Detester

Detester is a .NET library that enables you to write deterministic tests for AI-powered applications. It provides a fluent builder API for testing AI responses, ensuring consistency and reliability in your AI integrations.

Thank you for giving a ⭐ to this repo!

Features

• Fluent Builder API: Chain multiple prompts and assertions in a readable, intuitive way
• Any AI Provider Support: Works with any IChatClient implementation (OpenAI, Azure OpenAI, Ollama, etc.)
• Model Instructions: Set system messages to guide model behavior and responses
• Response Validation: Assert that AI responses contain expected keywords or text
• Function/Tool Call Verification: Verify that AI models call the correct functions with expected parameters
• JSON Response Validation: Deserialize and validate JSON responses from AI models with type-safe validation
• Method Chaining: Combine multiple prompts and assertions in a single test flow
• Extensible: Build on Microsoft.Extensions.AI abstractions for maximum flexibility

Installation

dotnet add package Detester

For OpenAI or Azure OpenAI support, also install:

dotnet add package Microsoft.Extensions.AI.OpenAI
dotnet add package Azure.AI.OpenAI

Quick Start

Check out the Wiki page for a detailed documentation

Using Azure OpenAI

using Azure.AI.OpenAI;
using Detester;
using Microsoft.Extensions.AI;
using System.ClientModel;

// Create Azure OpenAI client and wrap it as IChatClient
var azureClient = new AzureOpenAIClient(
    new Uri("https://your-resource.openai.azure.com"),
    new ApiKeyCredential("your-azure-api-key"));
var chatClient = azureClient.GetChatClient("gpt-4-deployment").AsIChatClient();

// Create a builder with the chat client
var builder = DetesterFactory.Create(chatClient);

// Or create a builder with ChatOptions (to configure tools, behavior, and settings)
var builder = DetesterFactory.Create(chatClient, new ChatOptions {/*setup*/});

// Execute a test
await builder
    .WithPrompt("Explain quantum computing in simple terms")
    .ShouldContainResponse("quantum")
    .AssertAsync();

Using Any IChatClient

Detester works with any IChatClient implementation from Microsoft.Extensions.AI:

using Detester;
using Microsoft.Extensions.AI;

// Use any IChatClient implementation (OpenAI, Azure OpenAI, Ollama, custom, etc.)
IChatClient chatClient = // your chat client implementation
var builder = DetesterFactory.Create(chatClient);

await builder
    .WithPrompt("Test prompt")
    .ShouldContainResponse("expected text")
    .AssertAsync();

Advanced Usage

Setting Model Instructions

Set custom instructions (system messages) to guide the model's behavior:

await new DetesterBuilder(chatClient)
    .WithInstruction("Answer concisely.")
    .WithPrompt("Summarize this long text you've provided.")
    .ShouldContainResponse("summary")
    .ShouldContainAll("short", "summary")
    .ShouldContainAny("tl;dr", "summary", "in short")
    .ShouldNotContainResponse("error")
    .ShouldNotContain("stack trace")
    .ShouldNotContainAnyResponse("SECRET_KEY", "password")
    .ShouldMatchRegex(@"^[A-Z].+")
    .ShouldBeEqualTo("Exact match", StringComparison.OrdinalIgnoreCase) // if you want strict equality
    .AssertAsync();

Multiple Prompts

Test conversational flows by chaining multiple prompts:

await builder
    .WithPrompt("Hello, I need help with coding")
    .WithPrompt("Can you explain what a variable is?")
    .ShouldContainResponse("variable")
    .AssertAsync();

OR Assertions

Use OrShouldContainResponse to create flexible response validation where at least one of the alternatives must match:

await builder
    .WithPrompt("What is the capital of France?")
    .ShouldContainResponse("capital")
    .OrShouldContainResponse("city")
    .OrShouldContainResponse("Paris")
    .AssertAsync();

In this example, the test passes if the response contains "capital" OR "city" OR "Paris". You can chain multiple OR conditions, and the test will pass if any one of them is found in the response.

