Microsoft’s POML Brings HTML-Style Structure to AI Prompt Engineering

Microsoft has introduced Prompt Orchestration Markup Language (POML), a new open-source framework that streamlines the design and coordination of AI prompts for developers. POML is built for multi-agent systems and it enables more structured, scalable, and adaptable AI workflows.

Prompt Orchestration Markup Language (POML) is an open-source framework developed to simplify the design and coordination of complex interactions between AI agents and tools. It provides a structured way to define how different components (such as language models, external functions, and memory systems) work together within a prompt-driven workflow. POML utilizes a markup-style syntax to enable developers to build scalable, modular, and maintainable AI systems that extend beyond single-turn conversations, supporting dynamic and multi-step reasoning across agents.

Key features that set POML apart

1. Structured prompting syntax

POML adopts a markup format similar to HTML (using tags like and ) to organize prompt components. This structure promotes clarity, modularity, and ease of reuse across different workflows.

2. Advanced data handling

It supports embedding various data types (such as documents, tables, and images) through dedicated tags. This allows prompts to reference external sources directly, with flexible formatting options to suit different contexts.

3. Decoupled presentation styling

POML introduces a styling system similar to CSS, which allows developers to adjust presentation aspects like verbosity or formatting without changing the core prompt logic.

4. Built-in templating capabilities

The language includes a native templating engine that supports variables, loops, conditionals, and variable definitions. This makes it easier to generate dynamic, data-driven prompts that respond to changing inputs.

5. Developer-friendly tools

POML offers extensions for Visual Studio Code that provide features like syntax highlighting, context-aware auto-completion, inline diagnostics, and live previews to streamline development. It also includes SDKs for Python and Node.js that allow seamless integration with existing applications and LLM frameworks.

Real-world applications of POML

POML is well-suited for applications that require dynamic and adaptable prompt generation, such as creating personalized content, running A/B tests on different prompt formats, or generating multi-modal instructions. Its ability to separate content from presentation allows developers to easily switch styles or formats without rewriting core logic. This makes POML highly compatible with various LLMs and robust across different use cases like automated reporting, instructional design, and interactive AI systems.

POML is an open-source project released under the MIT License and is freely available on the official GitHub repository. Developers can get started by installing the Visual Studio Code extension and using the official Node.js or Python SDKs to integrate POML into their applications.