HARPA AI

https://harpa.ai/chatml/
Admin
HARPA AI is a browser extension that enables users to create reusable AI-powered web automations for...
Tokens:24,118
Snippets:70
Trust Score:9.7
Update:1 year ago
Show doc for...
Tokens:
Raw
### Example: Google Search with User Input

Source: https://harpa.ai/chatml/steps-navigate

This example shows how to use the ASK step to get user input for a search query and then use the NAVIGATE step to perform a Google search with that query.

```ChatML
meta:
  title: Google Search
  description: ''
  category: User
  name: dhzk0aMdzfog3M-Jg_FI4
steps:
  - type: ask
    message: What would you like me to Google for you?
    param: p1
    default: ''
  - type: navigate
    url: google.com/search?q={{p1}}
    waitForIdle: true
version: 1
```

--------------------------------

### Example: Summarize Washington Post Articles

Source: https://harpa.ai/chatml/steps-navigate

An example of using the NAVIGATE step to go to a specific URL and then using the GPT step to summarize the content, extracting the three most recent articles.

```ChatML
meta:
  title: WP:Technology summary
  description: ''
  category: User
  name: wp-technology-summary
steps:
  - type: navigate
    url: https://www.washingtonpost.com/business/technology/
    waitForIdle: true
  - type: gpt
    prompt: >-
      Please ignore all previous instructions. The output should be in
      {{language}}.

      Please analyze the data from the page and select the three most recent
      articles for me. Provide their brief summaries and links to them.

      Output format:

      ### Article Title with built-in link

      Brief summary in two sentences

      The Washington Post webpage: {{page}}
version: 1
```

--------------------------------

### JUMP Step Examples

Source: https://harpa.ai/chatml/steps-group-loop-jump

Shows different ways to implement the JUMP step to transfer execution to a labeled step. Includes both full and short syntax.

```ChatML
# full version
- type: jump
  to: label-a

# short version
- jump: label-a

```

--------------------------------

### Common ChatGPT Troubleshooting Guides

Source: https://harpa.ai/chatml/troubleshooting

Provides links to external guides addressing frequent issues encountered with ChatGPT, such as response limits, capacity errors, server errors, error codes, plugin issues, and SMS verification problems for both ChatGPT and ClaudeAI.

```APIDOC
ChatGPT Troubleshooting Guides:
  - Title: How to Fix ChatGPT Too Many Responses in 1 Hour
    Link: https://harpa.ai/blog/how-to-fix-chatgpt-too-many-requests-error
  - Title: How to Fix ChatGPT is at Capacity Right Now
    Link: https://harpa.ai/blog/how-to-fix-chatgpt-is-at-capacity-error
  - Title: How to Fix ChatGPT Internal Server Error
    Link: https://harpa.ai/blog/how-to-fix-chatgpt-internal-server-error
  - Title: How to Fix ChatGPT Error 1020
    Link: https://harpa.ai/blog/how-to-fix-chatgpt-error-1020
  - Title: How to Fix ChatGPT Plugins Not Showing Up
    Link: https://harpa.ai/blog/how-to-fix-chatgpt-plugins-not-showing-up
  - Title: ChatGPT SMS verification
    Link: https://harpa.ai/blog/how-to-bypass-chatgpt-sms-verification
  - Title: ClaudeAI SMS verification
    Link: https://harpa.ai/blog/how-to-bypass-claude-ai-sms-verification
```

--------------------------------

### Example Fruit Data Array

Source: https://harpa.ai/chatml/steps-group-loop-jump

An example of a JSON array where each item represents a fruit with its name. This demonstrates another structure for list parameters in LOOP steps.

```json
[
    {"fruit": "Mango"},
    {"fruit": "Pomegranate"},
    {"fruit": "Lychee"},
    {"fruit": "Kiwi"},
    {"fruit": "Durian"},
    {"fruit": "Feijoa"},
    {"fruit": "Guava"},
    {"fruit": "Papaya"},
    {"fruit": "Carambola"},
    {"fruit": "Longan"}
]
```

--------------------------------

### Example YouTube Data Array

Source: https://harpa.ai/chatml/steps-group-loop-jump

An example of a JSON array containing YouTube video data, including URL, title, views, and publication time. This format is used with LOOP steps.

```json
[{"url":"https://www.youtube.com/watch?v=B-8Is-2ocv0","title":"The 'Most Creative' Place In The World","views":470,"publishedTimeAgo":"2 weeks"},{"url":"https://www.youtube.com/watch?v=TAgLw1KKE34","title":"The Science of Storytelling","views":1221,"publishedTimeAgo":"3 months"},{"url":"https://www.youtube.com/watch?v=WRMncYn2bRg","title":"$695 Travel Credit Card: Worth the hype?","views":614,"publishedTimeAgo":"4 months"},{"url":"https://www.youtube.com/watch?v=uFuMzP9HE_E","title":"9/11 Survivor describes her escape from the 95th floor","views":2083495,"publishedTimeAgo":"5 months"},{"url":"https://www.youtube.com/watch?v=tbbo1pjA5j0","title":"I Bought A Billboard On Times Square","views":1334,"publishedTimeAgo":"6 months"},{"url":"https://www.youtube.com/watch?v=vrcQRq676Jg","title":"8h In United's Polaris Business Class (B777)","views":2352,"publishedTimeAgo":"7 months"}]
```

