ollama-dotnet 0.6.1

Ollama .NET Library

Typed C# client for integrating .NET 10+ applications with Ollama.

Prerequisites

• .NET 10 SDK
• Ollama installed and running
• A model pulled locally:

ollama pull gemma3

See Ollama Models for available models.

Install

dotnet add package O_Llama

Basic usage

using ollama_dotnet;

using var client = new OllamaClient();

var response = client.Chat(new ChatRequest
{
    Model = "gemma3",
    Messages =
    [
        new Message
        {
            Role = "user",
            Content = "Why is the sky blue?"
        }
    ]
});

Console.WriteLine(response.Message.Content);

The default local endpoint is http://127.0.0.1:11434.

Streaming responses

foreach (var part in client.ChatStream(new ChatRequest
{
    Model = "gemma3",
    Messages =
    [
        new Message
        {
            Role = "user",
            Content = "Why is the sky blue?"
        }
    ]
}))
{
    Console.Write(part.Message.Content);
}

Synchronous streaming methods:

• GenerateStream
• ChatStream
• PullStream
• PushStream
• CreateStream

Cloud models

Use cloud models through a local Ollama installation:

ollama signin
ollama pull gpt-oss:120b-cloud

foreach (var part in client.ChatStream(new ChatRequest
{
    Model = "gpt-oss:120b-cloud",
    Messages =
    [
        new Message
        {
            Role = "user",
            Content = "Why is the sky blue?"
        }
    ]
}))
{
    Console.Write(part.Message.Content);
}

See Ollama Cloud Models for current model names.

Cloud API

Create an API key at Ollama.com > Settings > Keys, then set it in the environment.

PowerShell:

$env:OLLAMA_API_KEY = "your_api_key"

Other shells:

export OLLAMA_API_KEY=your_api_key

The client reads OLLAMA_API_KEY automatically:

using var client = new OllamaClient(host: OllamaConstants.WebApiHost);

var response = client.Chat(new ChatRequest
{
    Model = "gpt-oss:120b",
    Messages =
    [
        new Message { Role = "user", Content = "Why is the sky blue?" }
    ]
});

An explicit header can be supplied instead:

using var client = new OllamaClient(
    host: OllamaConstants.WebApiHost,
    headers: new Dictionary<string, string>
    {
        ["Authorization"] = "Bearer your_api_key"
    });

Custom client

Pass an existing HttpClient when your application owns handlers, proxies, diagnostics, or dependency injection:

var httpClient = new HttpClient
{
    BaseAddress = new Uri("http://localhost:11434")
};

using var client = new OllamaClient(httpClient: httpClient);

var response = client.Generate(new GenerateRequest
{
    Model = "gemma3",
    Prompt = "Why is the sky blue?"
});

An injected HttpClient is not disposed by OllamaClient.

using var client = new OllamaClient(
    host: "http://localhost:11434",
    followRedirects: true,
    timeout: TimeSpan.FromMinutes(5),
    headers: new Dictionary<string, string>
    {
        ["x-application-name"] = "my-app"
    });

Async client

AsyncOllamaClient supports asynchronous calls and cancellation:

 using var client = new AsyncOllamaClient();

var response = await client.ChatAsync(
    new ChatRequest
    {
        Model = "gemma3",
        Messages =
        [
            new Message { Role = "user", Content = "Why is the sky blue?" }
        ]
    },
    cancellationToken);

Console.WriteLine(response.Message.Content);

Async streaming returns IAsyncEnumerable<T>:

await foreach (var part in client.ChatStreamAsync(
    new ChatRequest
    {
        Model = "gemma3",
        Messages =
        [
            new Message { Role = "user", Content = "Why is the sky blue?" }
        ]
    },
    cancellationToken))
{
    Console.Write(part.Message.Content);
}

Async streaming methods:

• GenerateStreamAsync
• ChatStreamAsync
• PullStreamAsync
• PushStreamAsync
• CreateStreamAsync

API

The client follows the Ollama REST API.

