Prompty.Core 2.0.0-alpha.10

Prompty.Core

Prompty is an asset class and format for LLM prompts designed to enhance observability, understandability, and portability for developers. The .prompty file format combines YAML frontmatter (model config, inputs, tools) with a markdown body (your prompt template).

Prompty.Core is the foundation package — it loads .prompty files, renders templates, parses messages, and orchestrates the execution pipeline. Provider packages (OpenAI, Foundry, Anthropic) plug in via the invoker registry.

Installation

dotnet add package Prompty.Core

You'll also need at least one provider package:

dotnet add package Prompty.OpenAI      # OpenAI
dotnet add package Prompty.Foundry     # Azure OpenAI / Microsoft Foundry
dotnet add package Prompty.Anthropic   # Anthropic Claude

Quick Start

1. Create a .prompty file (chat.prompty):

---
name: basic-chat
model:
  id: gpt-4o-mini
  provider: openai
  connection:
    kind: key
    endpoint: https://api.openai.com/v1
    apiKey: ${env:OPENAI_API_KEY}
inputSchema:
  properties:
    - name: question
      kind: string
---
system:
You are a helpful assistant.

user:
{{question}}

2. Load and run it in C#:

using Prompty.Core;
using Prompty.OpenAI;

// One-time setup — registers renderers, parser, and providers
new PromptyBuilder()
    .AddOpenAI();

// Load, prepare messages, and execute
var agent = PromptyLoader.Load("chat.prompty");
var result = await Pipeline.InvokeAsync(agent, new Dictionary<string, object>
{
    ["question"] = "What is the meaning of life?"
});

Console.WriteLine(result);

Packages

Features

• .prompty file format — YAML frontmatter + Jinja2/Mustache template body
• Pipeline architecture — Render → Parse → Execute → Process
• Structured output — outputSchema converts to response_format automatically
• Streaming — PromptyStream (IAsyncEnumerable<object>) with tracing
• Agent mode — Automatic tool-call loop with configurable max iterations
• Tracing — Pluggable backends: console, .tracy files, OpenTelemetry
• Connection registry — Pre-register SDK clients for reuse across prompts
• Spec compliance — Passes Prompty spec test vectors

Core API

Loading

// Sync
var agent = PromptyLoader.Load("chat.prompty");

// Async
var agent = await PromptyLoader.LoadAsync("chat.prompty");

Pipeline

// Full invoke (load → render → parse → execute → process)
var result = await Pipeline.InvokeAsync("chat.prompty", inputs);

// Step-by-step control
var validated = Pipeline.ValidateInputs(agent, inputs);
var rendered  = await Pipeline.RenderAsync(agent, validated);
var messages  = await Pipeline.ParseAsync(agent, rendered);
var response  = await Pipeline.ExecuteAsync(agent, messages);
var result    = await Pipeline.ProcessAsync(agent, response);

// Or use prepare + run
var messages = await Pipeline.PrepareAsync(agent, inputs);  // render + parse
var result   = await Pipeline.RunAsync(agent, messages);     // execute + process

Agent Mode (Tool Calling)

var tools = new Dictionary<string, Func<string, Task<string>>>
{
    ["get_weather"] = async (args) => "72°F and sunny"
};

var result = await Pipeline.TurnAsync(
    agent, inputs, tools, maxIterations: 10);

Tracing

// Register tracers before invoking
ConsoleTracer.Register();         // Print spans to console
PromptyTracer.Register();         // Write .tracy JSON files
OTelTracer.Register();            // OpenTelemetry spans

// Wrap custom code in a trace span
var result = await Trace.TraceAsync("my-operation", async () =>
{
    // your code here
    return value;
});

Connection Registry

// Pre-register an SDK client
ConnectionRegistry.Register("my-openai", openAIClient);

// Reference it in .prompty frontmatter:
// connection:
//   kind: reference
//   name: my-openai

Documentation

Visit prompty.ai for full documentation, guides, and the language specification.

License

MIT