prompty

Prompty is a Rust runtime for the .prompty file format — a markdown-based asset class for LLM prompts. YAML frontmatter defines model configuration, connection details, tools, and schemas. The markdown body becomes templated instructions.

Quick Start

use prompty::{load, prepare, run, turn, register_defaults};

// Register built-in renderers and parsers
register_defaults();

// Load a .prompty file into a typed agent
let agent = load("path/to/prompt.prompty").await?;

// Render template + parse role markers → Vec<Message>
let messages = prepare(&agent, Some(&inputs)).await?;

// Execute LLM call + process response
let result = run(&agent, &messages).await?;

// Or do it all in one shot with agent loop support
let result = turn(&agent, Some(&inputs), None).await?;

Pipeline

The runtime provides 5 public functions:

Function	Description
`load()`	Parse a `.prompty` file → typed `Prompty` agent
`prepare()`	Render template + parse role markers → `Vec<Message>`
`run()`	Execute LLM call + process response (takes messages)
`invoke_agent()`	One-shot: load → prepare → execute → process
`turn()`	Conversation round with optional tool-calling agent loop

Providers

LLM providers are separate crates — register them before calling pipeline functions:

prompty-openai — OpenAI API
prompty-foundry — Azure OpenAI / Foundry
prompty-anthropic — Anthropic Claude

// Register OpenAI provider
prompty_openai::register();

Features

otel — Enables OpenTelemetry tracing backend

License