HeroML is an AI Prompt Chain/Workflow interpreter for Apps built on https://hero.page

For more documentation, visit Hero docs, and learn about AI App Workflows

Download VSCode Syntax Highlighter Ext. here

1. HeroML Documentation
2. How to use HeroML Interpreter in your project
3. HeroML Extension for Visual Studio Code
4. Formatting Rules

HeroML (Hero Markup Language) is a novel approach for setting up multi-step workflows to interact with AI models like OpenAI's GPT-3 and GPT-4.

In this version, you can set up workflows in a series of steps, where each step is a prompt that the AI will respond to. The responses from previous steps can be used in the prompts for later steps.

This tutorial guides you through the process of installing and using the HeroML CLI tool.

To install the HeroML CLI tool, you need to have Node.js and npm (comes with Node.js) installed on your machine. If you don't have these, please install them first.

Then, run the following command in your terminal:

npm install -g heroml

This will install the HeroML CLI tool globally on your system, allowing you to use the hero command from any directory.

If you don't want to install it globally, you can run:

npm install heroml

and use it like:

npx hero run ./path/to/file.heroml

We'll be using hero run ... in this tutorial for simplicty.

Before using the HeroML CLI tool, you need to configure your OpenAI API key.

Create a heroconfig.json file in your home directory with the following content:

{
  "openaiApiKey": "your-openai-api-key"
}

Replace "your-openai-api-key" with your actual OpenAI API key.

Join our discord, will try to be as helpful as possible!

To run a HeroML script, use the run command followed by the path to your script:

hero run /path/to/your/script.heroml

You can provide initial variable values as command-line options. For example, if your script expects a variable named number_of_colors, you can provide its value as follows:

hero run --number_of_colors 4 /path/to/your/script.heroml

If you do not provide a value for a variable, the HeroML CLI tool will prompt you to enter it interactively.

You can specify the output directory and the filename of the output file using command-line options:

• The -o or --output-dir option allows you to specify the output directory. By default, it is ./outputs/.

hero run --output-dir /custom/output/directory /path/to/your/script.heroml

• The -f or --filename option allows you to specify the filename of the output file. By default, it is response_TIMESTAMP.json.

hero run --filename custom_filename.json /path/to/your/script.heroml

The CLI tool writes the output to a JSON file in the specified directory. It will print the path of the output file to the console:

Success! Output written to /custom/output/directory/custom_filename.json

Each workflow is a series of steps separated by the "->>>>" delimiter.

Example:

Prompt for step 1 

->>>> 

Prompt for step 2 

->>>> 

Prompt for step 3

In the first step, you can define dynamic variables in the format {{variable_name}}. These variables can be used in any step of the workflow.

Example:

In a blog post about {{topic}}, include the following points: {{point_1}}, {{point_2}}, {{point_3}}

You can reference the response of a previous step in your current step's prompt using the format {{step_X}}, where X is the step number.

Example:

What is the capital of {{country}}? 

->>>> 

Write a paragraph about the country: {{step_1}}

There's only one action available in v0.0.1, which is "Loop". This action can be used to iterate over each item in an array from the response of a previous step.

In the first line of the prompt, write ACTION: Loop, followed by ForEveryItemDoThis: on the next line, and then the prompt. If you want each item in the loop to yield its own item in a UI list, you can add ManyItems: TRUE under the ACTION line.

Example:

ACTION: Loop
ManyItems: TRUE
ForEveryItemDoThis: Write a short bio for {{step_1}}
{{step_1}}

In the example above, {{step_1}} must be an array for the loop to work.

You must specify the format you expect the response to be in. For example, if you want the response to be an array of strings, you can specify "Return the list as an array of strings".

Example:

List the top 5 countries in terms of population. Return the list as an array of strings, like ["China", "India", "USA", "Indonesia", "Pakistan"]

1. When using {{step_X}} in a loop action, make sure that step_X is indeed an array. If it's not, the loop action will fail.

