Sylin.Koan.AI.Prompt 0.8.0

Koan.AI.Prompt

Uri-inspired prompt primitives for Koan: parse, create, version, and store prompts with variable extraction, few-shot examples, output schema, and A/B testing support.

• Target framework: net10.0
• License: Apache-2.0
• Version: 0.6.3

Install

dotnet add package Sylin.Koan.AI.Prompt

Quick Start

// Build a prompt
var prompt = Prompt.Create()
    .System("You are a technical writer.")
    .Instruct("Summarise the following article in {{tone}} tone.")
    .Constrain("Maximum 200 words.")
    .OutputAs<SummaryOutput>()         // JSON schema from type
    .Example("long article...", "Short summary.")
    .Default("tone", "professional");

// Resolve with variables
var resolved = prompt.Resolve(new { tone = "casual" });

// Send to AI
var result = await Client.ChatAsync(resolved.SystemPrompt, resolved.UserMessage, ct);

Prompt API

// Fluent builder
Prompt.Create()
    .System(string systemPrompt)
    .Instruct(string template)             // Supports {{variable}} interpolation
    .Constrain(string constraint)
    .OutputAs<TOutput>()                   // Extract JSON schema from type
    .OutputAs(OutputSpec spec)             // Or provide explicit spec
    .Example(string input, string output)  // Few-shot example
    .Default(string variable, string value)// Default variable value
    .Meta(string key, string value)        // Arbitrary metadata

// Loading from storage
Prompt.Load("my-prompt")                   // Latest version
Prompt.Load("my-prompt", version: "v3")    // Specific version
Prompt.Load("my-prompt", PromptStrategy.AbTest(["v2", "v3"])) // A/B test

// Instance methods
prompt.Resolve(object variables)           // → string (resolved template)
prompt.UnresolvedVariables()               // → string[] (missing variables)
prompt.With(string variable, string value) // → new Prompt (immutable update)

Variable Interpolation

Templates use {{variable}} syntax:

var prompt = Prompt.Create()
    .Instruct("Translate the following to {{language}}: {{text}}");

var resolved = prompt.Resolve(new { language = "French", text = "Hello world" });
// → "Translate the following to French: Hello world"

Prompt Storage

PromptEntry is a standard Entity<PromptEntry> — prompts are versioned and stored in the configured data backend automatically:

// Store
await new PromptEntry { Name = "my-prompt", Version = "v1", Template = "..." }.Save();

// Retrieve
var entry = await Prompt.Load("my-prompt");

Loading Strategies

Reference

• Related: Koan.AI (pipeline facade), Koan.AI.Orchestration (chain composition), Koan.AI.Training (training datasets)