OllamaSharp Documentation

For reasoning models (like deepseek-r1 or phi4-reasoning), you can enable thinking mode via the Think property. When enabled, internal reasoning is emitted through the OnThink event and is kept separate from the visible answer.

Think accepts a ThinkValue struct for budget control:

• ThinkValue.High: Maximum reasoning effort.
• ThinkValue.Medium: Balanced reasoning.
• ThinkValue.Low: Minimal reasoning.

var chat = new Chat(ollama) { Think = ThinkValue.High };

chat.OnThink += (_, thoughts) => Console.Write($"[thinking] {thoughts}");

await foreach (var token in chat.SendAsync("What is the square root of 144?"))
    Console.Write(token);

The Chat class is the recommended high-level interface for conversational use cases. It automatically manages message history (chat.Messages) so the model maintains context across multiple turns. Each call to SendAsync appends both the user's message and the assistant's response to the history.

var ollama = new OllamaApiClient("http://localhost:11434", "qwen3.5:35b-a3b");
var chat = new Chat(ollama);

while (true)
{
    Console.Write("You: ");
    var message = Console.ReadLine()!;

    Console.Write("Assistant: ");
    await foreach (var token in chat.SendAsync(message))
        Console.Write(token);

    Console.WriteLine();
}

OllamaSharp implements the IChatClient and IEmbeddingGenerator<string, Embedding<float>> interfaces from the Microsoft.Extensions.AI library. This allows you to use Ollama as a provider in applications designed for unified AI abstractions.

Note that while OllamaApiClient implements these interfaces, casting to IChatClient will hide Ollama-specific methods available in the native IOllamaApiClient interface.

// Requires 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);
}

The Chat class manages conversation state by automatically tracking message history (including roles and tool calls) across turns. This ensures the model maintains context throughout an interactive session. You can use SendAsync to send user messages and receive a stream of answer tokens.

// The chat object tracks history automatically 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);
}

The Chat class is the recommended abstraction for building conversational/multi-turn applications.

Key features:

• Automatic History Tracking: It automatically tracks the full message history, including roles and tool calls/results, across turns.
• Context Management: Because it tracks history, the model always has the full context of the conversation.
• Streaming: SendAsync returns an IAsyncEnumerable for real-time response streaming.

// 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);
}

To use Ollama cloud models, initialize the OllamaApiClient using an HttpClient that already contains your required authentication headers.

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

var ollama = new OllamaApiClient(client);

To use tools, define a class with methods decorated with the [OllamaTool] attribute (requires a source generator). Pass instances of these tool classes to SendAsync to enable function calling.

public class MyTools
{
    [OllamaTool]
    public static string GetWeather(string city) => $"Sunny and 22°C in {city}.";
}

var chat = new Chat(ollama);
await foreach (var token in chat.SendAsync("What's the weather in Berlin?", [new GetWeatherTool()]))
    Console.Write(token);