Appearance
Product Updates: August 28, 2025
Overview
This upgrade focuses on the product concept of "smarter, more efficient, and more flexible" introducing a series of innovative features and optimizations. It fully covers core modules such as AI Agent, Voice, Live Chat, Ticketing , and Knowledge Center, helping enterprises achieve a comprehensive improvement in service experience and operational efficiency.
Deep Empowerment by AI, Driving a New Paradigm in Customer Service
AI Live Chat: Added AI Live Chat assistant, which supports automatic answer generation, assisted Q&A, and knowledge base invocation; Added AI conversation summary function, which automatically extracts key information such as customer issues and resolution progress, reducing manual review time.
Intelligent Parameter Assignment: By passing in customer information, preset fields are automatically filled in, eliminating the need for customers to repeatedly answer the same questions, achieving an accurate and smooth service experience.
Enhanced Conversation Flow: Supports transferring to designated skill groups, proactively ending conversations based on customer semantics, and optimizing conversation timeout rules to improve efficiency in hotline service scenarios.
A New Experience of Global Connection, Smoother Communication Without Boundaries
Live Chat: Upgrade real-time translation feature to support automatic customer language detection and instant translation.
App Center: Rich channel access, supports Lazada e-commerce message integration and App Store / Google Play review integration. Allows centralized viewing of orders or reviews on the workbench and replying to customers.
Multilingual and Localization: Support various translation services, customize proper nouns, ensure accurate translation of specific terms (such as brand names, technical terms, etc.), and adapt to multilingual scenarios.
Knowledge Flow Upgraded, Operations Easier and More Efficient
Task Flow Optimization: Add new LLM task nodes, LLM judgment nodes, global memory mechanisms, and ticket creation nodes to enhance the flexibility and contextual coherence of multi-turn dialogue tasks. Optimize the multi-variable collection function and simplify the configuration of complex business processes.
Knowledge Management Optimization: Optimize customer issue learning, with more accurate clustering; Add a new LLM Action Management module to support custom or system action calls.
Sobot is committed to empowering global enterprises with intelligent products, activating new service efficiency, and leading the new future of customer interaction.
Note:
=Addition
=Optimization
=Addition =OptimizationLive Chat
Real-time Translation for Agent Workbench
Feature Description
Add Automatic Translation configuration to the Agent Workbench.
The Agent Workbench automatically translates the customer's language.
Automatically translate the native language of the agent into the customer's language when sending.
Management Workstation Supports Configuring the Online Message Forwarding Interface Address
Feature Description
Chat - Message Webhook, add configuration for message forwarding address.
Online Statistics API
Feature Description
New online statistics API added, including chat statistics, customer statistics, agent statistics, skill group statistics, satisfaction statistics, and service summary statistics.
Pre-query Form API
Feature Description
Add a new pre-query form API, including the message retrieval interface and the message forwarding interface.
H5 Visitor Side Allows Customers to Download Videos
Feature Description
Add a download video button on the H5 page.
SDK Supports Getting Unread Messages and Offline Messages
Feature Description
The SDK supports obtaining the number of unread messages and offline messages for use when customers have not entered the consultation list.
Multilingual Configuration
Function Background
Can automatically identify the language used by consulting customers and reply in the corresponding language; reduces the cost of enterprise staffing.
Feature Description
Online support references public-multilingual content. When customers contact support, the system automatically sends configuration content in the customer's language.
This period's adapted features: Chat - Customer Assignment Strategy - Queue Setting, Chat - Chatting - Timeout Settings, Chat - End of Chat - End Setting.
AI Summary
Function Background
After the robot transfers the human agent A, human agent A needs to review the previous customer communications one by one.
After human agent A transfers to human agent B, human agent B needs to review the previous customer communications one by one.
The conversation has a lot of information. The customer service may forget or miss the conversation details after serving other customers and checking again.
Feature Description
This update supports generating conversation summaries when robot reception transfers to human agents, after customer service transfers, and when there is a large amount of session information. This improves the efficiency of customer service reception.
- Configuration: Enable AI Summary in the Agent Workbench.
Select the timing for generating session summaries
a. Automatically generated after robot transfers to agent.
b. Automatically generated after transferring to the agent.
c. Manual click triggered by the agent.
