Introduction

The AI & Agents Actions are some of the most powerful tools available within AI Workflows. From pre-set AI activities such as text summarization and image generation, to creating personalized AI Agents, these actions enable human-like decision-making and autonomous task execution within your workflows.

At the moment, there are no Triggers related to AI & Agents. The Actions available in this section are described in this article.

All Actions in the AI & Agents rely on the LLMs connected to your platform via the AI Providers section.

---

Ask AI

This Action allows users to ask an AI to generate a response based on a customizable prompt.

Properties

Provider

Select your LLM from those you've already configured in the appropriate section.

Model

Select a model from those provided by the LLM you selected in the "Provider" field.

Prompt

Enter your prompt for the AI. You can combine static text with dynamic values entered via the Data Selector.

You can enter prompts in any language that is supported by your LLM of choice.

Conversation Key

In some cases, you may want multiple AI interactions to be part of a single conversation, allowing the AI to use its memory capabilities to recall previous interactions over time, saving you the trouble of having to repeat important context every time. This field allows you to do that by applying a custom ID or "name" to the conversation. By using the same Conversation Key across different Actions, these will refer to the same conversation and retain details from past interactions within it.

Creativity

Controls the creativity of the AI response. A higher value will make the AI more creative and a lower value will make it more deterministic. The default value is 100.

Max Tokens

Set the maximum amount of tokens that the request is allowed to consume. The default value is 2000.

Web Search

By enabling this option, you give the AI the ability to perform web searches if it deems it necessary to aid in its request. Certain requests might require this option to be enabled, such as requests about current-day news or those that require searching in public databases.

When enabling this option, you can set up other properties:

• Max Web Search Uses: Set the maximum number of searches that the AI is allowed to perform for each request. The default value is 5.
• Include Sources: If you enable this option, the AI will include in the response the sources that it found during its web searches. Useful for getting web search details (e.g. search queries, searched URLs, etc).
• Allowed Domains: List of domains to search (e.g., example.com, docs.example.com/blog). Domains should not include HTTP/HTTPS scheme. Subdomains are automatically included unless more specific subpaths are provided. Overrides Blocked Domains if both are provided.
• Blocked Domains: List of domains to exclude from search (e.g., example.com, docs.example.com/blog). Domains should not include HTTP/HTTPS scheme. Subdomains are automatically included unless more specific subpaths are provided. Overrided by Allowed Domains if both are provided.
• User Location - City: The AI can simulate a user located in a specific location for localized search results. In this field, you can specific a city (e.g., San Francisco)
• User Location - Region: Region or state for localized search results (e.g., California)
• User Location - Country: Country code for localized search results (e.g., US)
• User Location - Timezone: IANA-formatted timezone ID for localizing search results (e.g., America/Los_Angeles)

Continue on Failure

Enable this option to skip the current step and continue the flow normally if it fails.

Retry on Failure

Enable this option to automatically retry the current step up to four attempts when failed.

Output

The output of this Action will include the text of the AI-generated response. However, if your request had Web Search enabled with the "Include Sources" option turned on, you will receive the following objects in the response:

• text: The text of the response as generated by the AI.
• sources: An Array of all the sources the AI used in its response. Each source will include the following information:
  • type: Type of source. Usually just "source".
  • sourceType: Resource type. Usually "url".
  • id: Unique ID for the source for internal use by the AI provider.
  • url: URL of the source.
  • title: Title of the source.
  • providerMetadata: Metadata of the source as handled by the AI provider.

Example

Prompt (Web Search: On)

What is the current value of the Yen against US dollars?

Response

According to Yahoo Finance, the current value of USD/JPY (the US dollar-Japanese Yen exchange rate) is 158.8610. This represents a drop of 0.175 (0.11%) compared to the previous day closing.

---

Summarize Text

This Action allows users to ask an AI to summarize a given piece of text via a customizable prompt.

Properties

Provider

Select your LLM from those you've already configured in the appropriate section.

Model

Select a model from those provided by the LLM you selected in the "Provider" field.

Text

Enter the text you want to have summarized. You can combine static text with dynamic values entered via the Data Selector.

Prompt

A default prompt is already included:

Summarize the following text in a clear and concise manner, capturing the key points and main ideas while keeping the summary brief and informative.

You can personalize this prompt to change the instructions the AI must follow. You can also combine static text with dynamic values via the Data Selector.

You can enter prompts in any language that is supported by your LLM of choice.

Max Tokens

Set the maximum amount of tokens that the request is allowed to consume. The default value is 2000.

Continue on Failure

Enable this option to skip the current step and continue the flow normally if it fails.

