🔷 Common Tools and Actions in Agent Builder
When creating your AI Agent, you can build powerful automations by combining one or more of the following Action Types in your Agent Flow. Each action performs a distinct task—ranging from classification to advanced AI generation.
🔹 Classify Message
Automatically classify incoming text (emails, tickets, chat messages, or text) based on content, intent, topic, tone, or urgency.
How to configure
Add a Conversation Skills tool to your agent flow, then select “Classify Message” action and configure.
🔹 Classify Attachment or File
Use AI to classify documents such as PDFs, Word files, Excel sheets, or images based on content or metadata.
How to configure
Add a Document Skills tool to your agent flow, then select “Classify Attachment” action and configure.
🔹 Extract Text or Structured Data
You can extract the text of the files, or extract specific information from a file.
Extract text from files (attachments)
You can extract the text inside a file using Optical Character Recognition (OCR) technology and use it in the following steps. For example, you can insert the text of a file as a comment in Front App, Zendesk for Salesforce ServiceCloud so your human agents can access their text readily without opening the attached files one by one.
How to Configure
Add Document Skills Tool to your agent flow, and select “PDF/Image to Text (OCR)” action and configure it.
Extract structured data (field-values) from messages (text) and files
You may include steps to extract structured data from unstructured input such as:
Email body (available via ‘Extract Information’ action from Conversation Skills tool)
Documents like PDFs, images, Excel files (available via ‘Extract Information’ from’ Document Skills tool)
How to Configure
Add the Extract Information action from the appropriate skill type (Conversation Skills or Document Skills).
Choose the source (e.g., File URL or email body).
Select the file type: PDF, image, Excel, etc. (if extracting structured data from file) or text body (if extracting information from email or a text content)
Choose a prebuilt extractor (e.g., Standard Invoice Extractor) or one of your own custom extractors (you may create a custom extractor by following this guide in advance).
Test the extractor to ensure expected output.
Output: Structured data in a JSON object with the extracted fields including header fields and line item fields.
🔹 Entity Extraction
Identify and extract named entities such as names, dates, customer IDs, product codes, and more from text input. Supported entity types include:
Person Name
Date of Birth
Address
Email
Phone Number
Passport Number
Driver License Number
US Tax Identification Number (TIN)
US Social Security Number (SSN)
UK Unique Taxpayer Reference (UTR)
UK National Insurance Number (NINO)
UK National Health Service Number
Bank Account Number
International Banking Account Number (IBAN)
Canada Health Number
Canada Social Insurance Number
Vehicle Identification Number (VIN)
How to Configure
Add Conversation Skills or Document Skills tool to your agent flow (depending on whether processing message/text or files) and select ‘Find Mentions’ action and configure it. You can choose one or more available entity types.
Also, if you are looking for specific keywords, you can enter them in the ‘Keywords’ field.
🔹 Translation
Translate messages or document content into your desired language. Supports multilingual workflows.
How to Configure
Add Conversation Skills or Document Skills tool to your agent flow (depending on whether processing message/text or files) and select ‘Translate’ action and configure it.
🔹 Search, Query and Matching
In many situations, you may need to find specific values or query terms or values from the content of spreadsheets or UpBrains AI Collections (that are built on top of OpenSearch indexes as the storage for our RAG (Retrieval-Augmented Generation) agents.
You can include Search, Query and Matching capabilities in your agent flows using one of the following actions:
Partial Match (action available through Smart Classifier tool) – operates on the output from Google Sheet, Microsoft Excel Online or CSV content and offers an AI-powered partial matching capability to find values with approximate matching capability. It performs approximate or fuzzy matching within the data pulled from spreadsheets or data tables (e.g., match customer names or product SKUs with typos).
Query (action available through OpenSearch tool) – This offers a comprehensive, AI-powered search and query over an OpenSearch index (called an UpBrains AI Data Collection).
These actions are ideal for customer information lookup, product catalog search, or referencing predefined knowledge based, depending on where your data is located.
Note: you can create Data Collections in UpBrains AI and sync it with external data sources. The Data Collections will be available to AI agents to use as their knowledge base or perform search and query.