Generate key points
a. Customer question: Default to extracting and splitting multiple customer questions
b. Progress of resolution
c. Solution
d. Customization: You can fill in the information that needs to be collected for customization.
Fill in the example
a. Customer phone number: The phone number provided by the customer during the communication process
b. Customer order number: The order number provided by the customer during the communication process
Generate session summary
Note: The session summary can only be generated when the number of interaction rounds in the session information is greater than one.
AI Agent
The Agent Q&A Assistant Robot: AI Copilot
Function Background
When the agent receives clients through the workbench, it only supports active querying of the knowledge base to find relevant information in the knowledge base.
Feature Description
This AI Agent adds an AI copilot service assistant, which supports helping generate questions during Live Chat reception, improving agent efficiency.
Configuration
a. AI Agent: Configure AI Copilot
- Create AI Copilot
- Configure the business introduction, response style, and conversation guidance of the AI copilot. Bind the knowledge base and conduct Q&A debugging.
Note: The AI copilot can bind internal knowledge, internal/external knowledge for Q&A. The reception robot can only bind internal/external knowledge.
b. Live Chat: Bind AI Copilot
Enable and select the configured AI copilot.
Note: If the option to automatically generate answers is selected, answers will be automatically generated for customer questions regardless of whether the agent clicks on the AI copilot Q&A component or not.
Q&A: In the Agent Workbench, add an AI Copilot Q&A component to assist the agent with asking questions.
a. Question: Includes the following ways of asking
Agent asks questions proactively
Hover over customer issues and ask questions directly.
Check Auto-generate answers to automatically create answers based on customer questions.
b. When viewing the answer, you can check the knowledge snippet that this sentence comes from based on the hover number. Click the knowledge slice to open the corresponding knowledge details. After asking a question, you can send the answer directly or edit it in the input box before sending.
c. Attention:
During Q&A, answers are generated by filtering knowledge based on the customer's group and the agent representative's permissions.
a. Example:
Configuration: The knowledge center has two pieces of knowledge, VIP customer return policy, and regular customer return policy.
Online Consultation:
The agent does not have permission: When a VIP customer inquires about return policies, the agent has permissions to view only the return policies for regular customers. Therefore, the AI copilot cannot generate a corresponding VIP return policy answer.
The agent has permission: When a VIP customer inquires about return policies, the agent has permissions to view the VIP return policy. The AI copilot will generate a corresponding VIP return policy answer.
Statistics: Add AI Copilot data to AI Agent, and support viewing statistics of the AI Copilot.
a. In the assistant data, you can mainly view the number of assistant business chats, the number of assistant questions, the number of assistant agents, the number of answer usages, and the answer usage rate. Enterprise managers can comprehensively check the auxiliary effect of the AI copilot based on the above indicators.
b. Note: The number of assisted business chats refers to the number of online conversations and message counts assisted by the AI copilot in the online scenario.
Operation Optimization: Customer Question Learning
a. Enterprises can view relevant data in the customer question learning, including the number of AI copilot answer usages, answer usage rate, number of assistant business chats, and number of assistant agents. They can also identify which issues have a high number of occurrences but low response usage, and optimize further.
Reception records: You can view the details of the issues in the reception records.
Bot to Human: Supports Transferring to a Specific Skill Group of Live Chat
Function Background
Chatbot supports keyword and semantic transfer to a specified group. AI Agent do not support this feature currently.
Feature Description
In custom transfer scenarios, it supports configuring transfers to the Chatbot skill group.
When a customer's question matches a configured custom scenario for transferring to a human agent, and the scenario specifies the live skill group to transfer to, the system will transfer the customer to the corresponding skill group.
Note:
During online agent routing, the priority is: designated by the agent > transfer by the AI Agent > routing configuration in the Live Chat.
In the Live Chat, there is no skill group configured in the custom scenario. It hits the transfer to human agent, but cannot be transferred to the specified group.
Only one skill group can be filled in a custom scenario. The filling format is {Skill Group: XXX}.
Bot Ends Chat: Supports Ending Chat Based on Customer Feedback
Function Background
In the voice scenario, when the customer indicates on the phone that they have no other questions, the system currently cannot end the session based on the customer's semantics.
Feature Description
In the conversation guide, describe the situation where {End Conversation} is supported.