Chat

var response = client.Chat(new ChatRequest
{
    Model = "gemma3",
    Messages =
    [
        new Message { Role = "user", Content = "Why is the sky blue?" }
    ]
});

Generate

var response = client.Generate(new GenerateRequest
{
    Model = "gemma3",
    Prompt = "Why is the sky blue?"
});

List

var response = client.List();
foreach (var model in response.Models)
    Console.WriteLine(model.Name);

Show

var response = client.Show("gemma3");
Console.WriteLine(response.Details?.Family);

Create

var response = client.Create(new CreateRequest
{
    Model = "example",
    From = "gemma3",
    System = "You are a concise assistant."
});

Copy

var response = client.Copy("gemma3", "user/gemma3");
Console.WriteLine(response.Status);

Delete

var response = client.Delete("gemma3");
Console.WriteLine(response.Status);

Pull

var response = client.Pull(new PullRequest { Model = "gemma3" });

Push

var response = client.Push(new PushRequest { Model = "user/gemma3" });

Embed

var response = client.Embed(new EmbedRequest
{
    Model = "gemma3",
    Input = "The sky is blue because of Rayleigh scattering."
});

var vector = response.Embeddings[0];

Batch input is supported:

var response = client.Embed(new EmbedRequest
{
    Model = "gemma3",
    Input = new[]
    {
        "The sky is blue because of Rayleigh scattering.",
        "Grass is green because of chlorophyll."
    }
});

Legacy embeddings

var response = client.Embeddings(new EmbeddingsRequest
{
    Model = "gemma3",
    Prompt = "The sky is blue because of Rayleigh scattering."
});

Running models

var response = client.Ps();
foreach (var model in response.Models)
    Console.WriteLine($"{model.ModelName} - {model.SizeVram} bytes VRAM");

Web search and fetch

These operations require OLLAMA_API_KEY and use https://ollama.com:

var search = client.WebSearch("latest developments in space exploration");
foreach (var result in search.Results)
    Console.WriteLine($"{result.Title}: {result.Url}");

var page = client.WebFetch("https://ollama.com");
Console.WriteLine(page.Content);

Upload a blob

var digest = client.CreateBlob("model-data.bin");
Console.WriteLine(digest);

Images

Images can be supplied as Base64 data, bytes, or a file path:

var response = client.Generate(new GenerateRequest
{
    Model = "gemma3",
    Prompt = "Describe this image.",
    Images =
    [
        new Image("image-data-base64"),
        new Image(File.ReadAllBytes("image.jpg")),
        new Image("image.jpg", filePath: true)
    ]
});

Images are serialized as Base64 strings.

Request options

var response = client.Generate(new GenerateRequest
{
    Model = "gemma3",
    Prompt = "Write a short poem about the ocean.",
    Options = new Options
    {
        Temperature = 0.7,
        NumPredict = 128,
        TopP = 0.9,
        Stop = ["THE END"]
    }
});

Request properties use init accessors where possible, reducing accidental mutation after a request is configured.

Errors

Errors are raised for unsuccessful HTTP responses and errors reported in streaming responses:

try
{
    client.Chat(new ChatRequest
    {
        Model = "does-not-yet-exist",
        Messages =
        [
            new Message { Role = "user", Content = "Hello" }
        ]
    });
}
catch (ResponseException exception)
{
    Console.WriteLine($"Ollama error: {exception.Error}");
    Console.WriteLine($"Status code: {exception.StatusCode}");

    if (exception.StatusCode == 404)
        client.Pull(new PullRequest { Model = "does-not-yet-exist" });
}
catch (ConnectionException exception)
{
    Console.WriteLine(exception.Message);
}

ResponseException exposes the parsed error through Error and the HTTP status through StatusCode. ConnectionException indicates that Ollama could not be reached.