Retry on Failure

Enable this option to automatically retry the current step up to four attempts when failed.

Output

The output of this Action will consist of the AI-generated summary of the provided text, also provided a text value.

Example

Text

AI Agents in Automation\n\nArtificial intelligence agents represent a transformative leap in the field of automation, moving beyond traditional rule-based systems into the realm of autonomous, decision-making software entities. Unlike conventional automation tools that follow rigid, pre-programmed instructions, AI agents possess the ability to perceive their environment, reason about complex situations, and take actions to achieve specified goals. These intelligent systems are increasingly being deployed across industries, from customer service and finance to manufacturing and healthcare, fundamentally reshaping how organizations approach repetitive and even cognitively demanding tasks.

At the core of AI agents lies a sophisticated architecture that combines machine learning models, natural language processing, and decision-making frameworks. Modern agents, particularly those powered by large language models, can interpret unstructured inputs, draw upon vast knowledge bases, and dynamically plan multi-step actions to accomplish objectives. They can interact with external tools, query databases, browse the web, execute code, and communicate with other agents or humans in natural language. This combination of capabilities allows AI agents to handle ambiguous tasks that previously required human judgment, such as resolving customer disputes, conducting market research, or coordinating supply chain logistics.

The benefits of integrating AI agents into automation workflows are substantial and multifaceted. Organizations report significant reductions in operational costs, faster turnaround times, and improved consistency in service delivery. AI agents can operate around the clock without fatigue, scale instantly to meet demand surges, and continuously learn from new data to refine their performance. Moreover, by automating routine cognitive work, these agents free human employees to focus on higher-value creative and strategic activities. In sectors like software development, AI coding agents are accelerating product development cycles, while in healthcare, diagnostic agents are assisting clinicians in identifying patterns that might otherwise go unnoticed.

Despite these advantages, the deployment of AI agents in automation raises significant challenges and ethical considerations. Issues of reliability, transparency, and accountability become paramount when autonomous systems make consequential decisions. Agents can exhibit unpredictable behaviors, hallucinate incorrect information, or be manipulated through adversarial inputs, posing risks in sensitive domains. Furthermore, the displacement of human workers from automated roles raises pressing questions about workforce transitions, retraining programs, and the equitable distribution of productivity gains. Establishing robust governance frameworks, audit mechanisms, and human-in-the-loop safeguards is essential to ensure that AI agents operate safely and align with organizational and societal values.

Looking ahead, AI agents are poised to become even more sophisticated, collaborative, and ubiquitous in automation ecosystems. Emerging trends point toward multi-agent systems where specialized agents work in concert to solve complex problems, as well as deeper integration between agents and physical systems through robotics and the Internet of Things. As these technologies mature, the relationship between humans and AI agents will likely evolve into one of genuine partnership, with agents serving as tireless collaborators that augment human capabilities. Navigating this future will require thoughtful innovation, balancing the immense potential of AI agents with the responsibility to deploy them in ways that benefit both organizations and the broader public."

Response

AI agents mark a major advancement in automation, surpassing traditional rule-based systems by autonomously perceiving environments, reasoning, and making decisions to achieve goals. Powered by machine learning, natural language processing, and large language models, they can handle complex, ambiguous tasks—interacting with tools, databases, and humans across industries like finance, healthcare, and manufacturing.

Key Benefits: AI agents reduce costs, improve consistency, operate continuously, scale on demand, and free humans for higher-value work. They are accelerating fields such as software development and medical diagnostics.

Challenges: Concerns include reliability, transparency, accountability, unpredictable behavior, misinformation, vulnerability to manipulation, and workforce displacement. Robust governance, auditing, and human oversight are essential.

Future Outlook: AI agents will grow more advanced and collaborative, with multi-agent systems and integration into robotics and IoT. The human-AI relationship is expected to evolve into a true partnership, requiring responsible innovation to balance benefits with societal impact."

---

Generate Image

This Action allows users to use AI to generate an image according to their specifications.

Properties

Provider

Select your LLM from those you've already configured in the appropriate section.

Model

Select a model from those provided by the LLM you selected in the "Provider" field.

Prompt

Provide a description of the image you want the AI to generate for you. This description can combine static text with dynamic values via the Data Selector.

Input Images

If you wish the AI to base the image it will generate on one or more existing images, either as edits, merges or other variations of them, you can provide them here, formatted as Files.

Continue on Failure

Enable this option to skip the current step and continue the flow normally if it fails.

Retry on Failure

Enable this option to automatically retry the current step up to four attempts when failed.

Output

The output of this Action will consist of the AI-generated image, formatted as a File that can be used in later steps.