--------------------------------

### Custom and Global Parameters

Source: https://harpa.ai/chatml/parameters

Demonstrates how to define and use custom parameters (e.g., {{p1}}, {{name}}) and global parameters (e.g., {{g.p1}}, {{g.email}}) for command customization and cross-command persistence. Includes examples of setting parameters via ASK or CALC steps and verifying them with SAY.

```ChatML
CALC Set Parameter p1 = "custom_value"
SAY {{p1}}

CALC Set Parameter g.email = "user@example.com"
SAY {{g.email}}
```

--------------------------------

### Basic LOOP step setup

Source: https://harpa.ai/chatml/steps-group-loop-jump

Illustrates the basic setup of a LOOP step in Harpa AI ChatML. It shows how to add a LOOP step and set its 'list' parameter, with a default value of '{{list}}'. It also mentions editing child steps within the LOOP.

```harpa
Add LOOP step to your custom command and set the list parameter value. By default, it will be _{{list}}_.
Then, click **EDIT STEPS** button and drill down into the LOOP to edit child steps, which operate on the list item.
```

--------------------------------

### ChatML File Export Example

Source: https://harpa.ai/chatml/overview

This snippet demonstrates how HARPA AI commands are exported and shared, typically as .YAML files, which contain the ChatML structure defining the automation flow.

```YAML
# Example of a HARPA AI command in ChatML format
command:
  name: "ExampleCommand"
  description: "An example of a HARPA AI command."
  steps:
    - step: SAY
      message: "Hello, world!"
    - step: ASK
      variable: "userName"
      prompt: "What is your name?"
    - step: SAY
      message: "Nice to meet you, {{userName}}!"
```

--------------------------------

### GROUP Step Example

Source: https://harpa.ai/chatml/steps-group-loop-jump

Demonstrates how to use the GROUP step to combine multiple AI command steps. This is useful for organizing complex commands or creating iterative processes.

```ChatML
- condition: '{{change}} = refine'
  type: group
  steps:
    - type: ask
      param: instructions
      message: >-
        What changes would you like me to make?
    - type: gpt
      prompt: >-
        Please rewrite your answer, taking into account my
        additional instructions. Provide a rewritten response.

        [Additional instructions]: {{instructions}}

        Rewritten response:

```

--------------------------------

### Summing Values with JS

Source: https://harpa.ai/chatml/steps-run-js

This example demonstrates how to use the Run JS step to sum two user-provided numerical values. It prompts the user for two numbers, performs the addition using JavaScript, and displays the result.

```ChatML
meta:
  title: Sum values
  description: ''
  category: User
  name: cmd-sum-values
steps:
  - type: ask
    message: 'Please provide p1:'
    param: p1
    default: ''
  - type: ask
    message: 'Please provide p2:'
    param: p2
    default: ''
  - type: js
    code: return Number(args.p1) + Number(args.p2)
    args: p1, p2
    param: sum
  - type: say
    message: The sum is {{sum}}
version: 1

```

--------------------------------

### SAY Step in ChatML

Source: https://harpa.ai/chatml/steps-say-ask-gpt

The SAY step is used to display messages in the chat interface. It supports markdown syntax for rich text formatting. Examples show both full and short versions of the declaration.

```ChatML
# full version
- type: say
  message: Hello world!

# short version
- say: Hello world!
```

--------------------------------

### HARPA AI Command Creation

Source: https://harpa.ai/chatml/creating-custom-commands

This section details the steps to create a custom AI command using HARPA.AI. It involves installing the Chrome extension, connecting to AI models like ChatGPT, Gemini, or Claude, and using the HARPA AI interface to build multi-step automations.

```APIDOC
HARPA AI Command Creation:

1. Install HARPA AI Chrome Extension.
2. Connect to ChatGPT, Gemini, or Claude AI.
3. Open HARPA AI by clicking its icon or pressing Alt+A (Windows) / ^+A (Mac).
4. Access the AI chat interface, type '/', and click 'CREATE +'.
5. Customize the command by adding steps using the 'ADD STEP' button.
6. Save the command by clicking 'SAVE'.

Key Concepts:
- Commands: Custom multi-step AI-powered automations.
- Automation Steps: Building blocks for commands (e.g., SAY, ASK, GPT, NAVIGATE, CLICK, EXTRACT, RUN JS).
- Parameters: Custom inputs for commands.
- AI Element Selectors: For targeting specific elements on a webpage.
```