🔹 Transform / Map
Apply custom logic or JSON mapping and manipulation formulas to input data using programs, rules or scripting to transform some JSON data into the input of another flow step or an external application.
Here are key actions that you can use to transform or map data:
JSONata (advanced JSON manipulation) – This action is available from JSON tool sn support JSONata (https://docs.jsonata.org/overview.html) – a query and transformation language for JSONs – on input that can be referenced as Input_1, Input_2, etc in the Query input.
Static object building using “Build JSON” action from JSON tool by defining a set of key and values that will be formed into a JSON dictionary.
There are a number of other tools for operating and transforming different types of data, each offering many actions:
CSV
JSON
XML
Date Helper
Text Helper
File Helper
Image Helper
🔹 Draft a Response with AI
Through these actions, AI agents draft personalized replies using AI. You can configure these actions and provide context, tone, and goals to guide the response generation process.
Use this action in combination with classification, extraction results as well as Knowledge Bases as tools to AI agents for highly relevant answers.
🔹 Generate Text, Image or Voice with AI
Use generative AI to create:
Text summaries, intros, or explanations
Images (e.g., banners, thumbnails)
Voice messages (TTS)
This is useful for marketing automation, support scenarios, or creative flows.
How to Configure
You can use the “Open AI” tool, and use custom prompts and use different models available from Open AI in your Agent flow.
🔹 Call Large Language Models (LLMs) with Your Own Prompt
You can directly call and include major LLMs, the followings, in your Agent Flow:
OpenAI (GPT-4)
Azure OpenAI
DeepSeek
Gemini
Customize the prompt, provide system instructions, and retrieve structured or conversational responses. Ideal for advanced users who want full control over AI logic.
🔹 Send Email / Response
Using email and communication tools such as Front App, Zendesk, Microsoft Outlook, Salesforce ServiceCloud, Slack, Teams and many other tools available in the Agent Builder you can use these actions to send emails, messages, or notifications automatically based on the result of processing in the AI Agent in the earlier steps.
Major integrations include::
Microsoft Outlook
Front App
Zendesk
Freshdesk
Salesforce
Microsoft Teams
…
You can tag conversations, add comments, or send templated responses.
🔹 Call Actions from Third-Party Tools
Seamlessly connect with external systems to fetch, update, or act on external data. See the guide on Connecting External Tool and Data for more details.
Supported integrations include:
Storage platforms: Dropbox, Box, SharePoint, Amazon S3
Databases: MySQL, PostgreSQL, MongoDB, etc.
ERPs and CRMs: SAP Business One, NetSuite, Salesforce, and more
…
How to configure
From the list of tools, look for the desired tool by typing its name and select the appropriate actions and configure them as per required inputs.
🔷 Common Tools and Actions in Agent Builder
When creating your AI Agent, you can build powerful automations by combining one or more of the following Action Types in your Agent Flow. Each action performs a distinct task—ranging from classification to advanced AI generation.
🔹 Classify Message
Automatically classify incoming text (emails, tickets, chat messages, or text) based on content, intent, topic, tone, or urgency.
How to configure
Add a Conversation Skills tool to your agent flow, then select “Classify Message” action and configure.
🔹 Classify Attachment or File
Use AI to classify documents such as PDFs, Word files, Excel sheets, or images based on content or metadata.
How to configure
Add a Document Skills tool to your agent flow, then select “Classify Attachment” action and configure.
🔹 Extract Text or Structured Data
You can extract the text of the files, or extract specific information from a file.
Extract text from files (attachments)
You can extract the text inside a file using Optical Character Recognition (OCR) technology and use it in the following steps. For example, you can insert the text of a file as a comment in Front App, Zendesk for Salesforce ServiceCloud so your human agents can access their text readily without opening the attached files one by one.
How to Configure
Add Document Skills Tool to your agent flow, and select “PDF/Image to Text (OCR)” action and configure it.
Extract structured data (field-values) from messages (text) and files
You may include steps to extract structured data from unstructured input such as:
Email body (available via ‘Extract Information’ action from Conversation Skills tool)
Documents like PDFs, images, Excel files (available via ‘Extract Information’ from’ Document Skills tool)
How to Configure
Add the Extract Information action from the appropriate skill type (Conversation Skills or Document Skills).
