Product Updates: October 29, 2026

Overview

This release follows our product principle of "smarter, more efficient, more secure." It focuses on delivering LLM-powered Agents as a standalone product and on deepening our contact center, live chat and ticketing capabilities. Through product versioning, a closed loop for your own channels, and AI embedded across service workflows, this release improves onboarding efficiency and day-to-day operations for overseas customers, helping you build a more stable, efficient and scalable customer service system.

I. Standalone product delivery and channel expansion — ready to use out of the box

• Versions and subscription: the product is split by capability into three standard editions — Team, Professional and Enterprise — plus a Custom edition, available on both the Singapore and North America sites.

• Sign-up and sign-in: register and sign in with email or a Google account, with automatic routing to the nearest site based on the customer's IP.

• Your own channel loop: new Channel access, Visitor settings and Reception plan modules let you deploy the agent directly to your own website and app, plus a new HarmonyOS SDK.

II. Stronger Agents — more accurate answers, more flexible configuration

• Model and file parsing: a new model selection option, plus parsing of PDF, Word and Excel files, so answers fit your business scenarios more closely.

• Extended resource configuration: knowledge bases accept Markdown files, MCP supports OAuth authorization, and action parameters support multi-level nesting.

• A complete operations loop: customer question clustering is upgraded to scenario-based categorization, evaluation datasets support languages, evaluation results can be exported, and agent working hours can now be configured.

III. AI-powered contact center — more efficient agent service

• AI assistance: new AI transfer summary and AI task summary let agents grasp the preceding conversation immediately and generate task summaries in one click.

• Better recognition accuracy: new replacement word management, plus new recognition channels and recognition modes for real-time transcription, reduce homophone errors.

• Automatic ticket creation: call triggers can create tickets from a ticket template, and ticket content can be generated by AI from the call.

IV. Rounded-out messaging, ticketing and administration — smoother and better governed

• Visitor and agent efficiency: visitors can send images, files and text together, desktop sites support voice messages, and image-and-text messages can be translated; chats can be searched by customer name, and a Max Response Time metric has been added.

• Ticketing improvements: email tickets automatically send a satisfaction survey email, integration fields in the ticket plug-in can be passed via token to prevent leakage and tampering, and the submission list and ticket plug-in now offer Persian, Khmer and Hindi.

• Permissions and account management: a new Batch Transfer role permission, quality monitoring filters that include disabled and deleted accounts, self-service company closure for super administrators, and account deletion in the Agent App.

Sobot is committed to continuous innovation and product iteration, empowering businesses worldwide to build a new paradigm of intelligent, efficient and professional customer service in the digital era.

Note:  =Additions  =Optimization

Agents

Edition split: Team, Professional and Enterprise standard editions

Functional Background

The agent product previously had no editions — every customer saw the same set of capabilities, which left no basis for pricing self-service online subscriptions. This release splits the product by capability into three standard editions plus one custom edition. Each edition opens a different set of modules and quantity limits, and both the Singapore and North America sites are supported.

Features Description

Editions: Team, Professional and Enterprise standard editions, plus a Custom edition.

Key capability differences:

Capability	Team	Professional	Enterprise
Number of agents	1	10	100
Third-party bots, Skills, Actions, MCP	—	✓	✓
Channel API access	—	✓	✓
Audience rules, answer analytics	—	✓	✓
Workflows, memory, variables	—	—	✓
Knowledge from chats, knowledge tags, synonyms	—	—	✓
Evaluation	—	—	✓
Customer question clustering	—	—	✓
Satisfaction survey templates	—	—	✓
Workbench evaluation mode	Single-item test in debug panel	Single-item test in debug panel	✓
Visitor client style, FAQs, quick menus, reception plans	1	15	Multiple
Role permissions	Basic roles	Basic roles	Basic roles

See also: https://sobotwiki.yuque.com/fre92s/cf_records/kgph877z3kipg470#kgb1

Sign-up and sign-in with a Google account

Functional Background

Overseas customers commonly use Google accounts. Google sign-up and sign-in have been added to improve the registration and login experience.

Features Description

Entry point: Sobot Agents website → Free Trial or Login.