When a customer issue matches a configured session guide, and the scenario specifies ending the session, actively end the session.
Attention:
Only one {End Conversation} can be filled in a session guide.
Live and voice both support ending the session.
Bot Ends Chat: Rule Adds Second Unit
Function Background
In voice scenarios, if a customer does not respond for a long time, the system currently waits at least 1 minute before hanging up the call actively. This causes a waste of call charges.
Feature Description
Increase the session to end automatically after XX seconds.
Bot Adds Filipino Language
Feature Description
The AI Agent adds Filipino language.
Bot Variables Support Custom Parameter Assignment
Function Background
After a company passes business information through custom parameters such as params and multiParams, it hopes to store the information in variables for use by the AI Agent. For example: When a customer contacts the company and the company passes the parameter "Purchased model: X10", the robot can directly obtain "The current customer has purchased model X10" without asking the customer.
Feature Description
When the customer passes multiparams/params={"phone":"ios","type":"X10"}, and the customer has already configured two variables with the codes "phone" and "type" in AI Agent - Variable Management, the robot can parse the enterprise parameters during the actual conversation and use the parameter values in the current conversation.
Knowledge Operation: Customer Question Learning Optimization
Function Background
Enterprises encounter the following issues when using the customer question learning feature:
When enterprises query customer questions from the recent period, they find multiple identical questions.
Customer question is not fully displayed.
Feature Description
Optimize the logic for clustering customer iquestions. Each time clustering occurs, query the existing category names of the company to avoid having multiple identical questions.
Customer clustering question shows the rewritten question by the robot, completing the subject-verb-object information.
Knowledge Center
Task Workflow Node: LLM Task Node (Launch Planned for September 11th)
Function Background
In real customer service scenarios, multi-turn dialogues often face problems such as sudden topic shifts by the customer, irrelevant responses, or incomplete statements. These issues make it difficult for pre-set workflows to adapt flexibly, significantly impacting the completion rate of the workflow.
To enhance the task flow's ability to handle dynamic issues, we have introduced an LLM task node in the design. This node can carry out autonomous multi-round dialogues based on task objectives under the guidance of preset steps, and it can call external tools to assist in task progress. Its core value lies in responding to customer input in real time, dynamically adjusting execution strategies based on dialogue progress and tool feedback, until the task is completed. This mechanism significantly improves adaptability to uncertain dialogue scenarios and effectively ensures the smooth completion of complex task flows.
Feature Description
Example Scenario
In after-sales service in the smart manufacturing industry, customers often need to locate nearby repair or service locations for their products or equipment. Or, while searching for a location, they may suddenly ask questions about warranty policies or other issues. By configuring an LLM task node, the bot can autonomously call interfaces, assign variables, and search the knowledge base at this node to implement this scenario:
Task Objective: Query service location information for the customer.
Task Description:
When a customer requests "looking for an offline service location" but does not specify a location, guide the customer to provide the necessary information (such as their city or factory address);
Call the store/location query interface to retrieve official service locations in the area;
Return the store address, contact information, and business hours to the customer, and populate the variable "location to be booked" with the name of the location the customer wishes to book;
If there is no location in the area, ask the customer whether they would like mail-in repair service;
If the customer inquires about warranty policies or complaints about a location, provide a professional response by querying knowledge.
Knowledge Base Scope: Store and service location knowledge base.
End condition: When the customer explicitly indicates that he wants to go to a specific service outlet, the current LLM node ends.
Functional Configuration
Task Objective: Define the goal that the current node needs to achieve.
Task Steps: Define how the large model of this node executes the task. When writing task step prompts, you can reference actions and variables, which will prompt the large model to call external interfaces or store data at specific times.
Knowledge base scope: When a knowledge-based question is identified, the LLM task node will call the knowledge base to answer. You can define the scope of the knowledge base query. If you actively reference the Search Knowledge action in the task steps, it will guide the large model to directly query the knowledge base, and the query scope will also be within the configured knowledge base range.
Ending Conditions: You can customize the ending conditions. For example, "When the customer considers the question resolved," end the current LLM task node and follow the branch to the next node. The task will also be forcibly ended when the maximum number of question-answer rounds is exceeded.