---

Classify Text

This Action allows users to leverage AI to classify a provided text into one of multiple pre-established categories.

Properties

Provider

Select your LLM from those you've already configured in the appropriate section.

Model

Select a model from those provided by the LLM you selected in the "Provider" field.

Text to Classify

Provide the text that the AI must read and match with one of the categories provided in the "Categories" property. This text can combine static values with dynamic values via the Data Selector.

Categories

Establish the possible categories that the text provided in the "Text to Classify" property may be classified into.

Continue on Failure

Enable this option to skip the current step and continue the flow normally if it fails.

Retry on Failure

Enable this option to automatically retry the current step up to four attempts when failed.

Output

The output of this Action will consist of the Category that the AI determined for the provided text.

Example

Text to Classify

Hello, I would like to purchase an ElectroNeek license. What is your price range?

Categories

Sales Request

Technical Request

Security Request

Response

Sales Request

---

Extract Structured Data

This advanced AI-powered Action allows users to extract specific data points from provided text and/or files (PDF or images), and structure those data points in whichever specific way the user requires.

Properties

Provider

Select your LLM from those you've already configured in the appropriate section.

Model

Select a model from those provided by the LLM you selected in the "Provider" field.

Text

Provide the text that the AI must extract structured data from. This text can combine static values with dynamic values via the Data Selector.

Files

Provide one or multiple files that the AI must extract structured data from. Only PDF documents and images are supported for this field.

Guide Prompt

This is the prompt that will guide the AI in the extraction process. A default prompt is included:

Extract the following data from the provided data.

You can edit or replace the provided prompt with one of your choice. This prompt can combine static text with dynamic values via the Data Selector.

Data Definition

Use this section to provide the structure in which you want the AI to return the data it extracts from the given Text and/or Files. Each item in the Data Definition property consists of the following values:

• Name: Name of the value you want to extract from the Text and/or Files. It must be a unique value.
• Description: Brief description of the data that must fill this item. This acts as a hint for the AI on what to look for.
• Data Type: Type of parameter. Options are: Text, Number, Boolean.
• Fail if Not present?: If you enable this option, then this item must be present in order for the extraction to be successful. If it is not present, then the step will fail. If this option is not enabled, the respone may include an empty value for this item.

Max Tokens

Set the maximum amount of tokens that the request is allowed to consume. The default value is 2000.

Continue on Failure

Enable this option to skip the current step and continue the flow normally if it fails.

Retry on Failure

Enable this option to automatically retry the current step up to four attempts when failed.

Output

The output of this Action will be an object that contains the requested data, in the format that was defined by the user. Each of the keys of the output object will be named as per the Names of the Items in the "Data Definition" property, and their data types and values will correspond to your definition as well.

Example

Text

Dear Mr. Roger, there is a pending invoice for your subscription with OnePeace Life Insurance. This invoice is due on May 6, 2026. The total amount for payment is $576.

Guide Prompt

Extract the following data from the provided email about an invoice.

Data Definition

• Company (Text): Company that issued the invoice.
• Due Date (Text): Date by which the invoice must be paid. Return in YYYY-MM-DD format.
• Amount (Number): Amount to be paid.

Response

• Company = "OnePeace Life Insurance"
• Due Date = "2026-05-06"
• Amount = 576

---

Run Agent

This advanced action allows users to create and fully customize their own AI Agents to use within their AI Workflows.

Agents think in "steps", breaking down the user's request and solving it sequentially. In some of these steps, they can leverage user-provided Tools to autonomously perform actions related to their tasks, such as obtain information from external sources, search on the web, generate new records in ERP/CRM systems, leveraging other ElectroNeek products, among others.

Properties

Prompt

Enter your prompt for the AI. You can combine static text with dynamic values entered via the Data Selector. You can enter prompts in any language that is supported by your LLM of choice.

AI Model

Select your AI Provider from those you've already configured in the appropriate section, and afterwards, select a model from those available.

Agent Tools

In this section, you can provide Tools that the Agent will have access to. There are 3 types of Tools:

• Piece Tools: Leverage the existing Pieces (App connectors) to perform pre-configured Actions in their respective apps. You can add multiple Actions from the same Piece. For each Piece Tool Action added, you need to provide a Connection and other properties (just like with a regular App Action in your flow.) For the properties, you can determine whether you want to allow the Agent to fill it autonomously with the provided context ("Let agent decide"), fill it yourself ("Set value myself"), or leave it blank (only available for non-mandatory properties.)
• Flow Tools: Allow the Agent to execute one of your existing AI Workflows. Only flows that use the "MCP Tool" Trigger are available to use as Tools.
• MCP Tools: Generate a custom connection between your Agent and an external MCP server to leverage that server's built-in Tools. You will need to provide the following parameters: MCP Name, Server URL, Protocol (SSE, Simple HTTP, Streamable HTTP), and Authentication Type (None, Headers, Access Token, Api Key). Once you enter these details, press the "Validate Server" button to confirm the connection. After it's confirmed, you will see a list of available Tools from the MCP server, and you can choose which one to use.