• Sign-up methods: email + password, or Continue with Google. A company tenant is created automatically on successful registration.
• Regional routing: customers are routed by IP, with North America as the default. Southeast Asia, the Middle East, Hong Kong, Macao and Taiwan are routed to Singapore; everywhere else goes to North America.

Console: Notification center and Task center entry points

Functional Background

The console had no unified notification entry point, and the results of asynchronous operations such as bulk import and export were not collected anywhere, so users had to go back to the originating page and check repeatedly.

Features Description

Entry point: the right-hand side of the console's top toolbar.

• Notification center: a bell icon with a red dot in the top-right corner when there are unread messages. Clicking it opens a drawer from the right.
• Task center: a single place showing the results of asynchronous tasks such as bulk import and export.

Reception operations: Channel access

Functional Background

The agent used to be a built-in Sobot capability and could only be deployed on Sobot's own channels. Now that it is a standalone product, customers need to connect the agent to their own websites and apps, so the agent side needs its own channel management.

Features Description

Entry point: Reception operations → Channel access.

• Channel integration settings: multiple channels can be added; this release covers PC, H5 and SDK, with Sobot as the default channel. Each channel is configured independently, and once connected, reception plans and visitor-client touchpoints are bound per channel.
• Channel security settings: configure the access security policy for a channel.
• Differentiated display: when only the Agent product is purchased, the Channel security settings and Field mapping tabs are hidden and the organization scope defaults to all. Adding live chat opens everything up.

Reception operations: Visitor settings

Functional Background

Visitor client styles and touchpoints used to live under Sobot Live Chat settings and service apps, which customers registering directly on the website could not use. They have now been moved to the agent side, so the agent's own visitor client is configured there.

Features Description

Entry point: Reception operations → Visitor settings.

• Visitor client style: configure how the visitor client looks. When only the Agent product is purchased, customer information display, human-reception guidance text and message read status are hidden.
• FAQs: configure the list of questions recommended to visitors when a chat starts; the default timing is "when the visitor enters the bot chat." Once configured in a channel, the welcome message from bot management is migrated here.
• Quick menu: configure quick entry points on the visitor client; the default timing is "when the visitor enters the bot chat." When only the Agent product is purchased, the available actions exclude transfer-to-agent and leave-a-message.
• Audience grouping rules: an independent audience grouping module. Knowledge and knowledge augmentation both reference this audience library, and FAQs and quick menus are delivered by audience.

Reception operations: Reception plan

Functional Background

The reception plan decides who receives an incoming visitor, in which language, and under what conditions the chat is transferred to a human agent. This previously relied on Sobot's reception plan settings, with no entry point on the agent side.

Features Description

Entry point: Reception operations → Reception plan.

• Configuration items: trigger conditions, visitor client language, agent reception mode, bot reception, and transfer-to-agent rules.
• Differentiated display: when only the Agent product is purchased, the reception mode is fixed to "bot only" and the human-reception fields and transfer rules are hidden. Adding live chat opens up the human reception mode and all related fields.

HarmonyOS SDK: LLM-powered bot Q&A on the visitor client

Functional Background

The agent's own channels covered desktop websites, mobile websites and apps, but not HarmonyOS. Companies that want to offer support inside their own HarmonyOS app need a dedicated HarmonyOS SDK.

Features Description

Entry point: embed the HarmonyOS SDK in your own HarmonyOS app.

• Message types: text, tables, buttons, cards, images, video, hyperlinks and files; text supports rich-text formatting.
• Channel classification: HarmonyOS chats are counted under "APP" — no new channel enum is added. Chat data, answer data and reception records are all available.

Data: customer question clustering upgraded to scenario-based categorization

Functional Background

Today's customer question clustering uses an algorithm to aggregate raw customer questions into a flat list of topics. Topic names are generated by the algorithm and their granularity cannot be controlled, so operations teams simply have to accept them and cannot organize statistics around their own business taxonomy.

Some companies already have a well-structured business taxonomy and want the statistics to follow it. At the same time, companies using the feature for the first time often have no taxonomy ready, and companies that do have one keep encountering new questions that are not in their library. Those chats previously sank to the bottom of the list and could never be identified or captured.