Action Management Module (Launch Planned forSeptember 11th)
Function Background
Within the LLM task node, the LLM can reference "actions" to autonomously invoke external tools. Actions can be created in the Knowledge Center - Action Management module. There are two types of actions:
Custom Action: An action encapsulated using your own Interface, such as searching data within your system.
System action: Actions encapsulated using Sobot internal Interface, such as Creat Ticket, Remind Ticket, SMS Sending, etc.
Feature Description
Action Name: Define the name of the current action
Action Code: The code for the current action must have actual meaning, not random characters.
Action description: Define the purpose of the current action and possible return results to help the large model confirm when to invoke the action.
Interface request: After referencing the interface, fill in the descriptions of the request parameters and return parameters. This helps the large model collect the request parameters and analyze and understand the action return results.
Task Workflow Node: Create Ticket Node
Feature Background
After customers communicate with the robot, if there are business issues such as complaints, the robot cannot help customers create tickets.
Feature Description
Add a ticket creation node in the task flow. When the dialog process reaches this node, the bot will initiate the ticket creation action.
Select the work order category, automatically link to the corresponding work order template, and retrieve the work order fields associated with that template.
Title, Problem Description: Automatically generated after the large model summarizes historical conversations
Custom Field: Populate Bot Variables
Storage action information: Optional. You can store the number or ID of the created ticket for follow-up conversations.
Create Result Branch: Each create ticket node will have a Creation Successful and Creation Failed branch. You can proceed with different follow-up actions based on the creation result.
Task Workflow Node: LLM Judgment Node
Function Background
When a company has the following scenarios, it may need to use the LLM judgment node for judgment by a large model.
Customer intent judgment: Determine the customer's intention and whether they are willing to provide more information;
Business type judgment: Determine whether the customer's conversation content belongs to sale or product, in order to guide to different branches.
Feature Description
The enterprise describes each branch through semantics. The large model will analyze historical conversations at this node and make different branch decisions.
The Bot Adds Global Memory (Launch Planned forSeptember 11th)
Function Background
When a single conversation triggers multiple processes, due to excessive dialogue turns exceeding the context backtracking limit, the bot may forget "what it has already done" and "what information it has collected," leading to a false sense of memory loss in the conversation—forgetting what was just said. For example, after a customer has selected an order and checked the logistics progress based on that order, if the customer then requests an invoice for the same order, the bot often fails to automatically reuse the order number and instead repeats the inquiry, significantly impacting both experience and efficiency.
Feature Description
This release introduces a "memory" mechanism, supporting session-level persistent memory at the variable, process, and task levels. It enhances the robot's context awareness and coherence in complex multi-turn conversations.
Variable memory mechanism
Global variable: The value of the variable exists across processes throughout the session. For example: After the robot collects the global variable "order number," it will directly use that order number throughout the session and will not collect a new order number.
Process Variable: The value of the variable exists only within the current task process. After the process ends, the value of the variable iscleared. For example: The process variable "Is it resolved?" Each time a task process runs, the robot will ask the customer, "Has your question been resolved?"
Workflow and Task Memory
In addition to variable memory, this update also expands the process memory and LLM task memory capabilities. The system can automatically record triggered processes and task statuses, enabling more accurate conversation guidance.
Task Workflow Node: Conversation Collection Node Supports Collecting Multiple Variables
Function Background
The Conversation collection node can only collect one variable. When the enterprise workflow is complex and needs to collect 10+ variables, the configuration process becomes complicated.
Feature Description
The Conversation collection node supports collecting up to 5 variables.
The first message supports sending fixed scripts. Afterwards, the AI automatically adjusts and generates scripts based on the collected information.
Note: In voice scenarios, if one node collects multiple variables and the response time is longer, we still recommend that one node collect only one variable.
Task Workflow Node: Interface Request Node, Message Card Supports Re-sending After Collection Failure
Function Background
After the bot sends a message card, if the customer does not select the card but replies with some unrelated messages, it will directly move to the next node without triggering clarification or repeating the question.
Feature Description
The interface request node adds Action Rules, which can be configured to resend the card when information is not collected.
Knowledge Center Adds Filipino Language
Feature Description
The knowledge base and task flow in the Knowledge Center are expanded to include Filipino.