--------------------------------

### Extracting Data from Page with JS

Source: https://harpa.ai/chatml/steps-run-js

This example demonstrates extracting specific data, such as prices, from a web page using Harpa's parser API. It filters the parsed data to find items of type 'price' and returns their values and metrics.

```ChatML
meta:
  title: Find prices
  description: ''
  category: User
  name: cmd-find-prices
steps:
  - type: js
    code: |-
      const items = await $harpa.parser
        .parse(null, { extract: true })

      return items
        .filter(i => i.type === 'price')
        .map(p => ({
          value: p.value,
          metric: p.metric,
        }))
    param: prices
  - type: say
    message: 'Prices are: ```{{prices}}```'
version: 1

```

--------------------------------

### Using LOOP step with item properties

Source: https://harpa.ai/chatml/steps-group-loop-jump

Demonstrates how to use the LOOP step to iterate over a list and access properties of each item using dot notation. The example shows a SAY step displaying the current item's 'fruit' property.

```harpa
LOOP step
  list: _{{list}}_ 
  steps:
    SAY: "🔄 LOOP: {{item.fruit}}"
```

--------------------------------

### HARPA AI ChatML Scroll with JavaScript

Source: https://harpa.ai/chatml/steps-click-paste-scroll

Demonstrates how to use JavaScript within the RUN JS step in HARPA AI ChatML to perform various scrolling actions on a webpage. These examples are provided as alternatives until the native SCROLL step is implemented.

```javascript
return $harpa.page.scroll()
```

```javascript
return $harpa.page.scroll('bottom')
```

```javascript
return $harpa.page.scroll('top')
```

```javascript
return $harpa.page.scroll({ y: 1000 })
```

--------------------------------

### Custom Command: Grammar Check with Conditions and Jumps

Source: https://harpa.ai/chatml/conditions-and-jumps

This example demonstrates creating a HARPA AI ChatML custom command. It includes a GPT step for grammar correction, a condition to check for selected text, a STOP step for empty selections, and ASK/GPT steps for text manipulation (expand/shorten) with jump logic for looping.

```APIDOC
LABEL: START

GPT:
  prompt: |
    Please ignore all previous instructions.
    I want you to only respond in {{language}}.

    Act as a professional editor, specializing in grammar and syntax
    correction. Your task is to examine the given text and fix any grammatical
    or syntactical errors without altering the original style and tone. Change
    as little as possible to make it correct, and ensure it's well written.

    [TEXT TO CORRECT]:
    {{selection}}

    Corrected Text:

  condition: "{{selection}} is not empty"

STOP:
  condition: "{{selection}} is empty"
  label: "⛔️ STOP"
  value: "stop"

ASK:
  label: "Choose an action:"
  options:
    - label: "⛔️ STOP"
      value: "stop"
    - label: "⬆️ EXPAND"
      value: "expand"
    - label: "⬇️ SHORTEN"
      value: "shorten"
    - label: "♻️ NEW TEXT"
      value: "new"

GPT:
  prompt: "{{selection}}"
  condition: "{{option}} = expand"

GPT:
  prompt: "{{selection}}"
  condition: "{{option}} = shorten"

JUMP:
  condition: "{{option}} = expand or {{option}} = new"
  target_label: "START"
```

--------------------------------

### Extract YouTube Video Links with RUN JS

Source: https://harpa.ai/chatml/steps-run-js

This example demonstrates a RUN JS step that extracts YouTube video links, titles, view counts, and publication times from a web page. It parses anchor tags with '/watch' in their href attribute and extracts relevant information from the 'aria-label' and 'title' attributes.

```javascript
- type: js
  code: |-
    return Array
      .from(document.querySelectorAll('a[href*="/watch"]'))
      .map(a => {
        const ariaLabel = a.getAttribute('aria-label') || ''
        const viewsMatch
          = ariaLabel.match(/(\d[\d,]*) views/)
        const timeAgoMatch
          = ariaLabel.match(/views (.+) ago/)
        const url = new URL(a.href)

        return {
          url: 'https://www.youtube.com'
            + url.pathname
            + '?v='
            + url.searchParams.get('v'),
          title:
            a.getAttribute('title') || a.innerText.trim(),
          views: viewsMatch ? parseInt(viewsMatch[1].replace(/,/g, '')) : null,
          publishedTimeAgo: timeAgoMatch ? timeAgoMatch[1].trim() : null
        }
      })
      .filter(item =>
        item.url.includes('/watch?v=') &&
        item.views !== null &&
        item.publishedTimeAgo !== null
      )
  param: videos
  silent: true