This release unifies customer question clustering into scenario-based categorization, with the system deciding automatically between two cases: if a company has configured a scenario library, chats are categorized against it; if not, all chats are used to generate hierarchical new-scenario suggestions as a fallback.

Features Description

Entry point: Agents → Data → Customer question clustering.

Once scenario tagging is enabled, the workflow has four steps: configure the scenario library → review scenario data → review new-scenario suggestions → adopt suggestions back into the library. Questions not yet covered keep being discovered and captured, closing the loop.

• Step 1 — Enable scenario tagging: use the toggle at the top of Scenario configuration in the upper right of the Customer question clustering page. Once enabled, chats are tagged according to the scenario configuration as soon as they end; each chat receives exactly one scenario tag. Chats that match no scenario in the library go into the built-in "Other" uncovered pool, which is the source of new-scenario suggestions. Turning it off stops automatic tagging and billing and removes scenario statistics from the clustering page; existing tags and the scenario library are retained and reappear when it is turned back on.
• Step 2 — Configure the scenario library: entry point is Scenario configuration in the upper right of the Customer question clustering page. The library supports up to three levels; the first-level category is required, and each scenario can carry a "definition and key notes" field used as the basis for categorization. To rewrite history against the latest library, click Re-run and select a chat range in the dialog to re-tag immediately.
• Step 3 — Review scenario data: the page has two tabs, Scenario statistics and New scenario suggestions. Companies that have configured a scenario library land on Scenario statistics by default. If no library is configured, or a chat does not match any configured scenario description, new-scenario suggestions are generated. Suggestions that do not fit your business rules can be ignored — their chats are then marked "Other"; suggestions that do fit can be adopted into the library.
• Step 4 — Manual correction: every chat record supports manual correction of its scenario tag. The Edit tag entry point appears on the chat record detail page, the customer question detail page and the chat detail page.

Note: tagging a chat is a model call and costs 1 credit per chat, drawn from the unified Sobot AI credits pack. In the AI credits usage log the scenario is shown as "Customer question analysis." Manual tag correction involves no model call and consumes no credits. Re-running is billed again.

Evaluation: language added to evaluation datasets

Functional Background

Evaluation datasets previously contained only "question / reference answer" and "scenario description / customer persona," with no concept of language at all. The product targets overseas markets, where knowledge bases and chat records are naturally multilingual, so evaluation data needs to be organized by language. In multi-turn evaluations the language spoken by the user simulator was also uncontrolled, making it impossible to verify an agent's multi-turn performance in a given language.

Features Description

Entry point: Agents → Evaluation → Datasets.

• Set a language: a language can be set when generating data with AI, adding data manually, or importing in bulk.
• Multi-turn evaluation: the user side of the conversation follows the language set on the data.

Evaluation: better filtering and export

Functional Background

Evaluation reports could only be viewed online, and the overview lacked metrics showing how far bound resources were actually used and how often chats were transferred to agents. Item-level filtering was weak, making it hard for operations teams to locate a specific result among hundreds.

Features Description

Entry point: Agents → Evaluation → Plans → Plan details.

• New data cards in the report overview: Resource adoption rate — the share of items whose replies actually used a bound resource such as knowledge, a QA pair or a workflow (shown for single-turn dataset plans only); Handoff rate — the share of items where a transfer to a human agent was triggered (for multi-turn, a trigger in any turn counts for the item).
• Stronger item-level filtering: a question search box and a manual review filter have been added. Question search matches the question for single-turn and the scenario description for multi-turn; the manual review filter offers All / Reviewed / Not reviewed, defaulting to All.
• Export evaluation results: an Export button on the plan details page exports the plan's report as xlsx in one click. Completed and cancelled plans can be exported; once the export finishes, download it from the Task center.

Agent setting: model selection

Functional Background

Bot Q&A was previously fixed to a single underlying LLM, with no way for operations teams to switch models by business need. Different scenarios weigh "fast" and "accurate" differently: voice is a real-time conversation and is sensitive to latency, while messaging and ticketing can tolerate slightly slower responses in exchange for more accurate answers. One model cannot satisfy both.

