Implementation

Prompty has runtime implementations in multiple languages. Each follows the same pipeline architecture, uses generated model types from the in-repository schema, and is validated by shared conformance vectors for observable behavior.

Language	Status	Package
Python	✅ Stable	prompty on PyPI
TypeScript	✅ Stable	@prompty/core on npm
C#	✅ Stable	Prompty.Core, Prompty.OpenAI, Prompty.Foundry, Prompty.Anthropic on NuGet
Rust	✅ Stable	prompty, prompty-openai, prompty-foundry, prompty-anthropic on crates.io
Java	✅ Alpha	Java runtime packages
Go	✅ Alpha	Go module under runtime/go/prompty
Swift	✅ Alpha	Prompty, PromptyOpenAI, PromptyFoundry, PromptyAnthropic SwiftPM packages

Choosing a Runtime

All runtimes share the same .prompty file format and generated model contract — a prompt written for one runtime should load and prepare the same way in any other runtime. The differences are in installation, API style, host integration, and provider packages.

Canonical vectors cover load, render, parse, provider wire contracts, response processing, agent/tool behavior, memory behavior, memory context injection, and live provider acceptance when environment variables are present. Missing live provider credentials skip those checks; present credentials plus provider rejection fail them.

Python

• Most mature runtime with all features implemented
• Async support via _async variants of every function
• Entry-point-based plugin discovery
• Install: uv pip install "prompty[all]"

TypeScript

• Modular package architecture (@prompty/core + provider packages)
• Native async/await throughout
• Provider registration via side-effect imports
• Install: npm install @prompty/core @prompty/openai

C#

• Stable packages available on NuGet: Prompty.Core, Prompty.OpenAI, Prompty.Foundry, Prompty.Anthropic
• Static Pipeline API with async-first design
• Same four-stage render → parse → execute → process architecture
• Install: dotnet add package Prompty.Core

Rust

• Async-only runtime (Tokio) with zero-cost abstractions
• Modular crate architecture (prompty core + provider crates)
• Trait-based plugin system (Executor, Processor, Renderer, Parser)
• Full agent loop with events, cancellation, guardrails, and steering
• Install: cargo add prompty prompty-openai

Provider Packages

Provider	Python	TypeScript	C#	Rust	Swift
OpenAI	prompty[openai]	@prompty/openai	Prompty.OpenAI	prompty-openai	PromptyOpenAI
Microsoft Foundry	prompty[foundry]	@prompty/foundry	Prompty.Foundry	prompty-foundry	PromptyFoundry
Anthropic	prompty[anthropic]	@prompty/anthropic	Prompty.Anthropic	prompty-anthropic	PromptyAnthropic

Provider feature coverage is not identical in every runtime. For model/deployment discovery support, see Discover Available Models. For endpoint-backed verification, see Run Live Provider Tests.

Code Generation

Prompty’s runtime model types and schema reference are generated from the in-repository TypeSpec model via the Typra emitter. For details on how the generator works, see Code Generation.