Getting Started

MonadicSharp.AI extends MonadicSharp with five building blocks for working with LLMs — all returning Result<T> so they compose cleanly with the rest of your pipeline.

Install

dotnet add package MonadicSharp.AI

Requires: .NET 8.0+, MonadicSharp ≥ 1.5.

Quick Example

using MonadicSharp.AI;

// Call an LLM with retry and structured output validation
var result = await RetryResult<WeatherReport>
    .ExecuteAsync(
        maxAttempts: 3,
        action: () => llmClient.CompleteAsync(prompt),
        shouldRetry: err => err is AiError.RateLimit or AiError.Timeout)
    .BindAsync(raw => ValidatedResult<WeatherReport>.ParseAndValidateAsync(raw))
    .MapAsync(report => report with { FetchedAt = DateTime.UtcNow });

result.Match(
    onSuccess: report => Console.WriteLine($"Temp: {report.Temperature}°C"),
    onFailure: err  => logger.LogError("LLM call failed: {Err}", err));

AiError types

// All LLM failures are typed — no stringly-typed exceptions
AiError.RateLimit(retryAfter: TimeSpan.FromSeconds(30))
AiError.Timeout(elapsed: TimeSpan.FromSeconds(10))
AiError.TokenExhausted(tokensUsed: 4096)
AiError.ContentFiltered(reason: "policy violation")
AiError.InvalidResponse(raw: responseBody)

Streaming

await foreach (var chunk in StreamResult.FromAsync(llmClient.StreamAsync(prompt)))
{
    chunk.Match(
        onToken: token => Console.Write(token),
        onError: err   => logger.LogWarning("Stream interrupted: {Err}", err));
}

AgentResult — trace your pipeline

var traced = await AgentResult<string>
    .StartAsync("summarize", input)
    .StepAsync("validate",   ValidateInputAsync)
    .StepAsync("call-llm",   CallLlmAsync)
    .StepAsync("parse",      ParseOutputAsync);

Console.WriteLine($"Steps: {traced.Steps.Count}, Tokens: {traced.TotalTokens}");

Next steps

• AiError reference
• RetryResult reference
• AgentResult reference