Getting Started with AIML

This guide will help you create your first AI agent using the AIML language.

Coming soon: AIML Language support for VS Code, Cursor, Vim, Emacs and more!

Basic Structure

At the end of the day, an AIML file is just a markdown (MDX to be precise) file with a more forgiving syntax parser and some intrinsic XML/JSX elements.

Here's a minimal example:

---
name: my-agent
model: account/fireworks/model/deepseek-v3
---
 
You are a helpful assistant that provides concise, accurate responses. Always begin your responses with "Agent:"
 
<if condition={({lastElement}) => !lastElement.output.startsWith("Agent:")}>
    <script language="javascript">
        return `Agent: ${lastElement.output}`;
    </script>
</if>

Here if the model doesn't start with "Agent:" we add it to the output using a script element.

AIML will handle conversion to/from streams making easy things easy and hard things possible!

Key Concepts

Frontmatter

The frontmatter section at the top of the file (between --- markers) defines default settings:

---
model: account/fireworks/model/deepseek-v3
---

System Prompt

The text immediately following the frontmatter serves as the system prompt for your agent:

You are a helpful assistant that provides concise, accurate responses.

Elements

Elements are XML/JSX-style tags that define your agent's behavior:

<llm>
    <instructions>
        Process the user's input carefully and provide a helpful response.
    </instructions>
    <prompt>
        {({userInput}) => userInput.message}
    </prompt>
</llm>

Comments

Use JSX-style comments to add notes that won't be sent to the LLM:

{/* This comment won't be included in the LLM context */}

Creating Your First Agent

Let's create a simple agent that thinks about a response before answering:

---
model: account/fireworks/model/deepseek-v3
---
 
{/* This is a system prompt and acts as the first step in the workflow */}
You are a helpful assistant that thinks carefully before responding.
 
 
{/* Then provide the actual response */}
<llm model="accounts/fireworks/models/qwen3-30b-a3b">
    <instructions>
        {({lastElement}) => `Based on the previous analysis (${lastElement.output}), 
        provide a clear and helpful response to the user.`}
    </instructions>
    <prompt>
        {({userInput}) => userInput.message}
    </prompt>
</llm>

Next Steps

Learn more about Language Syntax
Explore available Core Elements
Check out more Examples
Dive into Advanced Features

Getting Started

On this page