Core Concepts

This section covers the fundamental concepts behind Prompty. Each page explains a key aspect of how Prompty processes .prompty files into LLM results.

The Basics

• The .prompty File Format — anatomy of a .prompty file: frontmatter, body, role markers, template syntax
• Pipeline Architecture — the four-stage pipeline: render → parse → execute → process

Configuration

• Connections — how Prompty authenticates with LLM providers (API keys, Microsoft Entra ID, registries)
• Conversation History (Threads) — multi-turn conversation via kind: thread inputs, nonce-based expansion, and best practices
• Tools — function, MCP, OpenAPI, and custom tool definitions

Advanced Features

• Tracing & Observability — pluggable tracer backends, @trace decorator, OpenTelemetry
• Streaming — real-time chunk delivery, PromptyStream, async iteration
• Structured Output — typed JSON responses via output schemas
• Agentic Concepts — turns, loop controls, guardrails, context compaction, steering, cancellation, and retries