guidance.

Listing all models that are available locally

var models = await ollama.ListLocalModelsAsync();

Pulling a model and reporting progress

await foreach (var status in ollama.PullModelAsync("qwen3.5:35b-a3b"))
    Console.WriteLine($"{status.Percent}% {status.Status}");

Generating a completion directly into the console

GenerateAsync maps to the /api/generate endpoint and is ideal for single-turn, context-free completions.

await foreach (var stream in ollama.GenerateAsync("How are you today?"))
    Console.Write(stream.Response);

Building interactive chats

The Chat class is the recommended way to build conversational applications. It automatically tracks the full message history (including tool calls and their results) across turns so the model always has full context.

// messages including their roles and tool calls will automatically be tracked within the chat object
// and are accessible via the Messages property

var chat = new Chat(ollama);

while (true)
{
    var message = Console.ReadLine();
    await foreach (var answerToken in chat.SendAsync(message))
        Console.Write(answerToken);
}

You can also set a system prompt, send images for vision models, request structured JSON output, and enable thinking mode for reasoning models. See the Chat and Generate documentation for the full guide.

Answering typed questions with System One

SystemOneAsync maps to the /v1/systemone endpoint (Ollama v0.35.0 or later). Instead of generating text, a System One model like nimble answers choice, yes/no and score questions about a given state and returns probabilities.

var response = await ollama.SystemOneAsync(new SystemOneRequest
{
    Model = "nimble",
    State = "Our checkout has returned 500 errors since 9am.",
    Questions = new()
    {
        ["label"] = new SystemOneChoiceQuestion
        {
            Instructions = "Which label fits this ticket?",
            Criteria = new() { ["billing"] = "Payments and refunds", ["bug"] = "Software errors", ["account"] = "Login and account access" }
        },
        ["refund"] = new SystemOneNoulQuestion { Instructions = "Is the customer requesting a refund?" },
        ["urgency"] = new SystemOneScoreQuestion
        {
            Instructions = "How urgently does this ticket need a response?",
            Criteria = ["Routine", "Soon", "Immediate"]
        }
    }
});

var label = (SystemOneChoiceAnswer)response.Answers["label"];     // label.Choice == "bug"
var refund = (SystemOneNoulAnswer)response.Answers["refund"];     // refund.Noul == probability of "yes"
var urgency = (SystemOneScoreAnswer)response.Answers["urgency"];  // urgency.Score from 0 to 2

See the Ollama decision guide for more details.

Usage with Microsoft.Extensions.AI

Microsoft built an abstraction library to streamline the usage of different AI providers. This is a really interesting concept if you plan to build apps that might use different providers, like ChatGPT, Claude and local models with Ollama.

I encourage you to read their accouncement Introducing Microsoft.Extensions.AI Preview – Unified AI Building Blocks for .NET.

OllamaSharp is the first full implementation of their IChatClient and IEmbeddingGenerator that makes it possible to use Ollama just like every other chat provider.

To do this, simply use the OllamaApiClient as IChatClient instead of IOllamaApiClient.

// install package Microsoft.Extensions.AI.Abstractions

private static IChatClient CreateChatClient(Arguments arguments)
{
  if (arguments.Provider.Equals("ollama", StringComparison.OrdinalIgnoreCase))
    return new OllamaApiClient(arguments.Uri, arguments.Model);
  else
    return new OpenAIChatClient(new OpenAI.OpenAIClient(arguments.ApiKey), arguments.Model); // ChatGPT or compatible
}

The OllamaApiClient implements both interfaces from Microsoft.Extensions.AI, you just need to cast it accordingly:

• IChatClient for model inference
• IEmbeddingGenerator<string, Embedding<float>> for embedding generation

Cloud models aka Ollama Turbo

OllamaSharp can be used with Ollama cloud models as well. Use the constructor that takes an HttpClient and set it up to send the api key as default request header.

var client = new HttpClient();
client.BaseAddress = new Uri("http://localhost:11434");
client.DefaultRequestHeaders.Add(/* your api key here */);
var ollama = new OllamaApiClient(client);

OllamaSharp vs. Microsoft.Extensions.AI vs. Semantic Kernel

It can be confusing which library to use with AI in C#. The following paragraph should help you decide which library to start with.

Prefer OllamaSharp if ...

• you plan to use Ollama models only
• you want to use the native Ollama API, not only chats and embeddings but model management, usage information and more

Prefer Microsoft.Extensions.AI if ...

• you only need chat and embedding functionality
• you want to be able to use different providers like Ollama, OpenAI, Hugging Face, etc.

Prefer Semantic Kernel if ...

• you need the highest flexibility with different providers, plugins, middlewares, caching, memory and more
• you need advanced prompt techniques like variable substitution and templating
• you want to build agentic systems

No matter which one you choose, OllamaSharp should always be the bridge to Ollama behind the scenes as recommended by Microsoft (1) (2) (3).

Thanks

I would like to thank all the contributors who take the time to improve OllamaSharp. First and foremost mili-tan, who always keeps OllamaSharp in sync with the Ollama API.

The icon and name were reused from the amazing Ollama project.