Voice
Optimization of Seat Allocation Strategy within the Call Skill Group
Feature Description
In the location of Voice - Setting - Call Skill Group - Edit Skill Group -In-Group Agent Assignment -Basic Assignment Strategy, the Basic Assignment Strategy option of"Preferentialliy assign agents with min answered calls in the day" follows the enterprise's primary time zone for the "same day" time, as shown in the figure.
IVR Process Adds SMS Node
Function Background
Enterprises hope to add SMS nodes during the IVR process, for sending SMS in scenarios like self-service inquiries.
Feature Description
At the location of Voice - Setting - IVR Flow - Edit - Advanced Node, add SMS node.
At the location of Voice - Record - Call Record- Call Detail - Link Record - SMS Record, add downstream SMS records for call records.
IVR Track Analysis Adds Data Statistics for SMS Node
Function Background
For the SMS node in the IVR Flow, count the number of incoming/outgoing calls to analyze the usage efficiency of this node.
Feature Description
In Voice - Statistics - IVR Track Analysis, add data statistical indicators for the new SMS node.
Inbound Voicebot (General Version), Character Limit Optimization for Some Script Input Boxes
Feature Description
In the Main Workflow, Knowledge Base, multi-round dialogue, advanced settings, and other areas involving custom script content in Bot Management, the character limit for input has been adjusted to 1500 characters.
Inbound Voicebot (General Version), Some Script Nodes Support Interruption Function
Feature description
Main Workflow & multi-round process information collection nodes now have a new call interruption function. This supports interruption by key press or voice.
Inbound Voicebot (General Version), Flow Nodes Support Operations Such as Copy and Paste for Single and Multiple Template Nodes
Function Background
Some companies have various service categories under the same type of business. This may lead to repeated operations for a large number of businesses when configuring robot processes.
Feature Description
To improve the efficiency of business process configuration for business managers, support operations such as copy and paste for single or multiple nodes. (Support shortcut key copy and paste operations)
OutboundVoicebot (General Version)
Character Limit Optimization for Some Script Input Boxes
Feature Description
In the Main Flow, Knowledge Base, multi-round, advanced settings, and other areas involving custom script content in Robot Management, the character limit for input has been adjusted to 1500 characters.
Some Script Nodes Support Interruption Function
Feature Description
Main Workflow & multi-round process Information Collection nodes now have a new call interruption function. This supports interruption by key press or voice.
Some Script Nodes Support Variable Import
Function Background
Some businesses need to verify or use existing information when calling customers. To support this, we have added the ability to import variables. This allows companies to define relevant variable data for different customer numbers and apply it to the call flow.
Feature Description
Outbound Voicebot now support creating and referencing imported variables.
Flow Nodes Support Operations Like Copy and Paste for Single and Multiple Nodes in Templates
Funtion Background
Some companies have various service classifications under the same type of business. This may lead to repeated operations for a large number of businesses when configuring robot processes.
Feature Description
To improve the efficiency of business process configuration for business managers, support operations such as copy and paste for single or multiple nodes. (Supports shortcut-based copy and paste operations)
Developer Documentation - Voice API - Voice V6 - Basic Management - Outbound Task Management, New API Interfaces Added
Feature Description
Developer Documentation - Voice Product API - Voice V6 - Basic Management - Outbound Task Management, the following API interface are added:
Add API interface for modifying task information
Add API interface for querying outbound task details
Add API interface for querying individual task detail list
Add API interface for querying details of a single task
Ticketing
The View's ReceivingAgent and Receiving Skill Group Conditions Support Multiple Selections
Feature Description
The conditions of the view, the Receiving Agent, and the Receiving Skill Group support multiple selections to support more complex scenarios.
Option to Record Internal Notes as First Response (Only after SLA Upgrade)
Feature Description
After checking the Internal Notes, both the Internal Notes and Public Reply will be recorded as the First Reply.
Note: This feature is only supported after upgrading to the latest SLA.
Admin Center
The Phone Charges of the Three Products - Voice for Sales V6, Voice and Outbound Voicebot (General Edition) - Have been Added to the Company Account
Feature Description
Admin Center - Company Account adds Voice for Sales V6, Voice, Outbound Voicebot (General Version) with new Billing Detail data (Customer call billing, Agent call billing), and adds Export operation to support summary export by billing department and agent.