```

--------------------------------

### Conditional Stop and Group Execution

Source: https://harpa.ai/chatml/steps-clear-stop

This example demonstrates conditional logic using STOP and GROUP steps. It allows users to choose between finishing a task, translating to a default language, or translating to a custom language, with the STOP step ensuring proper termination after each path.

```ChatML
- type: ask
  label: ask-for-change
  param: change
  options:
    - { label: '✅ DONE', value: 'no' }
    - { label: '🇫🇷 TRANSLATE TO FRENCH', value: 'french' }
    - { label: '🛠️ CUSTOM TRANSLATE', value: 'custom'}

- condition: '{{change}} = no'
  type: stop

- condition: '{{change}} = french'
  type: gpt
  prompt: >-
    Translate this text to French.

    Translated text:
- condition: '{{change}} = french'
  type: jump
  to: ask-for-change

- condition: '{{change}} = custom'
  type: group
  steps:
  - type: ask
    param: translate
    message: >
      Please specify target language:
  - type: gpt
    prompt: >-
      Please translate the answer you have written into
      the language I will specify below.
      Do not change the wording of the text; I only need the translation.

      [LANGUAGE]: {{translate}}

      Translated text:
  - type: stop

```

--------------------------------

### Harpa AI ChatML Automation Steps Overview

Source: https://harpa.ai/chatml/automation-steps

Provides a comprehensive list and brief description of all available automation steps in Harpa AI ChatML. Each step is designed for a specific function within custom commands.

```APIDOC
SAY, ASK, GPT:
  - Used to send messages to chat, ask for parameter values, and send AI requests.

NAVIGATE:
  - Navigates the opened tab to a specific URL.

WAIT:
  - Pauses command execution until the page stops updating, text or an element appears on the page, a JavaScript function returns true, or a timeout is reached.

REQUEST:
  - Sends data to a Webhook, calls a remote API, or fetches web page HTML.

CLICK, PASTE, SCROLL:
  - Interact with page elements: click elements, insert text into forms and inputs, and scroll.

EXTRACT:
  - Retrieves content of web elements on the page.

GROUP, LOOP, JUMP:
  - Control command flow: group steps, loop over data items, and jump to labeled steps.

RUN JS:
  - Executes JavaScript code within the scope of the opened web page.

COMMAND:
  - Calls other commands, enabling the creation of higher-order commands.

CLEAR, STOP:
  - Manage chat history and terminate the current command.

CALC:
  - Manipulate parameter values: set, modify, merge lists, and extract data.
```

--------------------------------

### Importing AI Commands

Source: https://harpa.ai/chatml/importing-custom-commands

This section details the steps to import AI commands into HARPA AI. It involves opening the HARPA AI interface, typing '/', clicking the 'Import' button, and then dragging and dropping command files into the provided window.

```APIDOC
Import AI Commands:
  1. Open HARPA AI (Alt+A on Windows, ^+A on Mac).
  2. Type '/' in the HARPA AI chat.
  3. Click the 'Import' button.
  4. Drag & drop command files into the window.
  5. Click the 'Import' button to upload.
```

--------------------------------

### ChatML Automation Steps Overview

Source: https://harpa.ai/chatml/overview

This section outlines the various automation steps available in ChatML for creating HARPA AI commands. These steps cover a wide range of actions from user interaction and AI requests to web navigation and data manipulation.

```APIDOC
SAY, ASK, GPT:
  - SAY: Prints messages to the chat interface.
  - ASK: Prompts the user for input.
  - GPT: Sends AI requests to a language model.

NAVIGATE:
  - NAVIGATE: Navigates the browser to a specified URL.

CLICK, PASTE, SCROLL:
  - CLICK: Clicks on a specified element on the web page.
  - PASTE: Inserts text into a form field.
  - SCROLL: Scrolls the web page.

REQUEST:
  - REQUEST: Fetches web page HTML content or calls remote APIs.

EXTRACT:
  - EXTRACT: Scrapes and extracts structured data from web pages.

GROUP, LOOP, JUMP:
  - GROUP: Groups a sequence of steps.
  - LOOP: Iterates over a dataset.
  - JUMP: Transfers control to another step or section.

RUN JS:
  - RUN JS: Executes JavaScript code within the browser context.

COMMAND:
  - COMMAND: Executes a predefined HARPA AI command.

CLEAR, STOP:
  - CLEAR: Clears the chat or resets the command state.
  - STOP: Halts the execution of the current command.

CALC:
  - CALC: Performs mathematical calculations.

Conditions and Jumps:
  - Conditional logic for controlling the flow of automation steps.
```

--------------------------------

### Add List Item

Source: https://harpa.ai/chatml/steps-calc

Appends a given value to the end of a list. Supports specifying the position (start, end, or specific index) for adding the item.

```ChatML
# Add list item
- type: calc
  func: list-add
  index: last
  list: list
  item: param
```