Structured Output

If you want your Agent to provide specific, structured data points as part of its response, you can define them in this section. For each item, you must provide 3 values:

• Field Type: Text, Number, or Yes/No (Boolean)
• Field Name: Name of the data point you want returned.
• Field Description: Explain to the AI what this data point must contain.

Max Steps

Determine the maximum amount of steps that the AI will be allowed to perform in its analysis ("thought process") before it provides a resolution. Typically, each different request within the prompt takes up one step, as well as each instance of Tool usage. The default value is 20.

Web Search

By enabling this option, you give the AI the ability to perform web searches if it deems it necessary to aid in its request. Certain requests might require this option to be enabled, such as requests about current-day news or those that require searching in public databases.

When enabling this option, you can set up other properties:

• Max Web Search Uses: Set the maximum number of searches that the AI is allowed to perform for each request. The default value is 5.
• Include Sources: If you enable this option, the AI will include in the response the sources that it found during its web searches. Useful for getting web search details (e.g. search queries, searched URLs, etc).
• Allowed Domains: List of domains to search (e.g., example.com, docs.example.com/blog). Domains should not include HTTP/HTTPS scheme. Subdomains are automatically included unless more specific subpaths are provided. Overrides Blocked Domains if both are provided.
• Blocked Domains: List of domains to exclude from search (e.g., example.com, docs.example.com/blog). Domains should not include HTTP/HTTPS scheme. Subdomains are automatically included unless more specific subpaths are provided. Overrided by Allowed Domains if both are provided.
• User Location - City: The AI can simulate a user located in a specific location for localized search results. In this field, you can specific a city (e.g., San Francisco)
• User Location - Region: Region or state for localized search results (e.g., California)
• User Location - Country: Country code for localized search results (e.g., US)
• User Location - Timezone: IANA-formatted timezone ID for localizing search results (e.g., America/Los_Angeles)

Continue on Failure

Enable this option to skip the current step and continue the flow normally if it fails.

Retry on Failure

Enable this option to automatically retry the current step up to four attempts when failed.

Output

AI Agents provide two kinds of response: a Timeline and a standard Output object.

Timeline

AI Agents perform multi-layered analysis, similar to a human thought process, and this process is separated into steps. The steps are illustrated in the Timeline. There's 3 types of step:

• Prompt: The first step is always the Prompt, with all dynamic data parsed.
• Content: Natural-language descriptions of the ongoing process by the Agent, formatted as Markdown.
• Tool execution: Descriptions of Tools that the Agent chooses to use, and all input and output data involved.

Output object

As most other Actions, the "Run Agent" Action returns a standard object for its output, which is updated in real time as the Agent processes the tasks given. This object includes the following data:

• status: Status of the Agent. As soon as the Agent starts, it's set to "IN_PROGRESS". When the Agent completes its task, this is changed to "COMPLETED".
• steps: Array with the details of each step in the timeline. Each step consists of:
  • type: For each of the steps (Prompt not included), this type may be "MARKDOWN" (for Content steps) or "TOOL_CALL" (for Tool execution steps.)
  • markdown: On Content steps, the Markdown-formatted contents of the step.
  • toolName: On Tool execution steps, the name of the Tool being used.
  • toolCallId: On Tool execution steps, a unique ID for the Agent-Tool interaction.
  • status: On Tool execution steps, the status of the Tool execution.
  • input: On Tool execution steps, the input values that the Agent passed to the Tool.
  • output: On Tool execution steps, the output values returned to the Agent by the Tool.
  • startTime: On Tool execution steps, the date and time when the execution began.
  • toolCallType: On Tool execution steps, describes the type of Tool: "PIECE", "FLOW" or "MCP".
  • pieceName: On Tool execution steps where toolCallType is "PIECE", contains the name of the Piece used.
  • pieceVersion: On Tool execution steps where toolCallType is "PIECE", contains the Piece version number.
  • actionName: On Tool execution steps where toolCalltype is "PIECE", contains the name of the Action that the Agent performed.
  • endTime: On Tool execution steps, the date and time when the execution ended.
• structuredOutput: Provides the values for the items requested in the "Structured Output" property, using the provided names.
• prompt: Contains the full Prompt, with parsed dynamic values.