# Using PromptLoop Tasks Tasks as PromptLoops action tools. Tasks allow you to run a workflow on an input(s), often from a row of data, and return one or more outputs. Tasks allow you to define a workflow or task, and run it on a single input while testing or thousands of rows of data using the exact same formatting and sequence. PromptLoop plans and executes tasks in a repeatable way to build out reliable business flows to automate processes. At a high level, tasks allow you to generate consistent, formatted results to fill in blanks and augment your data. They work by taking inputs (typically rows of data) and using them for a search, labeling, or other AI task. You create tasks with a simple prompt that can be used on every row of your data. **Helpful Links** - [Create a task](/account/custom) - [Example Tasks to Copy](/account/custom/library) - [Your Accounts Tasks](/account/custom) <Callout title="Getting Started?"> If you are just getting started, copying a task from one of the standard templates is usually the best way to get started. You can submit edits, and creating and editing tasks is free and included with all plans. Tasks can also be deleted easily once you are done exploring or if you make a mistake </Callout> ## Getting Started <VideoLink video_src='https://share.descript.com/view/zn86eArPgah' image_src='https://web-public-photos.s3.amazonaws.com/testing-new.jpg'/> ## Step 1 - Create a task or use a template Start by creating a task using a prompt in our web editor. Explain what you want and generate a trained and deployed new task in 90 seconds. You can use Excel or Google Sheets, or upload a dataset. The AI will be trained based on your task and requirements. **Options to Creating a Task** - **Template** - Copy a template and get started right away - **Form** - Use the form to identify the inputs and outputs you will use. What data or columns will you select for the model (what info do you have) and what new data points are you looking for. - **Prompt** - Describe in plain language, like you would in an email, what you are looking for and a task will generate. Please review the capabilities to ensure the model will be able to accomplish what you describe. ### Use a Template Starting with one of our templates and editing it is something the fastest way to get started. Select an option in the drop down or navigate to the template library. Once you copy a template you can open it up in the editor to modify it for your needs. This is often the fastest way to get started. Instant templates include: - **Find Information from a Website** - This is a general QA task to take a link and extract info. This can include live data, nuanced questions, or items that require browsing. - **Extract List from Website** - These tasks let you build a list of entities from a website that lists items. This could be a product page, address book, or profile list. You can define the items you want (name, address, phone) and the model will return a formatted list for each input site. - **Text Gen (Non Web)** - This is a template for a single step non web browsing task. You can edit the instructions (prompt) when you copy it. <Image alt={'Custom Tasks AI job automation'} src={'https://web-public-photos.s3.amazonaws.com/dropdown.jpg'} width={800} height={420} /> ### Use the Form If you know what the inputs and outputs will be for each row, you can use the form for a faster more accurate task. For example, if you have a list of website and want to return a few questions about them, you can use the form to create a task. When starting you will be presented with two options - **Web Browsing** - These tasks will access the internet for research and crawling sites with live data. You can use either a specific link, a domain, or a search term as an input(s) - **Language Tasks** - These are tasks that do not use internet models or access web pages. This encompasses general AI work like summarizing or reformatting content that you already have. <Image alt={'Custom Tasks AI job automation'} src={'https://web-public-photos.s3.amazonaws.com/CleanShot%202024-05-12%20at%2012.12.21%402x.jpg'} width={800} height={420} /> **Selecting Inputs** You then need to identify what the inputs to the model are. Often this is text you have in a column in your dataset. It could be a website or multiple search terms, or unformatted text. **Selecting Outputs** PromptLoop is task driven. For each task you identify what you want the model to return, and how you want it returned including the desired format. Our systems then identify the best models and steps to achieve that. You can add, edit and delete these steps along the way with new versions of the task. ### Use a Prompt If you are using our chat assistant or just want to describe what you need, you can use the simple prompt form. You can think of this like an email request for the data you are looking to build out. Once your task generates successfully. You can use the editor to modify it further. <Image alt={'Custom Tasks AI job automation'} src={'https://web-public-photos.s3.amazonaws.com/CleanShot%202024-05-08%20at%2010.34.23%402x.jpg'} width={800} height={420} /> ## Step 2 - Test Task The next step once you have created or copied a task is to test it. Enter an input (often a website) and click **Run** to kick off the PromptLoop task. You will see results returned and any issue or error will be displayed. PromptLoop allows unlimited task creation in case you aren't getting the results you want to. <Image alt={'Custom Tasks AI job completion'} src={'https://web-public-photos.s3.amazonaws.com/testing-new.jpg'} width={800} height={420} /> You can try adjusting your creation instructions or reach out to us for help. You can refine your tasks as much as you need to ensure the AI model is trained to your specific requirements. Any time you make an edit you can immediately test the live task using the test page. You will be able to see the inputs and outputs that the task generated, as well as the steps it will take to return your data. ## Step 3 Editing Tasks Because tasks are fast and flexible, it is simple to submit multiple edits to tasks as your needs evolve and you want to improve them. If you need to make a change to a task, like adding an item to find, can toggle between **Edit** and **Test** using the buttons on the top of the task test page or on the more option on the [task list](/account/custom). <Image alt={'Custom Tasks AI job completion'} src={'https://img.promptloop.com/edit-toggle.jpg'} width={800} height={420} /> Editing is fast and simple but only available for small changes to tasks. If you need to change the inputs or operation, we suggest making a new task which you can do [here](/account/custom/new). <Image alt={'Editing Tasks'} src={'https://web-public-photos.s3.amazonaws.com/CleanShot%202024-05-08%20at%2010.37.24%402x.jpg'} width={800} height={420} /> Editing works similarly to using the form to create a task. You will be able to modify output items and instructions to help the task better achieve your goals. ### Search Items Search Items are the results