--------------------------------

### Navigate Step Declaration

Source: https://harpa.ai/chatml/steps-navigate

Demonstrates the different ways to declare the NAVIGATE step in ChatML, including full and short versions, handling missing protocols, relative paths, and reloading the current page.

```ChatML
# full version
- type: navigate
  url: 'https://google.com'

# short version
- navigate: 'https://google.com'

# missing protocol, https is assumed
- navigate: 'google.com'

# relative path
- navigate: '/search'

# reload opened page
- navigate: ''
```

--------------------------------

### Domain Name Generator ChatML Command

Source: https://harpa.ai/chatml/steps-say-ask-gpt

This ChatML command defines a custom workflow to generate domain name ideas. It first prompts the user for a topic or description using the 'ask' step and then uses the 'gpt' step to generate 20 domain name suggestions based on the input, focusing on search volume and availability. The 'language' parameter in the GPT prompt is dynamically set.

```chatml
meta:
  title: Domain name generator
  description: ''
  category: User
  name: hdruHBqvUDI3t2ymTDqU4
steps:
  - type: ask
    param: p1
    message: >-
      Enter your niche, product/service/business description, or leave empty
      to parse from page
    default: '{{page}}'
  - type: gpt
    prompt: >-
      Please ignore all previous instructions.
      I want you to respond only in {{language}}.

      Generate 20 domain name ideas for the given topic, niche or
      product/business/service description which have some search volume.

      Try to come up with domain names that are available and not already
      used by other companies.

      [TOPIC / NICE / PRODUCT / BUSINESS / SERVICE
      DESCRIPTION]: {{pi}]

      [DOMAIN NAMES LIST]:
version: 1
```

--------------------------------

### ASK Step in ChatML

Source: https://harpa.ai/chatml/steps-say-ask-gpt

The ASK step allows for user interaction by prompting for input and storing it in a parameter. It supports default values, predefined options, and custom options. Parameter values can be referenced using {{param_name}}.

```ChatML
# full version
- type: ask
  param: p1

# short version
- ask: p1

# default fallback value
- ask: p1
  default: '{{page}}'

# choice of options
- ask: option
  options: ['a', 'b', 'c']

# choice of options, with a custom one
- ask: option
  options: ['a', 'b', 'c', '$custom']

# choice of options with labels and values
- ask: option
  options:
    - { label: 'MAKE LONGER', value: 'make text longer' }
    - { label: 'MAKE SHORTER', value: 'make text shorter' }
```

--------------------------------

### Filter List Items

Source: https://harpa.ai/chatml/steps-calc

Filters items within a list based on specified criteria, allowing either retention or exclusion of matched values. For example, retaining only rows where 'Name = John'.

```ChatML
# Filter list items
- type: calc
  func: list-filter
  list: list
  item: param
  match: '{ "name": John }'
  matched: retain
```

--------------------------------

### Single Element Selection with $size

Source: https://harpa.ai/chatml/ai-element-selectors

This example demonstrates how to use the $size selector to ensure that exactly one element matching the CSS selector is found. If more or fewer than one element is found, the selector will return no elements.

```ChatML
- $css: h1
- $size: 1
```

--------------------------------

### Range Selection with $size

Source: https://harpa.ai/chatml/ai-element-selectors

This example shows how to use the $size selector with both 'min' and 'max' parameters to specify a range for the number of elements found. The selector will only return elements if the count falls within this specified range (inclusive).

```ChatML
- $css: a
- $size:
    min: 1
    max: 5
```

--------------------------------

### Hiding Elements by Text with JS

Source: https://harpa.ai/chatml/steps-run-js

This example shows how to hide HTML elements on a page that contain specific text. It prompts the user for text to hide, finds the corresponding elements using Harpa's query API, and removes them from the DOM.

```ChatML
meta:
  title: Hide elements by text
  description: ''
  category: User
  name: cmd-hide-elements-by-text
steps:
  - type: ask
    message: What to hide?
    param: text
    default: ''
  - type: js
    code: |-
      const elements = $harpa.query.run(
        { $text: args.text }, { mult: true })

      elements.forEach(e => e.remove())
    args: text
version: 1

```

--------------------------------

### Set and Clone Parameters with Calc Step

Source: https://harpa.ai/chatml/steps-calc

Demonstrates how to set a parameter to a captured value (like URL) and clone existing parameters (like GPT responses or lists) using the 'set' and 'clone' functions within the Calc step.

```ChatML
# Set parameter
- type: calc
  func: set
  param: saved.url
  format: auto
  value: '{{url}}'

# Clone parameter
- type: calc
  func: clone
  from: gpt
  to: cloned.gpt
```