2. You can only use variables defined in the first step ({{variable_name}}) throughout the entire workflow. New variables cannot be introduced in subsequent steps.

3. In the loop action, if ManyItems: TRUE is not specified, all responses will be added to one item.

Here's an example of a workflow:

For a blog post about {{topic}}, 
I want you to return an array of strings, of keywords 
(both short & long-tail) that are related to the following points:
{{point_1}}, {{point_2}}, {{point_3}}.

->>>>

ACTION: Loop
ManyItems: TRUE
ForEveryItemDoThis: Regarding my blog post about {{topic}}, write a paragraph about:
{{step_1}}

In this workflow, the first step is a prompt asking the AI to write a blog post about a certain topic, including three points. The second step is a loop action that asks the AI to write a paragraph about each point. The result of the second step will be an array of paragraphs, each about one point from the first step.

This guide will show you how to import and use the HeroML Interpreter in your own projects.

Firstly, you need to import the main function from the HeroML Interpreter. You can do this with the following line of code:

import { main } from './compilers/node/compile';

Ensure that the path in the import statement correctly points to the location of the compile.js file in your project.

Before you can use the main function, you need to set up the initial values for your dynamic variables. These are the variables that will be used in your HeroML script.

Here's an example of how to set up the initial values:

const initialValues = {
    blog_title: 'SEO Optimization: A short Guide',
    keywords: "SEO,",
    tone: 'informative',
    number_of_main_points: 1
};

In this example, we're setting up four variables: blog_title, keywords, tone, and number_of_main_points.

After setting up your initial values, you can now use the main function to interpret your HeroML script. You need to pass in two arguments to the main function:

1. The HeroML script as a string
2. The initial values you just set up

Here's an example of how to use the main function:

const heroml = `Your HeroML script goes here`;

async function run() {
    try {
        const finalEnvironment = await main(heroml, initialValues);
    } catch (error) {
        console.error('Error:', error);
    }
}

run();

In this example, we're wrapping the call to main inside an asynchronous function called run. This is because main returns a promise that resolves with the final environment — the output of the last step in the HeroML script. The run function is then called to initiate the interpretation process.

The response will be dependant on each step. If step_1 is a normal prompt, but step_2 is a loop, of 3 items, it will look like:

let finalEnvironment = {
  blog_title: 'SEO Optimization: A short Guide',
  keywords: 'SEO,',
  tone: 'informative',
  number_of_main_points: 1,
  step_1: '**Array of 3 objects**',
  step_2_1: '...',
  step_2_2: '...',
  step_2_3: '...',
}

To use your OpenAI API key, add:

OPENAI_API_KEY="YOUR_KEY_HERE"

to your .env file, and make sure not to push it up... like me.

To use the TypeScript compiler, run:

npx ts-node index.ts

I recommend making a virtual environment and sourcing it since the openai library has a lot of dependencies. If you don't understand what this means, don't worry about it; it's an optional step that doesn't make much of a difference.

cd compilers/py
python -m venv venv
source venv/bin/activate

pip install -r requirements.txt

python compile.py ./../../examples/Colors.heroml

If python doesn't work, try python3

This is all there is to using the HeroML interpreter! You can now build your own applications using HeroML. Remember to replace Your HeroML script goes here with your actual HeroML script.

Note: If an error occurs during the interpretation process, it will be caught in the catch block and logged to the console.

This extension provides support for HeroML in Visual Studio Code, including syntax highlighting.

Follow these steps to install the HeroML extension:

1. Download the heroml-0.0.1.vsix file from the heroml/ directory in this repository.

2. Open Visual Studio Code.

3. Click on the Extensions view icon on the Sidebar or press Ctrl+Shift+X (or Cmd+Shift+X on macOS) to open the Extensions view.

4. Click on the three dots ... at the top right corner of the Extensions view, next to the Clear Extensions Input icon.

5. From the dropdown, select Install from VSIX....

6. Navigate to the directory where you downloaded the heroml-0.0.1.vsix file, select it, and click Open.

7. The extension will be installed and you will get a notification in VS Code.

To start using the extension, simply open any .heroml file in VS Code. The syntax highlighting will be applied automatically.