JSON Response Validation

Detester supports validating JSON responses from AI models by deserializing them to C# types and optionally validating the deserialized objects. This is useful for testing structured outputs from language models.

Basic JSON Validation

Verify that the response can be deserialized to a specific type:

public class User
{
    public string? FirstName { get; set; }
    public string? LastName { get; set; }
    public int Age { get; set; }
    public DateTime JoinDate { get; set; }
}

await builder
    .WithPrompt("Who is the last user joined?")
    .ShouldHaveJsonOfType<User>(new JsonSerializerOptions { PropertyNameCaseInsensitive = true })
    .AssertAsync();

JSON Validation with Custom Validation

Add custom validation logic to verify the deserialized object:

await builder
    .WithPrompt("Who is the last user joined?")
    .ShouldHaveJsonOfType<User>(
        new JsonSerializerOptions { PropertyNameCaseInsensitive = true },
        user => user.Age > 30 && user.FirstName!.Contains("Jo"))
    .AssertAsync();

Complex JSON Validation

Combine multiple validations:

await builder
    .WithPrompt("Get user details")
    .ShouldContainResponse("Joe")  // Text assertion
    .ShouldHaveJsonOfType<User>(
        new JsonSerializerOptions { PropertyNameCaseInsensitive = true },
        user => user.Age > 18)  // JSON validation
    .ShouldHaveJsonOfType<User>(
        new JsonSerializerOptions { PropertyNameCaseInsensitive = true },
        user => user.LastName == "Doe")  // Additional JSON validation
    .AssertAsync();

Note:

• The JSON validation uses System.Text.Json for deserialization
• Deserialization exceptions are caught and wrapped in DetesterException with helpful error messages
• If validation fails, the test throws DetesterException with details about what went wrong
• For case-insensitive property name matching, use JsonSerializerOptions { PropertyNameCaseInsensitive = true }

Function/Tool Call Verification

Detester supports verifying that AI models call the correct functions/tools with expected parameters. This is useful for testing AI applications that use function calling capabilities.

Basic Function Call Verification

Verify that a specific function is called:

await builder
    .WithPrompt("What's the weather in Paris?")
    .ShouldCallFunction("get_weather")
    .AssertAsync();

Verify Function Parameters

Check that functions are called with the correct parameters:

await builder
    .WithPrompt("What's the weather in Paris in celsius?")
    .ShouldCallFunctionWithParameters("get_weather", 
        new Dictionary<string, object?> 
        { 
            { "location", "Paris" },
            { "units", "celsius" }
        })
    .AssertAsync();

Multiple Function Calls

Verify multiple function calls in a single response:

await builder
    .WithPrompt("Compare the weather in Paris and London")
    .ShouldCallFunction("get_weather")
    .ShouldCallFunction("get_weather")
    .AssertAsync();

Combined Verification

Combine function call verification with text response assertions:

await builder
    .WithPrompt("What's the capital of France?")
    .ShouldCallFunction("get_capital")
    .ShouldContainResponse("Paris")
    .AssertAsync();

For more detailed information and examples, see the Function Calling Guide.

Error Handling

Detester throws DetesterException when:

• No prompts are provided before execution
• Expected text is not found in the response
• None of the OR alternatives are found in the response
• Expected function call is not found
• JSON deserialization fails or validation fails

Detester throws InvalidOperationException when:

• OrShouldContainResponse is called without a prior assertion

Example:

try
{
    await builder
        .WithPrompt("What is AI?")
        .ShouldContainResponse("impossible text that won't appear")
        .AssertAsync();
}
catch (DetesterException ex)
{
    Console.WriteLine($"Test failed: {ex.Message}");
}

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

This project is licensed under the MIT License.

Acknowledgments

Built on top of Microsoft.Extensions.AI for seamless integration with AI services.