Features Description

Entry point: Agents → Agent setting → Q&A capability → Model selection, placed after Capabilities & resources.

• Two options: the accurate model answers more precisely but responds slightly slower; the fast model responds more quickly. The accurate model is the default for both newly created and migrated bots.
• Scope: supported for both autonomous decision and knowledge-augmented modes of reception bots and agent copilots. Third-party bots only forward messages to an external server and do not use the platform model, so this setting is not shown for them.

Agent Q&A: image and file parsing configuration

Functional Background

When companies use an agent for scenarios such as placing orders or verifying information, customers often send a prepared PDF contract or an order screenshot directly. The existing capability only recognized images, and used a generic recognition prompt — it could neither read the contents of files such as PDFs nor extract the information a specific business actually needs. Customers had to repeat, in the chat, information already present in the file, which was slow, frustrating and prone to transcription errors.

Features Description

Entry point: Agents → Agent setting → Q&A capability → Image and file parsing.

• Enable toggle: on by default. When off, nothing is parsed and the model asks the customer to describe the file or image in text before replying.
• Dedicated prompt: required once enabled, up to 2,000 characters, pre-filled with a general default prompt that can be edited. Use it to describe your parsing requirements and the information to extract, such as customer number, product category or quantity.
• Supported formats: documents — PDF, Word, Excel; images — PNG, JPG, JPEG. Video, audio and archives are not yet supported.

Note: supported in both autonomous decision and knowledge-augmented modes.

Resource: knowledge bases accept Markdown files

Functional Background

Upload file in the knowledge base previously supported pdf, docx, xlsx and txt, but not Markdown. A great deal of content — technical documentation, help centers, READMEs — is written in md, so users had to convert it to Word or PDF first. That added a manual step and lost the heading hierarchy, which degraded the quality of the resulting chunks.

Features Description

Entry point: Agents → Resource → Knowledge → Knowledge bases → Knowledge base details → Add knowledge → File.

• Supported formats: .md and .markdown have been added to Upload file.

Resource \ Tools: multi-level action parameters

Functional Background

Request and response parameters for an action could only be configured as a flat, single level. The object nesting and array structures common in API payloads could not be described fully, so operations teams had to explain them in the parameter description field, which made model extraction and value resolution inaccurate.

Features Description

Entry point: Agents → Resource → Tools → Actions.

• Extended parameter types: request and response parameters can be String, Number, Integer, Boolean, Object or Array, defaulting to String.
• Nested structures: up to 5 levels of nesting; at the 5th level the parameter type is limited to String, Number, Integer or Boolean.

Resource \ Tools: MCP supports OAuth authorization

Functional Background

The first MCP release supported two authentication methods, None and Custom header. However, a growing number of remote MCP services accept only OAuth authorization and no longer issue static tokens, so those services could not be connected. Static tokens maintained through a custom header also have to be obtained manually and rotated periodically, and the service fails silently once they expire.

Features Description

Entry point: Agents → Resource → Tools → MCP.

• OAuth added as an authentication method: a third option alongside None and Custom header. When selected, the configuration area shows only the authorization status and an Authorize button.
• Zero-effort authorization: operations teams only need to enter the service address. After clicking Authorize, the system automatically performs authorization server discovery and client registration — no client details need to be filled in. A new window opens the third-party authorization page hosted by the service provider; once the user signs in and consents, they are returned automatically, the status changes to Authorized with the authorizing account and time shown, and connectivity verification and tool discovery run automatically.
• Error messages: cancelled authorization, timeouts, insufficient scope and credential exchange failures each produce a specific message, and the status returns to Not authorized.

Agent setting: agent working hours

Functional Background

Once transfer-to-agent was configured, the system had to attempt the transfer before checking whether any human agent was online. If none was, the customer was told after a wait that no agent was available and the transfer could not be completed — so the customer was first told they were being transferred and then told it was impossible.

Features Description

Agent working hours can now be configured. The transfer action can only be triggered during working hours; outside them, the customer is told that no agent is currently online and the bot continues to answer their questions.

Entry point: Agent setting → Conversation experience → Transfer to agent → new Agent working hours tab (empty by default).