Remember to reload VS Code after installing the extension to ensure it's properly activated.

Enjoy writing your HeroML workflows with syntax highlighting!

1. Only one action (Loop) is available in v0.0.1.

2. The responses from the AI are dependant on your own prompting. If you need a specific format (like an array of strings), you need to specify it in the prompt. I will make this easier in future versions.

3. The variables ({{variable_name}}) are static and must be defined in the first step. They cannot change throughout the workflow.

This documentation provides a basic overview of how to create and use workflows in HeroML v0.0.1. As the language evolves, new features and functionalities will be added. For the most up-to-date information, always refer to the latest documentation.

Different steps in a HeroML script are separated by ->>>>.

Steps do not need to be numbered.

You're not writing to the user, but rather to an AI model. When you include dynamic variables in the first step, the UI will already prompt the user to fill them in, and they will be replaced as if a natural part of the sentence.

You can only create new custom dynamic variables in the first step. In any following steps, the dynamic variables are either from the first step (re-used to maintain context) or step variables, like {{step_1}}.

You can only reference previous steps. For example, after step 1 finishes running, in step 2, you can reference the output from step 1, like so: "Write something about {{step_1}}".

There are two modes in HeroML:

The default mode, where no action is present, and it is a single prompt that yields a response.

A loop action, denoted by:

ACTION: Loop
ManyItems: TRUE
ForEveryItemDoThis: for every item in this array, write 15 words:
{{step_2}}

• Tutorial: Using HeroML Instruction Flags
  • Why use instruction flags?
  • How to use instruction flags?
  • Available Instructions
    • --return-json-array-strings
    • --return-json-array-numbers
    • --return-json-array-boolean
    • --return-json-array-dates
    • --return-json-array-objects
    • --return-json-array-objects-two-keys
    • --return-json-array-objects-three-keys
    • --return-json-array-objects-mixed
    • --return-json-array-objects-nested
    • --return-json-array-objects-arrays
    • --return-json-array-mixed

In HeroML, we often use AI prompts to direct our models in specific ways. We can ask the model to return responses that are JSON arrays, strings, numbers, or even more complex structures. To make this process more efficient, we have predefined certain instructions, called instruction flags, that you can insert into your prompts.

Instruction flags help make the prompts to the AI more explicit, ensuring that the AI model returns the desired type of response. This is particularly useful when you're dealing with loop actions and need your responses in a specific format. For example, if you're trying to generate a list of items with specific properties, it's much easier to guide the AI by using an instruction flag.

Using instruction flags is easy. You simply insert the flag into your prompt. For example:

Generate a list of random fruits --return-json-array-strings

When your prompt is parsed, the instruction flag --return-json-array-strings will be replaced with explicit instructions to the AI model, helping it understand exactly what you want.

Here's how different instruction flags work:

1. --return-json-array-strings: This flag instructs the AI to return a json array of strings as a plain text response. For example:

["item 1", "item 2", "item 3"]

2. --return-json-array-numbers: This flag tells the AI to return a json array of numbers. For example:

[1, 2, 3, 4]

3. --return-json-array-boolean: This flag directs the AI to return a json array of boolean values. For example:

[true, false, true]

...and so on. Each instruction flag guides the AI in producing a different type of output. Using these flags helps you get the results you want more consistently and with less manual tweaking of the prompts.