of a task. These are returned as new columns when you run a series of inputs or a dataset. Each search item has several sections. - Search Item Name - This is the column header and must be completed - Query - The query is the action or question for the column. This can be short and concise as should not be a long AI style prompt. - Action Instructions - Any additional specifics like formatting, nuance or error avoidance can be added here <Callout title="Answer Format Options"> Below are the different output formats you can specify when creating AI tasks: - **Text** Free-form text response. Ideal for descriptions, summaries, and detailed explanations. - **True/False** Binary response format. Perfect for simple verification tasks and qualifying criteria. - **Number** Numeric output. Use for counts, measurements, statistics, or any numerical data points. - **Link** URL output. For tasks that need to extract or verify web links and resources. - **Image Link** Image URL output. Best for tasks requiring image source extraction or validation. - **JSON** Structured data output in JSON format. Useful when you need multiple data points in a standardized format. - **Script** Code or script output. For tasks that generate executable code snippets or commands. - **Single Category** One category selection from predefined options. Great for classification tasks. - **Multiple Categories** Multiple category selections allowed. Ideal for tagging and multi-label classification tasks. Each format ensures consistent outputs across your datasets, making it easier to process and analyze results at scale. </Callout> <Image alt={'Editing Tasks'} src={'https://img.promptloop.com/editor-options.jpg'} width={400} height={220} /> <Callout title="Not all Tasks are Editable"> Not all tasks are editable. In some cases, your task will have generate with too many steps to edit manually. You can reach out to the team for help or start over with a new task. Remember, tasks are free to create. </Callout> ## Step 4 - Upload your data to use the task Once you have a task, you can upload data with the appropriate inputs to run it on your files. When you run a task on a file, you will get a new version of your data with all of the new columns that the task provide. You can then explore this in the [datasets](/docs/autoloop) tab where you can search and filter your results <Image alt={'Datasets'} src={'https://web-public-photos.s3.amazonaws.com/CleanShot%202024-05-08%20at%2010.25.10%402x.jpg'} width={800} height={420} /> ## Chained Tasks Transform complex workflows into simple automated processes by connecting multiple tasks together. Chained tasks are perfect for when you need to: - Enrich data through multiple steps - Process information sequentially - Combine different data sources ### How It Works Let's look at a real example: <Callout title="Website Data Enrichment Chain"> 1. Task 1: Extract company details from a website URL 2. Task 2: Search task to find linkedin company link using company name Input: website URL ↓ Task 1 Output: company_name, company_email ↓ Task 2 Output: social_profiles Output: Company Details + Linkedin Link </Callout> Chains are a simple way to make running PromptLoop tasks easier once you have several set up that you are running on a recurring basis. If you need help setting up your first chain please reach out. ### Building Your First Chain 1. **Start with Inputs**: Add your initial data input (e.g., Website URL) 2. **Add Tasks**: Click "Add Step" to include each task in your sequence 3. **Map Data**: Connect outputs from previous steps to inputs of the next task 4. **Create Chain**: Save your workflow as a single automated process ### Current Limitations ⚠️ Note: Once created, chains cannot currently be edited. We recommend testing your workflow thoroughly before saving. ### What our customers use us for today We built custom tasks to meet the enormous variety of nuanced searches that our customers are using PromptLoop for every day. Custom Tasks allow you to craft and drive an AI task in as simple a form factor as possible - and then let it work for you on top of your datasets like a well-trained analyst. Some examples of research jobs that have made significant custom tasks - Domain-specific research - Does this company require HIPAA compliance software solutions? Who are their customers? What recent blog posts have they published - Find information about small businesses like contact info, locations, and services. - Market Sizing - Does this company offer X product or operate in Y space? (Scale of 1-10) - Taking a list of addresses - what public ownership information exists ### Need more? PromptLoop is a growing platform. Our tasks can currently cover various customer tasks but we are constantly adding capabilities. Custom Tasks will handle your job and will be completed per your specifications. You can monitor the progress and make any necessary adjustments along the way. This is a powerful tool that can revolutionize the way you handle business tasks. Get started with Custom Tasks today! <Callout title="Get Started"> To get started with Custom Tasks, request a [demo here](/demo) </Callout> ## Capabilities Detail PromptLoop tasks draw on a wide range of capabilities that are combined to provide you a simple command and tool to use to accomplish your goals. Each task can draw on AI language models, search services, and our web browsers to provide precise and customized solutions. Current capabilities give an overview of what PromptLoop Tasks can accomplish well, and the roadmap indicates areas under development. If you have specific questions or would like to enquire about capabilities, please reach out to the team [here](/demo). ### Current Capabilities **Web Research Focused** - Keyword Extraction - Classification of websites based on criteria - Summarization of websites or data - Specific, structured data extraction from websites - Google reviews retrieval **Text Analysis and Transformation** - Fixed example labeling - Sentiment and summarization of natural language processing - Dynamic content creation - Translation - Keyword extraction ### Sample Use Cases - Investment / Search fund target research and business data enrichment - Filtering unformatted data for use in quantitative studies and decision-making - Classifying 1000s of websites into various categories based on their content in less than an hour. - Structured summarization of medical papers. - Keyword and Tag Extraction from unstructured text data - Multi Step Natural Language Processing for text data - Classifying companies based on the number of reviews - Enriching business data across unique and customized data points and questions ### Roadmap Near-term releases or capabilities available on request: - Full page comparison (given 1 or 2 websites classify similar businesses) - Proprietary data integration (bring your data) - Continued improvements in strict typing and structuring of data to match desired outputs - Search services on enriched data - List aggregation (generate a list matching specifications) - Full dataset creation