Add configuration:

• Agent time zone: taken from the company time zone.
• Daily working hours: set the working hours for each day, for example Monday 8:00–12:00 and Monday 14:00–17:00.
• Holiday exceptions: set working hours for specific dates, or mark a specific date as closed all day — for example 2026-9-30 8:00–23:00, or 2026-10-1 closed all day.

Only one working-hours configuration can be added per time zone; daily hours and holiday exceptions are filled in together within that configuration.

Autonomous decision bots: the message shown outside working hours can be configured through the prompt.

Note: each customer is shown this notice at most once per day; afterwards no further notice is shown and the bot continues to handle the chat.

Agents Resource: knowledge parsing preserves automatic numbering in docx

Functional Background

Automatic numbering in docx files uploaded to the knowledge base was lost during parsing, leaving chunk content incomplete and degrading the quality of later knowledge base answers.

Features Description

Automatic numbering in docx files is now preserved during parsing.

Note: files uploaded previously must be re-uploaded and re-parsed before automatic numbering is preserved.

Agents Resource: variables can reference enum-type system variables

Functional Background

Bot configuration already supported referencing system variables (CRM customer fields), but not enum-type fields, which interrupted the configuration flow.

Features Description

Entry point: Agents → Resource → Variables → Add variable, with Reference system variable as the method.

When selecting a variable name from the dropdown, enum-value variables can now be added.

Agents voice settings: AI-generated filler script while waiting

Functional Background

When an agent takes a while to produce an answer, it first plays a filler script. Only a preset, randomly chosen fixed script was supported; the new AI-generated script makes the filler sound more natural.

Features Description

Entry point: Agents → Agent setting → Voice settings → Call settings → Play buffer script while waiting → new AI-generated script option.

• AI-generated script: the model generates a bridging line in real time from the current conversation context.
• Fallback script: if generation is too slow or fails, the fallback script is played instead, so the agent is never silent.

Note: each AI-generated script costs 1 credit (counted as AIGC usage).

Agents action management: DTMF validation for voice key collection

Functional Background

Key collection previously only supported ending input by digit count or by #, with no way to judge whether the input was correct. The new DTMF validation checks what the customer entered against a regular expression (for example, the quantity collected must be greater than 2).

Features Description

Entry point: Agents → Action management → Collect information by keypad → new DTMF validation.

• Regular expression: validates the keys entered by the customer.
• Error prompt: the message played when validation fails; the customer can interrupt it by pressing a key.

Agents debug panel: voice scenario testing

Functional Background

Once an agent was configured, only text chat could be tested. The new call test lets you verify speech recognition, playback, interruption and keypad input while still configuring.

Features Description

Entry point: Agents → Agent setting → Debug panel → new Call test; the former Text debug is renamed Chat test.

• Microphone: talk to the agent through your computer's microphone.
• Real phone call: enter a number to place an outbound call and test with a real phone.
• Text input: type text to simulate what the customer says, tested over the voice pipeline.

Note: real phone tests generate actual outbound call records and are billed normally. They can be identified in call records by Call scenario = Test call.

Agents voice settings: Channel Al model upgraded, hot words supported

Features Description

Channel Al has been upgraded to a new model.

• Hot words: Channel Al now supports hot words.

Agents conversation experience: in voice scenarios, the silence timer starts after the agent finishes speaking

Agents details: transfer-to-agent results in chat records now cover voice scenarios

Agents workflows: in voice scenarios, subsequent nodes run only after the Send reply node has finished playing

Nexus Voice

IVR Flow: call language added as a system variable and passed to Agents automatically

Functional Background

In multilingual scenarios, customers want the language determined during the IVR stage. Previously the language could not be passed to Agents. The new Language system variable lets the agent serve the customer in that language — for example, in an Agents workflow configured with three languages, the agent uses the language determined in the IVR.

Features Description

Entry point: IVR Flow → System variables → new Language.

• Language: holds a language code such as zh or en, and can be assigned in a variable assignment node, from API response parameters, and elsewhere.
• The value is carried automatically when the call is transferred to Agents, so the agent serves the customer in that language.