Adjustment of the Admin Center Menu Structure
Function Background
To make the menu structure more organized, we have adjusted the menu structure for account management in this update.
Feature Description
Added a Localization menu, which includes Multilingual, Time zone, and Brand.
The content included in the brand consists of the enterprise information and workbench logo settings from the original Company Account.
Additionally, we expect that in early September, we will remove the Enterprise Info section from Account Management - Company Account and place the Company ID, or SYSNUM, under the Enterprise Name in Account Management - Company Account - Product Order Info."
In the October version, the menu structure of the management center will be further optimized.
Dropdown List and Related Support Excel Import
Feature Description
Problem solved: For scenarios that require setting a large number of list options, such as chain stores nationwide;
Note: Due to the large amount of data making sorting inconvenient, when the drop-down list has more than 100 options, the page's sorting will be disabled. You can use the Excel import method to complete the sorting.
Support for Large Model Translation and Proper Noun Settings
Feature Description
Amazon uses machine translation, which is fast but performs poorly for minority languages and dialects;
Large model translation can understand context well and has a better understanding of newly emerged internet slang, but the speed is slower.
For some industry-specific terms and names, companies can upload a name list to specify the translation content. For example, they can treat "Sobot" as a proper noun and specify that it should be translated into English as "Sobot". This will prevent it from being translated incorrectly as "wisdom tooth".
This feature is suitable for translating Ticketing, Live Chat and app reviews.
CRM
Create Customer Component Supports Creating Company
Feature Description
When creating a customer in the Ticketing or on the Workbench, support for creating a company is available.
Applicable scenario: This feature is very suitable for TOB customer inbound consultation scenarios. If not needed, you can disable the company field in the CRM field settings.
Batch Import Customers Add Privacy Agreement
Feature Description
Before importing customers in Excel, you need to agree to the Privacy Agreement first.
Inspection
Chat Inspection Supports2-Level Evaluation Screening Condition
Feature Description
Inspection Statistics Support Counting Disabled and Deleted Agents
Feature Description
App Center
Support Pulling and Replying to App Reviews in App Store and Google Play
Feature Description
You can view and reply to App Store/Google Play reviews without leaving the Agent Workbench.
This feature is a paid function. Enterprises can install one application for free to try out. For pricing, please consult your customer service manager.
Start Integration
Take the App Store as an example:
In the App Center - App Store - App Integration, select App Store and fill in the required parameters (see the instruction document on the page for parameter acquisition) to complete the integration.
Handle App Reviews
After the first successful authorization, we will initially fetch some app data (6 months from App Store and 7 days from Google Play). After that, we will i ncrementally pull the latest data every 2 hours.
Enterprises can view and reply to comment information on the Agent WorkbenchFor explanations of each term and metric, please refer to the product documentation.
Additionally, we also provide convenient tools such as AI rewriting, translation assistant, and quick replies to improve the work efficiency of enterprises.
- We provide detailed data analysis reports to reflect the performance of enterprise applications and the agents.
Support for Connecting with the Lazada Channel, Enabling Customer Message Reception
Feature Description
Agent can directly reply to Lazada buyer messages in the Agent Workbench and view the corresponding order details.
Application Integration
In the App Center - Marketplaces - Marketplaces App, select Lazada. Follow the introduction and instructions in the app's documentation to complete the integration process.
After the integration is completed, the Agent Workbench will be able to handle private messages from Lazada customers and display order information.
Broadcast
WhatsApp Sending Record Interface Supports Single-page Query of 1000 Records
Feature Description
Query sending records API, supports querying 1000 data entries at a time.
Facebook Messenger Supports Horizontal Carousel Messages
Feature Description
Facebook Messenger supports horizontal carousel messages, which can send up to 10 templates at once, providing a more detailed and comprehensive introduction of products.
Added Customer Nickname Field to the Sending Record List
Feature Description
Customer Group Supports the Selection Condition of Customer Language
Feature Description
Support businesses in sending template messages in different languages to customers who speak different languages.
SMS Center
Developer Documentation - Message Class API - International SMS, New Query SMS Template Interface
PublicServices
Reminder Added for Fogin Gasswords not Changed in 3 Months
Feature Description