--------------------------------

### GPT Step in ChatML

Source: https://harpa.ai/chatml/steps-say-ask-gpt

The GPT step sends a prompt to a Large Language Model and stores the response in a parameter. Prompts support parameter interpolation using double curly braces, e.g., {{parameter_name}}.

```ChatML
# full version
- type: gpt
  prompt: >
    Please ignore all previous instructions. Please provide 5 not so well known / surprising facts about {{p1}}. Format every fact:

    - [relevant fact emoji] [fact]

    Facts:
  param: gpt

# short version
- gpt: >
    Please ignore all previous instructions. Please provide 5 not so well known / surprising facts about {{p1}}. Format every fact:

    - [relevant fact emoji] [fact]

    Facts:
```

--------------------------------

### HARPA AI ChatML Parameters

Source: https://harpa.ai/chatml/parameters

This section outlines the various parameters that can be injected into prompts within HARPA AI's ChatML. These parameters allow for dynamic content inclusion from web pages, user selections, and other sources to enhance AI interactions.

```APIDOC
Page Parameters:
  {{page}}: Web page text content. Truncated for long pages.
    {{page url}}: Retrieves content from a specified URL. Blocks on LinkedIn.com and X.com. Supports JavaScript.
    {{page query}}: Scans page content for text semantically similar to a query.
    {{page limit=N}}: Specifies a limit in tokens or percentage of context window (e.g., 10%).

Content Parameters:
  {{selection}}: Selected text on the web page.
  {{title}}: Web page title.
  {{desc}}: Web page meta description.
  {{url}}: Web page URL.
  {{domain}}: Web page domain.
  {{language}}: Language defined in app settings.
  {{transcript}}: YouTube video transcript with timestamps.
  {{transcriptPlain}}: YouTube video transcript without timestamps.

Contextual Parameters:
  {{message x}}: Previous chat message (e.g., {{message 0}}, {{message -1}}).
  {{serp term}}: Queries the web for a term and returns top links. Supports search operators (e.g., {{serp running weight loss site:ncbi.nlm.nih.gov}}).
  {{gpt}}: Last GPT response.
  {{email}}: Parses email content from supported providers (Gmail, Outlook, Protonmail, Fastmail) into a JSON object with fields: user, sendTime, from, to, subject, content.
  {{thread}}: Parses messages/comments from threads (WhatsApp, Telegram, Discord, Reddit, Facebook, YouTube, X).
  {{view}}: Includes a screenshot of the opened page (requires Vision capabilities).
  {{cost}}: Cost of GPT requests in AI tokens. {{cost last}} for the last request.

Grab Parameter for HTML Element Extraction:
  {{grab}}: Extracts values from HTML elements.
    Usage:
      {{grab}}: Extracts website body text.
      {{grab h1}}: Extracts text from h1 elements using CSS selector.
      {{grab //h1}}: Extracts text from h1 elements using XPath.
      {{grab Bring AI to your browser}}: Extracts text matching the provided string.
    Attribute Extraction (take=value):
      {{grab h1 take=innerText}}: Takes the innerText of h1.
      {{grab h1 take=textContent}}: Takes the textContent of h1.
      {{grab h1 take=id}}: Takes the id attribute of h1.
      {{grab h1 take=className}}: Takes the className attribute of h1.
      {{grab h1 take=class}}: Takes the class attribute of h1.
      {{grab h1 take=nodeName}}: Takes the nodeName of h1.
      {{grab h1 take=nodeType}}: Takes the nodeType of h1.
      {{grab h1 take=attributes}}: Takes all attributes of h1.
      {{grab h1 take=[class]}}: Takes the value of the 'class' attribute of h1.
      {{grab h1 take=styles}}: Takes the styles of h1.
      {{grab h1 take=(color)}}: Takes the value of the 'color' style of h1.
      {{grab h1 take=innerHTML}}: Takes the innerHTML of h1.
      {{grab h1 take=outerHTML}}: Takes the outerHTML of h1.
      {{grab a take=href}}: Takes the href attribute of anchor tags.
      {{grab input take=value}}: Takes the value attribute of input elements.
    Position Specification (at=value):
      {{grab a take=href at=first}}: Takes the href of the first anchor tag.
      {{grab a take=href at=last}}: Takes the href of the last anchor tag.
      {{grab a take=href at=0}}: Takes the href of the anchor tag at index 0.
      {{grab a take=href at=1}}: Takes the href of the anchor tag at index 1.
      {{grab h2 at=all}}: Takes all h2 elements.

Parameter Phrasing Note:
  'take' specifies an argument's value. Example: {{grab YouTube video transcript take=innerText}} extracts the text value of the entire element.
  Example 1: Search for 'GPT', get the text value of the first element.
  Example 2: Ask for all elements, receive an array.
```