    Using PromptLoop Tasks

    Tasks as PromptLoops action tools. Tasks allow you to run a workflow on an input(s), often from a row of data, and return one or more outputs.

    Tasks allow you to define a workflow or task, and run it on a single input while testing or thousands of rows of data using the exact same formatting and sequence. PromptLoop plans and executes tasks in a repeatable way to build out reliable business flows to automate processes.

    At a high level, tasks allow you to generate consistent, formatted results to fill in blanks and augment your data. They work by taking inputs (typically rows of data) and using them for a search, labeling, or other AI task. You create tasks with a simple prompt that can be used on every row of your data.

    Helpful Links

    Getting Started?

    If you are just getting started, copying a task from one of the standard templates is usually the best way to get started. You can submit edits, and creating and editing tasks is free and included with all plans. Tasks can also be deleted easily once you are done exploring or if you make a mistake

    Getting Started

    Watch the video

    Step 1 - Create a task or use a template

    Start by creating a task using a prompt in our web editor. Explain what you want and generate a trained and deployed new task in 90 seconds. You can use Excel or Google Sheets, or upload a dataset. The AI will be trained based on your task and requirements.

    Options to Creating a Task

    • Template - Copy a template and get started right away
    • Form - Use the form to identify the inputs and outputs you will use. What data or columns will you select for the model (what info do you have) and what new data points are you looking for.
    • Prompt - Describe in plain language, like you would in an email, what you are looking for and a task will generate. Please review the capabilities to ensure the model will be able to accomplish what you describe.