To see all the instructions, or to contribute even more, see utils/instructions.ts

Here are the available instructions and examples of how to use each one:

1. --return-json-array-strings

   • Usage: Generate a list of random fruits --return-json-array-strings
   • Expect: A list of strings

   [
       "Apple",
       "Banana",
       "Cherry"
   ]
   

2. --return-json-array-numbers

   • Usage: Generate a list of random numbers --return-json-array-numbers
   • Expect: A list of numbers

   [
       1, 
       2, 
       3, 
       4, 
       5
   ]
   

3. --return-json-array-boolean

   • Usage: Give me a list of boolean values --return-json-array-boolean
   • Expect: A list of boolean values

   [
       true, 
       false, 
       true
   ]
   

4. --return-json-array-dates

   • Usage: Give me a list of the next three dates --return-json-array-dates
   • Expect: A list of date strings

   [
       "2023-06-14", 
       "2023-06-15", 
       "2023-06-16"
   ]
   

5. --return-json-array-objects

   • Usage: Give me a list of users --return-json-array-objects
   • Expect: A list of objects with simple and consistent keys

   [
       {
           "name":"John", 
           "age":30
       }, 
       {
           "name":"Anna", 
           "age":27
       }
   ]
   

6. --return-json-array-objects-two-keys

   • Usage: Generate random objects with two keys --return-json-array-objects-two-keys
   • Expect: A list of objects each having exactly two keys

   [
       {
           "key1":"value1", 
           "key2":"value2"
       }, 
       {
           "key1":"value3", 
           "key2":"value4"
       }
   ]
   

7. --return-json-array-objects-three-keys

   • Usage: Generate random objects with three keys --return-json-array-objects-three-keys
   • Expect: A list of objects each having exactly three keys

   [
       {
           "key1":"value1", 
           "key2":"value2", 
           "key3":"value3"
       }, 
       {
           "key1":"value4", 
           "key2":"value5", 
           "key3":"value6"
       }
   ]
   

8. --return-json-array-objects-mixed

   • Usage: Generate random objects with mixed key-value types --return-json-array-objects-mixed
   • Expect: A list of objects with mixed key-value types

   [
       {
           "key1":"value1", 
           "key2":2
       }, 
       {
           "key1":"value3", 
           "key2":4, 
           "key3": true
       }
   ]
   

9. --return-json-array-objects-nested

   • Usage: Generate a list of users with their address --return-json-array-objects-nested
   • Expect: A list of objects, some of which contain nested objects

   [
       {
           "name":"John", 
           "age":30, 
           "address": {
               "city": "New York", 
               "country": "USA"
           }
       }, 
       {
           "name":"Anna", 
           "age":27, 
           "address": {
               "city": "Berlin", 
               "country": "Germany"
           }
       }
   ]
   

10. --return-json-array-objects-arrays

    • Usage: Generate a list of users with their hobbies --return-json-array-objects-arrays
    • Expect: A list of objects, some of which contain arrays

    [
        {
            "name":"John", 
            "age":30, 
            "hobbies": ["swimming", "reading"]
        }, 
        {
            "name":"Anna", 
            "age":27, 
            "hobbies": ["painting", "coding"]
        }
    ]
    

11. --return-json-array-mixed

    • Usage: Generate a mixed type list --return-json-array-mixed
    • Expect: A list with mixed value types (strings, numbers, objects, arrays)

    [
        "String", 
        42, 
        {
            "key": "value"
        }, 
        [
            "Item1", 
            "Item2"
        ]
    ]
    

Note: It's important to always test your AI prompts to make sure they return the expected results as different AI versions might behave differently.

Hello there! My name is Sam Chahine, and I'm the creator of HeroML. I built this project during the #AISF hackathon, an event that brings together innovators from all walks of life to collaborate and create something amazing in the field of AI. The hackathon was hosted by Founders, Inc., an organization based in San Francisco that loves AI! (Who doesn't?)

If you'd like to reach out, my personal email is sechahi at gmail dot com!

HeroML is an AI Prompt Chain/Workflow interpreter that can be a game changer for developers who work with AI models like OpenAI's GPT-3 and GPT-4. I plan to add support for more models from various platforms in the future. My hope is that HeroML can become a valuable tool that helps bridge the gap between AI and human creativity.

Want to build the python compiler? PULL REQUEESSSST

Update 17/07/2023 I want to thank @Shinji322 a thousand times for writing an amazing Python compiler for HeroML! You are amazing, check out his github profile down below 😊

• Python compiler: Shinji322.