Choose the source (e.g., File URL or email body).
Select the file type: PDF, image, Excel, etc. (if extracting structured data from file) or text body (if extracting information from email or a text content)
Choose a prebuilt extractor (e.g., Standard Invoice Extractor) or one of your own custom extractors (you may create a custom extractor by following this guide in advance).
Test the extractor to ensure expected output.
Output: Structured data in a JSON object with the extracted fields including header fields and line item fields.
🔹 Entity Extraction
Identify and extract named entities such as names, dates, customer IDs, product codes, and more from text input. Supported entity types include:
Person Name
Date of Birth
Address
Email
Phone Number
Passport Number
Driver License Number
US Tax Identification Number (TIN)
US Social Security Number (SSN)
UK Unique Taxpayer Reference (UTR)
UK National Insurance Number (NINO)
UK National Health Service Number
Bank Account Number
International Banking Account Number (IBAN)
Canada Health Number
Canada Social Insurance Number
Vehicle Identification Number (VIN)
How to Configure
Add Conversation Skills or Document Skills tool to your agent flow (depending on whether processing message/text or files) and select ‘Find Mentions’ action and configure it. You can choose one or more available entity types.
Also, if you are looking for specific keywords, you can enter them in the ‘Keywords’ field.
🔹 Translation
Translate messages or document content into your desired language. Supports multilingual workflows.
How to Configure
Add Conversation Skills or Document Skills tool to your agent flow (depending on whether processing message/text or files) and select ‘Translate’ action and configure it.
🔹 Search, Query and Matching
In many situations, you may need to find specific values or query terms or values from the content of spreadsheets or UpBrains AI Collections (that are built on top of OpenSearch indexes as the storage for our RAG (Retrieval-Augmented Generation) agents.
You can include Search, Query and Matching capabilities in your agent flows using one of the following actions:
Partial Match (action available through Smart Classifier tool) – operates on the output from Google Sheet, Microsoft Excel Online or CSV content and offers an AI-powered partial matching capability to find values with approximate matching capability. It performs approximate or fuzzy matching within the data pulled from spreadsheets or data tables (e.g., match customer names or product SKUs with typos).
Query (action available through OpenSearch tool) – This offers a comprehensive, AI-powered search and query over an OpenSearch index (called an UpBrains AI Data Collection).
These actions are ideal for customer information lookup, product catalog search, or referencing predefined knowledge based, depending on where your data is located.
Note: you can create Data Collections in UpBrains AI and sync it with external data sources. The Data Collections will be available to AI agents to use as their knowledge base or perform search and query.
🔹 Transform / Map
Apply custom logic or JSON mapping and manipulation formulas to input data using programs, rules or scripting to transform some JSON data into the input of another flow step or an external application.
Here are key actions that you can use to transform or map data:
JSONata (advanced JSON manipulation) – This action is available from JSON tool sn support JSONata (https://docs.jsonata.org/overview.html) – a query and transformation language for JSONs – on input that can be referenced as Input_1, Input_2, etc in the Query input.
Static object building using “Build JSON” action from JSON tool by defining a set of key and values that will be formed into a JSON dictionary.
There are a number of other tools for operating and transforming different types of data, each offering many actions:
CSV
JSON
XML
Date Helper
Text Helper
File Helper
Image Helper
🔹 Draft a Response with AI
Through these actions, AI agents draft personalized replies using AI. You can configure these actions and provide context, tone, and goals to guide the response generation process.
Use this action in combination with classification, extraction results as well as Knowledge Bases as tools to AI agents for highly relevant answers.
🔹 Generate Text, Image or Voice with AI
Use generative AI to create:
Text summaries, intros, or explanations
Images (e.g., banners, thumbnails)
Voice messages (TTS)
This is useful for marketing automation, support scenarios, or creative flows.
How to Configure
You can use the “Open AI” tool, and use custom prompts and use different models available from Open AI in your Agent flow.