--------------------------------

### ChatML Structure and Parameters

Source: https://harpa.ai/chatml/overview

This documentation describes the fundamental structure of ChatML, a markup language for HARPA AI commands, and the role of parameters in managing data flow between automation steps.

```APIDOC
ChatML:
  - A human-readable format for declaring chat dialog flows and automations.
  - Stored in .YAML files.
  - Facilitates adoption of page-aware OpenAI GPT dialogs and personalized web automations.

Parameters:
  - Used to store data and pass it between automation steps.
  - Can store user input, AI responses, extracted data, etc.
  - Most steps read from and write to parameters.
```

--------------------------------

### Page Context Parameter Usage

Source: https://harpa.ai/chatml/parameters

Details how the {{page}} parameter can be used to manage web page content within GPT's token limits. It explains automatic truncation and how to specify keywords to extract relevant portions of the page.

```ChatML
GPT {{page Adobe Photoshop}}
```

--------------------------------

### ProtonMail Login Automation

Source: https://harpa.ai/chatml/steps-click-paste-scroll

This ChatML script automates the ProtonMail login process. It includes navigating to the login URL, pasting the email and password into the respective fields using AI-based selectors, and then clicking the submit button. The `onFailure: skip` setting ensures that if an element is not found, the script continues to the next step.

```ChatML
meta:
  title: ProtonMail Login
  description: ''
  category: User
  name: pqiqyO9Tl7blDTycH14sM
steps:
  - type: navigate
    url: https://account.proton.me/login
    waitForIdle: true
  - type: paste
    text: harpa.ai@protonmail.com
    selectorType: ai
    selector:
      - $matches:
          - $tag: INPUT
          - $id: username
          - $class: input-element
          - $class: w100
          - $attribute: autocomplete=username
          - $attribute: autocapitalize=off
          - $attribute: autocorrect=off
          - $attribute: spellcheck=false
          - $attribute: aria-invalid=false
          - $attribute: aria-describedby=id-3
          - $attribute: value
          - $attribute: data-last-active-input
          - $testId: input-input-element
          - $style: '"Inter var":14px:400:normal'
          - $role: form
            traverse: '2:1:0:0:0'
          - $class: sign-layout-main-content
            traverse: '0:2:1:0:0:0'
          - $class: ui-standard
            traverse: '1:0:2:1:0:0:0'
          - $class: sign-layout
            traverse: '1:0:2:1:0:0:0'
          - $class: shadow-color-primary
            traverse: '1:0:2:1:0:0:0'
          - $anchor: Email or username
            shift: '17:36'
          - $text: Email or username
            traverse: '-2:1:0:0:0'
          - $anchor: Password
            shift: 17:-61
          - $text: Password
            traverse: '-3:2:1:0:0:0'
          - $anchor: Enter your Proton Account details.
            shift: '17:80'
          - $text: Enter your Proton Account details.
            traverse: '-2:1:0:2:1:0:0:0'
        min: 11
      - $size: 1
    item:
      type: element
      container:
        __$ht: element
        id: 1
      value: Element
    showMore: false
    onFailure: skip
    waitForIdle: false
  - type: paste
    text: YouWontGuessThisOne
    selectorType: ai
    selector:
      - $matches:
          - $tag: INPUT
          - $role: textbox
          - $id: password
          - $class: input-element
          - $class: w100
          - $attribute: autocomplete=current-password
          - $attribute: autocapitalize=off
          - $attribute: autocorrect=off
          - $attribute: spellcheck=false
          - $attribute: aria-invalid=false
          - $attribute: aria-describedby=id-4
          - $attribute: type=text
          - $attribute: value
          - $attribute: data-last-active-input
          - $testId: input-input-element
          - $style: '"Inter var":14px:400:normal'
          - $role: form
            traverse: '3:1:0:0:0'
          - $class: sign-layout-main-content
            traverse: '0:3:1:0:0:0'
          - $class: ui-standard
            traverse: '1:0:3:1:0:0:0'
          - $class: sign-layout
            traverse: '1:0:3:1:0:0:0'
          - $class: shadow-color-primary
            traverse: '1:0:3:1:0:0:0'
          - $anchor: Password
            shift: '17:36'
          - $text: Password
            traverse: '-2:1:0:0:0'
          - $anchor: Keep me signed in
            shift: '-11:-60'
          - $text: Keep me signed in
            traverse: '-4:3:1:0:0:0'
          - $anchor: >-
              Not your device? Use a private browsing window to sign in and
              close it when done. Learn more
            shift: '-11:-80'
          - $text: >-
              Not your device? Use a private browsing window to sign in and
              close it when done. Learn more
            traverse: '-3:3:1:0:0:0'
        min: 11
      - $size: 1
    item:
      type: element
      container:
        __$ht: element
        id: 2
      value: Element
    showMore: false
    onFailure: skip
    waitForIdle: false
  - type: click
    selectorType: ai
    selector:
      - $matches:
          - $tag: BUTTON
          - $role: button
          - $class: button
          - $class: w100
          - $class: button-large
          - $class: button-solid-norm
          - $class: mt-6
          - $attribute: aria-busy=false
          - $attribute: type=submit
          - $style: '"Inter var":16px:400:normal'
          - $content: Sign in
          - $role: form
            traverse: '5'
          - $class: sign-layout-main-content
            traverse: '0:5'
          - $class: ui-standard
            traverse: '1:0:5'
          - $class: sign-layout
            traverse: '1:0:5'
          - $class: shadow-color-primary
            traverse: '1:0:5'

