InstructSharp 1.0.1

InstructSharp

InstructSharp is a high‑performance, provider‑agnostic .NET SDK that turns large‑language‑model requests into one‑line calls and structured JSON responses. ✨

Seamlessly swap between OpenAI ChatGPT, Anthropic Claude, Google Gemini, X.AI Grok, DeepSeek, or Meta LLaMA without rewriting a single line of business logic.

TL;DR – Install the package, define a POCO, call QueryAsync<T>(), get strongly‑typed results. ✅

📑 Table of Contents

1. Key Features
2. Quick Install
3. Hello, World
4. Provider Matrix
5. Advanced Usage
6. Performance Notes
7. Roadmap
8. Contributing
9. License

Key Features

Quick Install

# Package Manager
Install-Package InstructSharp

# .NET CLI
dotnet add package InstructSharp

⚡ Tip – add --prerelease to grab nightly builds from CI.

Hello, World

using InstructSharp.Clients.ChatGPT;
using InstructSharp.Core;

class QuestionAnswer
{
    public string Question { get; set; }
    public string Answer   { get; set; }
}

var chat = new ChatGPTClient("YOUR_OPENAI_API_KEY");

var req = new ChatGPTRequest
{
    Model       = ChatGPTModels.GPT4oMini,
    Instruction = "Talk like a pirate.",
    Input       = "What is 2 + 2?"
};

var res = await chat.QueryAsync<QuestionAnswer>(req);
Console.WriteLine($"A• {res.Result.Answer}");

Want raw text? Simply use string instead of a POCO:

var text = await chat.QueryAsync<string>(req);

Provider Matrix

Streaming support is on the roadmap – follow the issues to vote.

Advanced Usage

🔒 Secure Configuration

var http = new HttpClient {
    Timeout = TimeSpan.FromSeconds(15)
};
var chat = new ChatGPTClient(Environment.GetEnvironmentVariable("OPENAI_KEY"), http);

• HttpClient injection lets you share retry policies, logging handlers, or proxies.
• Add standard headers globally via DefaultRequestHeaders.

🎛️ Tuning Parameters

Every request type exposes vendor‑specific knobs such as Temperature, TopP, MaxTokens, etc. Set only what you need – defaults are sane.

🗂️ Error Handling

try
{
    var res = await chat.QueryAsync<MyType>(req);
}
catch(HttpRequestException ex) when (ex.StatusCode == HttpStatusCode.TooManyRequests)
{
    // handle 429 rate limit
}

📉 Token Usage

LLMResponse<T>.Usage returns prompt, response & total tokens. Use it for cost tracking or throttling.

Performance Notes

• No dynamic – all JSON is parsed with System.Text.Json. Fast.
• Schema cache – Generated JSON schemas are cached per‑type to avoid regeneration.
• One HTTP round‑trip – no second prompt to "format JSON"; the schema is sent in the first call.

Benchmarks live under /benchmark – PRs welcome! 🏎️💨

Roadmap

• 🔄 Streaming completions
• 🧩 Function & tool call helpers
• 🏗️ Automatic retries / exponential back‑off
• 📝 DocFX site with full API reference
• 🏆 Benchmarks vs raw vendor SDKs

Have a feature in mind? Open an issue or send a PR!

Contributing

1. Fork the repo
2. git clone & dotnet build – tests should pass
3. Create your branch: git checkout -b feature/my-awesome
4. Commit & push, then open a PR

Dev Environment

• .NET 8 SDK
• Optional: direnv / dotenv for API keys
• EditorConfig + Roslyn analyzers enforce style