🔹 Call Large Language Models (LLMs) with Your Own Prompt
You can directly call and include major LLMs, the followings, in your Agent Flow:
OpenAI (GPT-4)
Azure OpenAI
DeepSeek
Gemini
Customize the prompt, provide system instructions, and retrieve structured or conversational responses. Ideal for advanced users who want full control over AI logic.
🔹 Send Email / Response
Using email and communication tools such as Front App, Zendesk, Microsoft Outlook, Salesforce ServiceCloud, Slack, Teams and many other tools available in the Agent Builder you can use these actions to send emails, messages, or notifications automatically based on the result of processing in the AI Agent in the earlier steps.
Major integrations include::
Microsoft Outlook
Front App
Zendesk
Freshdesk
Salesforce
Microsoft Teams
…
You can tag conversations, add comments, or send templated responses.
🔹 Call Actions from Third-Party Tools
Seamlessly connect with external systems to fetch, update, or act on external data. See the guide on Connecting External Tool and Data for more details.
Supported integrations include:
Storage platforms: Dropbox, Box, SharePoint, Amazon S3
Databases: MySQL, PostgreSQL, MongoDB, etc.
ERPs and CRMs: SAP Business One, NetSuite, Salesforce, and more
…
How to configure
From the list of tools, look for the desired tool by typing its name and select the appropriate actions and configure them as per required inputs.
🔷 Common Tools and Actions in Agent Builder
When creating your AI Agent, you can build powerful automations by combining one or more of the following Action Types in your Agent Flow. Each action performs a distinct task—ranging from classification to advanced AI generation.
🔹 Classify Message
Automatically classify incoming text (emails, tickets, chat messages, or text) based on content, intent, topic, tone, or urgency.
How to configure
Add a Conversation Skills tool to your agent flow, then select “Classify Message” action and configure.
🔹 Classify Attachment or File
Use AI to classify documents such as PDFs, Word files, Excel sheets, or images based on content or metadata.
How to configure
Add a Document Skills tool to your agent flow, then select “Classify Attachment” action and configure.
🔹 Extract Text or Structured Data
You can extract the text of the files, or extract specific information from a file.
Extract text from files (attachments)
You can extract the text inside a file using Optical Character Recognition (OCR) technology and use it in the following steps. For example, you can insert the text of a file as a comment in Front App, Zendesk for Salesforce ServiceCloud so your human agents can access their text readily without opening the attached files one by one.
How to Configure
Add Document Skills Tool to your agent flow, and select “PDF/Image to Text (OCR)” action and configure it.
Extract structured data (field-values) from messages (text) and files
You may include steps to extract structured data from unstructured input such as:
Email body (available via ‘Extract Information’ action from Conversation Skills tool)
Documents like PDFs, images, Excel files (available via ‘Extract Information’ from’ Document Skills tool)
How to Configure
Add the Extract Information action from the appropriate skill type (Conversation Skills or Document Skills).
Choose the source (e.g., File URL or email body).
Select the file type: PDF, image, Excel, etc. (if extracting structured data from file) or text body (if extracting information from email or a text content)
Choose a prebuilt extractor (e.g., Standard Invoice Extractor) or one of your own custom extractors (you may create a custom extractor by following this guide in advance).
Test the extractor to ensure expected output.
Output: Structured data in a JSON object with the extracted fields including header fields and line item fields.
🔹 Entity Extraction
Identify and extract named entities such as names, dates, customer IDs, product codes, and more from text input. Supported entity types include:
Person Name
Date of Birth
Address
Email
Phone Number
Passport Number
Driver License Number
US Tax Identification Number (TIN)
US Social Security Number (SSN)
UK Unique Taxpayer Reference (UTR)
UK National Insurance Number (NINO)
UK National Health Service Number
Bank Account Number
International Banking Account Number (IBAN)
Canada Health Number
Canada Social Insurance Number
Vehicle Identification Number (VIN)
How to Configure
Add Conversation Skills or Document Skills tool to your agent flow (depending on whether processing message/text or files) and select ‘Find Mentions’ action and configure it. You can choose one or more available entity types.