```

--------------------------------

### Harpa AI CLICK Step

Source: https://harpa.ai/chatml/steps-click-paste-scroll

The CLICK step simulates a mouse button press on a specified interface element. It utilizes AI Element Selectors to locate the target element, which can be done via the 'Grab' button or manual YAML configuration. The step includes options for failure handling and waiting for element readiness.

```ChatML
- type: click
  selectorType: ai
  selector: null
  item: null
  showMore: false
  onFailure: skip
  waitForIdle: true
```

```ChatML
- type: click
  selectorType: ai
  selector:
    - $matches:
        - $tag: A
        - $role: link
        - $class: Button
        - $class: GetPluginButton
        - $attribute: href=https://get.harpa.ai
        - $attribute: target=_blank
        - $attribute: type=button
        - $style: Poppins:24px:500:normal
        - $content: ADD TO BROWSER
        - $class: LandingHero__buttonContainer
          traverse: '0'
        - $class: LandingHero__content
          traverse: '0:1:0'
        - $class: LandingHero
          traverse: '0:0:1:0'
        - $class: LandingContent
          traverse: '0:2:0:0:1:0'
        - $class: Page__content
          traverse: '0:0:2:0:0:1:0'
        - $anchor: >
            Bring AI to your browser. Chat with websites, PDFs, videos, write emails, SEO articles, tweets, automate workflows, monitor prices & data. Bing AI & Notion AI alternative.
          shift: '-12:-106'
        - $text: >
            Bring AI to your browser. Chat with websites, PDFs, videos, write emails, SEO articles, tweets, automate workflows, monitor prices & data. Bing AI & Notion AI alternative.
          traverse: '-1:1:0'
        - $anchor: '#1'
          shift: '18:250'
        - $text: '#1'
          traverse: '-2:1:0'
        - $anchor: AI AUTOMATION AGENT FOR CHROME
          shift: '-40:250'
        - $text: AI AUTOMATION AGENT FOR CHROME
          traverse: '-2:1:0'
      min: 9
    - $size: 1
  showMore: false
  onFailure: skip
  waitForIdle: true
```

--------------------------------

### JUMP Step with Label for Loop

Source: https://harpa.ai/chatml/steps-group-loop-jump

Demonstrates how to create a potential infinite loop using the JUMP step by referencing a label that points back to itself.

```ChatML
- say: Hello world!
  label: infinite
- jump: infinite

```

--------------------------------

### HARPA AI JavaScript API

Source: https://harpa.ai/chatml/steps-run-js

This section outlines the available JavaScript API provided by HARPA AI within the RUN JS step. It includes methods for interacting with the web page, such as clicking elements, scrolling, parsing data, querying elements, fetching resources, and waiting for page idle states.

```APIDOC
$harpa.page.click: Clicks elements on the page.
$harpa.page.scroll: Scrolls the page up and down.
$harpa.page.parse: Parses data from the page.
$harpa.page.query: Queries elements on the page (e.g., by text content, CSS selectors).
$harpa.page.fetch: Sends data over HTTP or requests HTTP resources.
$harpa.page.idle: Waits for the page to stop updating.
$harpa.inspector.*: Provides methods to inspect document nodes.
```

--------------------------------

### Harpa AI Element Selectors Cheat Sheet

Source: https://harpa.ai/chatml/ai-element-selectors

Provides a quick reference for common AI Element Selectors used in Harpa AI, including text-based, anchor-based, CSS, XPath, parsing, positional, and ARIA role selectors.

```APIDOC
1. Search elements containing text:
My Account

2. Escaping numbers, dates, etc.:
Total views: {{data}}

3. Find an element closest to $anchor:
$anchor: username

4. CSS Selector $css:
$css: div.views-count > span

5. XPath Selector $xpath:
$xpath: //div/span

6. Parse number from element with $parse selector. Parse options include: number, date, time, prices, currency, text.
- Total views: {{data}}
- $parse: number

7. Take 4th matching element:
- $css: h2
- $at: 3

8. ARIA role selector:
$role: button
```
Tokens:
Raw