    Use a Template

    Starting with one of our templates and editing it is something the fastest way to get started. Select an option in the drop down or navigate to the template library.

    Once you copy a template you can open it up in the editor to modify it for your needs. This is often the fastest way to get started.

    Instant templates include:

    • Find Information from a Website - This is a general QA task to take a link and extract info. This can include live data, nuanced questions, or items that require browsing.
    • Extract List from Website - These tasks let you build a list of entities from a website that lists items. This could be a product page, address book, or profile list. You can define the items you want (name, address, phone) and the model will return a formatted list for each input site.
    • Text Gen (Non Web) - This is a template for a single step non web browsing task. You can edit the instructions (prompt) when you copy it.
    Custom Tasks AI job automation

    Use the Form

    If you know what the inputs and outputs will be for each row, you can use the form for a faster more accurate task. For example, if you have a list of website and want to return a few questions about them, you can use the form to create a task.

    When starting you will be presented with two options

    • Web Browsing - These tasks will access the internet for research and crawling sites with live data. You can use either a specific link, a domain, or a search term as an input(s)
    • Language Tasks - These are tasks that do not use internet models or access web pages. This encompasses general AI work like summarizing or reformatting content that you already have.
    Custom Tasks AI job automation

    Selecting Inputs You then need to identify what the inputs to the model are. Often this is text you have in a column in your dataset. It could be a website or multiple search terms, or unformatted text.

    Selecting Outputs PromptLoop is task driven. For each task you identify what you want the model to return, and how you want it returned including the desired format. Our systems then identify the best models and steps to achieve that.

    You can add, edit and delete these steps along the way with new versions of the task.

    Use a Prompt

    If you are using our chat assistant or just want to describe what you need, you can use the simple prompt form. You can think of this like an email request for the data you are looking to build out.

    Once your task generates successfully. You can use the editor to modify it further.

    Custom Tasks AI job automation

    Step 2 - Test Task

    The next step once you have created or copied a task is to test it. Enter an input (often a website) and click Run to kick off the PromptLoop task. You will see results returned and any issue or error will be displayed. PromptLoop allows unlimited task creation in case you aren't getting the results you want to.

    Custom Tasks AI job completion

    You can try adjusting your creation instructions or reach out to us for help. You can refine your tasks as much as you need to ensure the AI model is trained to your specific requirements. Any time you make an edit you can immediately test the live task using the test page.

    You will be able to see the inputs and outputs that the task generated, as well as the steps it will take to return your data.

    Step 3 Editing Tasks

    Because tasks are fast and flexible, it is simple to submit multiple edits to tasks as your needs evolve and you want to improve them.

    If you need to make a change to a task, like adding an item to find, can toggle between Edit and Test using the buttons on the top of the task test page or on the more option on the task list.

    Custom Tasks AI job completion

    Editing is fast and simple but only available for small changes to tasks. If you need to change the inputs or operation, we suggest making a new task which you can do here.

    Editing Tasks

    Editing works similarly to using the form to create a task. You will be able to modify output items and instructions to help the task better achieve your goals.

    Search Items

    Search Items are the results of a task. These are returned as new columns when you run a series of inputs or a dataset. Each search item has several sections.

    • Search Item Name - This is the column header and must be completed
    • Query - The query is the action or question for the column. This can be short and concise as should not be a long AI style prompt.
    • Action Instructions - Any additional specifics like formatting, nuance or error avoidance can be added here
    Answer Format Options

    Below are the different output formats you can specify when creating AI tasks:

    • Text Free-form text response. Ideal for descriptions, summaries, and detailed explanations.
    • True/False Binary response format. Perfect for simple verification tasks and qualifying criteria.
    • Number Numeric output. Use for counts, measurements, statistics, or any numerical data points.
    • Link URL output. For tasks that need to extract or verify web links and resources.
    • Image Link Image URL output. Best for tasks requiring image source extraction or validation.
    • JSON Structured data output in JSON format. Useful when you need multiple data points in a standardized format.
    • Script Code or script output. For tasks that generate executable code snippets or commands.
    • Single Category One category selection from predefined options. Great for classification tasks.
    • Multiple Categories Multiple category selections allowed. Ideal for tagging and multi-label classification tasks.