Also, if you are looking for specific keywords, you can enter them in the ‘Keywords’ field.
🔹 Translation
Translate messages or document content into your desired language. Supports multilingual workflows.
How to Configure
Add Conversation Skills or Document Skills tool to your agent flow (depending on whether processing message/text or files) and select ‘Translate’ action and configure it.
🔹 Search, Query and Matching
In many situations, you may need to find specific values or query terms or values from the content of spreadsheets or UpBrains AI Collections (that are built on top of OpenSearch indexes as the storage for our RAG (Retrieval-Augmented Generation) agents.
You can include Search, Query and Matching capabilities in your agent flows using one of the following actions:
Partial Match (action available through Smart Classifier tool) – operates on the output from Google Sheet, Microsoft Excel Online or CSV content and offers an AI-powered partial matching capability to find values with approximate matching capability. It performs approximate or fuzzy matching within the data pulled from spreadsheets or data tables (e.g., match customer names or product SKUs with typos).
Query (action available through OpenSearch tool) – This offers a comprehensive, AI-powered search and query over an OpenSearch index (called an UpBrains AI Data Collection).
These actions are ideal for customer information lookup, product catalog search, or referencing predefined knowledge based, depending on where your data is located.
Note: you can create Data Collections in UpBrains AI and sync it with external data sources. The Data Collections will be available to AI agents to use as their knowledge base or perform search and query.
🔹 Transform / Map
Apply custom logic or JSON mapping and manipulation formulas to input data using programs, rules or scripting to transform some JSON data into the input of another flow step or an external application.
Here are key actions that you can use to transform or map data:
JSONata (advanced JSON manipulation) – This action is available from JSON tool sn support JSONata (https://docs.jsonata.org/overview.html) – a query and transformation language for JSONs – on input that can be referenced as Input_1, Input_2, etc in the Query input.
Static object building using “Build JSON” action from JSON tool by defining a set of key and values that will be formed into a JSON dictionary.
There are a number of other tools for operating and transforming different types of data, each offering many actions:
CSV
JSON
XML
Date Helper
Text Helper
File Helper
Image Helper
🔹 Draft a Response with AI
Through these actions, AI agents draft personalized replies using AI. You can configure these actions and provide context, tone, and goals to guide the response generation process.
Use this action in combination with classification, extraction results as well as Knowledge Bases as tools to AI agents for highly relevant answers.
🔹 Generate Text, Image or Voice with AI
Use generative AI to create:
Text summaries, intros, or explanations
Images (e.g., banners, thumbnails)
Voice messages (TTS)
This is useful for marketing automation, support scenarios, or creative flows.
How to Configure
You can use the “Open AI” tool, and use custom prompts and use different models available from Open AI in your Agent flow.
🔹 Call Large Language Models (LLMs) with Your Own Prompt
You can directly call and include major LLMs, the followings, in your Agent Flow:
OpenAI (GPT-4)
Azure OpenAI
DeepSeek
Gemini
Customize the prompt, provide system instructions, and retrieve structured or conversational responses. Ideal for advanced users who want full control over AI logic.
🔹 Send Email / Response
Using email and communication tools such as Front App, Zendesk, Microsoft Outlook, Salesforce ServiceCloud, Slack, Teams and many other tools available in the Agent Builder you can use these actions to send emails, messages, or notifications automatically based on the result of processing in the AI Agent in the earlier steps.
Major integrations include::
Microsoft Outlook
Front App
Zendesk
Freshdesk
Salesforce
Microsoft Teams
…
You can tag conversations, add comments, or send templated responses.
🔹 Call Actions from Third-Party Tools
Seamlessly connect with external systems to fetch, update, or act on external data. See the guide on Connecting External Tool and Data for more details.
Supported integrations include:
Storage platforms: Dropbox, Box, SharePoint, Amazon S3
Databases: MySQL, PostgreSQL, MongoDB, etc.
ERPs and CRMs: SAP Business One, NetSuite, Salesforce, and more
…
How to configure
From the list of tools, look for the desired tool by typing its name and select the appropriate actions and configure them as per required inputs.
Connecting External Tools & Data