    Each format ensures consistent outputs across your datasets, making it easier to process and analyze results at scale.

    Editing Tasks
    Not all Tasks are Editable

    Not all tasks are editable. In some cases, your task will have generate with too many steps to edit manually. You can reach out to the team for help or start over with a new task. Remember, tasks are free to create.

    Step 4 - Upload your data to use the task

    Once you have a task, you can upload data with the appropriate inputs to run it on your files. When you run a task on a file, you will get a new version of your data with all of the new columns that the task provide. You can then explore this in the datasets tab where you can search and filter your results

    Datasets

    Chained Tasks

    Transform complex workflows into simple automated processes by connecting multiple tasks together. Chained tasks are perfect for when you need to:

    • Enrich data through multiple steps
    • Process information sequentially
    • Combine different data sources

    How It Works

    Let's look at a real example:

    Website Data Enrichment Chain
    1. Task 1: Extract company details from a website URL
    2. Task 2: Search task to find linkedin company link using company name

    Input: website URL

    ↓ Task 1 Output: company_name, company_email ↓ Task 2 Output: social_profiles

    Output: Company Details + Linkedin Link

    Chains are a simple way to make running PromptLoop tasks easier once you have several set up that you are running on a recurring basis. If you need help setting up your first chain please reach out.

    Building Your First Chain

    1. Start with Inputs: Add your initial data input (e.g., Website URL)
    2. Add Tasks: Click "Add Step" to include each task in your sequence
    3. Map Data: Connect outputs from previous steps to inputs of the next task
    4. Create Chain: Save your workflow as a single automated process

    Current Limitations

    ⚠️ Note: Once created, chains cannot currently be edited. We recommend testing your workflow thoroughly before saving.

    What our customers use us for today

    We built custom tasks to meet the enormous variety of nuanced searches that our customers are using PromptLoop for every day. Custom Tasks allow you to craft and drive an AI task in as simple a form factor as possible - and then let it work for you on top of your datasets like a well-trained analyst. Some examples of research jobs that have made significant custom tasks

    • Domain-specific research - Does this company require HIPAA compliance software solutions? Who are their customers? What recent blog posts have they published
    • Find information about small businesses like contact info, locations, and services.
    • Market Sizing - Does this company offer X product or operate in Y space? (Scale of 1-10)
    • Taking a list of addresses - what public ownership information exists

    Need more?

    PromptLoop is a growing platform. Our tasks can currently cover various customer tasks but we are constantly adding capabilities. Custom Tasks will handle your job and will be completed per your specifications. You can monitor the progress and make any necessary adjustments along the way.

    This is a powerful tool that can revolutionize the way you handle business tasks. Get started with Custom Tasks today!

    Get Started

    To get started with Custom Tasks, request a demo here

    Capabilities Detail

    PromptLoop tasks draw on a wide range of capabilities that are combined to provide you a simple command and tool to use to accomplish your goals. Each task can draw on AI language models, search services, and our web browsers to provide precise and customized solutions. Current capabilities give an overview of what PromptLoop Tasks can accomplish well, and the roadmap indicates areas under development. If you have specific questions or would like to enquire about capabilities, please reach out to the team here.

    Current Capabilities

    Web Research Focused

    • Keyword Extraction
    • Classification of websites based on criteria
    • Summarization of websites or data
    • Specific, structured data extraction from websites
    • Google reviews retrieval

    Text Analysis and Transformation

    • Fixed example labeling
    • Sentiment and summarization of natural language processing
    • Dynamic content creation
    • Translation
    • Keyword extraction

    Sample Use Cases

    • Investment / Search fund target research and business data enrichment
    • Filtering unformatted data for use in quantitative studies and decision-making
    • Classifying 1000s of websites into various categories based on their content in less than an hour.
    • Structured summarization of medical papers.
    • Keyword and Tag Extraction from unstructured text data
    • Multi Step Natural Language Processing for text data
    • Classifying companies based on the number of reviews
    • Enriching business data across unique and customized data points and questions

    Roadmap

    Near-term releases or capabilities available on request:

    • Full page comparison (given 1 or 2 websites classify similar businesses)
    • Proprietary data integration (bring your data)
    • Continued improvements in strict typing and structuring of data to match desired outputs
    • Search services on enriched data
    • List aggregation (generate a list matching specifications)
    • Full dataset creation