AI Assistance: AI transfer summary

Functional Background

When a call is transferred, the receiving agent has to read through the conversation with the previous agent or with Agents. The new AI transfer summary lets them grasp the key points of the earlier conversation the moment they pick up.

Features Description

Entry point: AI Assistance → new AI transfer summary.

• When it is generated: bot-to-agent transfers, agent-to-agent transfers, or both.
• What it covers: customer issue, customer sentiment, resolution progress and solution, plus custom points.
• Where to view it: generated automatically once the transfer succeeds; the receiving agent sees it in the call screen pop.

Note: requires an AI resource pack. AI transfer summaries draw on the AIGC resource pack. Summaries are not generated for transfers to third-party (external) numbers.

AI Assistance: AI task summary

Functional Background

The task summary for manual outbound campaigns could only be filled in by the agent. The new AI task summary fills it in automatically from the call.

Features Description

Entry point: AI Assistance → new AI task summary.

• AI generation: a Generate with AI button is added to the task summary of manual outbound campaigns, filling in the notes and custom fields from the call in one click.

Note: requires an AI resource pack. Each generation draws on the AIGC resource pack.

AI Assistance: recognition channels and recognition modes for real-time transcription

Features Description

Entry point: AI Assistance → ASR transcription → Real-time transcription of agent calls → new recognition channels and Recognition mode.

• Recognition channels: Channel De (trial), Channel By (trial) and Channel Al (trial) have been added.
• Recognition mode: speed first or accuracy first, supported by some channels.

Call Trigger \ Create ticket: create from a ticket template, with template fields bound to variables or constants

Functional Background

Creating a ticket from a call trigger previously only allowed a category, title and description. The new ticket template field mapping lets information collected in the IVR be written straight into the ticket.

Features Description

Entry point: Call Trigger → Action → Create ticket, and IVR Flow → Transfer to voicemail node → new Ticket template field mapping.

• Ticket template: once a ticket category is selected, the template bound to that category is pulled in automatically.
• Field mapping: template fields can be bound to a variable or filled with a constant.
• Trigger records: if ticket creation fails, the reason is shown.

Call Trigger \ Create ticket: AI generation mode

Functional Background

Creating a ticket from a call trigger previously only allowed binding variables or entering constants. The new AI generation mode fills the ticket in automatically from the call.

Features Description

Entry point: Call Trigger → Action → Create ticket → new AI generation mode.

• AI generation: generates the ticket title, issue description and template fields from the conversation in this call, following the prompt written for each field.
• Prompt description: write the generation requirements for each field; the title and description come with default prompts.
• Scope: calls on Nexus Voice, inbound Agents, inbound Agents (standard) and outbound Agents.

Note: requires an AI resource pack. Each AI generation draws 2 entries from the AIGC resource pack. The IVR Transfer to voicemail node does not yet support AI generation.

AI Assistance: ASR transcription split into real-time and recording configurations

Features Description

Entry point: AI Assistance → ASR transcription — Model settings is split into two independent configurations.

• Real-time transcription of agent calls: choose the recognition channel and model directly, without selecting a language first.
• Recording file transcription: configured by language, recognition channel, recognition model and hot words, unchanged.

Call Trigger \ Create ticket: per-field time zone for date and time, and variables in the issue description

Call records and bot call records: new call scenario filter

Features Description

Entry point: Record → Call Record and Record → Bot Call Record → new Call scenario filter.

• Call scenario: All, Normal call or Test call, defaulting to Normal call.
• Test call: call records produced by the Real phone call option in the Agents call test.
• Export: the export range matches the filter result.

IVR Flow: underscores allowed in transfer-to-bot node parameter names, with the length limit raised to 20 characters

Call Trigger: the list status filter now defaults to Enabled

IVR Flow: the flow canvas now opens in a new browser tab

Inbound and outbound calls: no call balance check when the trunk number is the customer's own number

Note: whether a number belongs to the customer is marked by the platform when the number is provisioned. Contact your customer success manager to confirm.

Inbound / Outbound Agents (Standard)

Bot management: Voice settings tab and new recognition channels

Functional Background

Customer speech recognition settings used to live under Advanced settings. The new Voice settings tab brings them together in one place and adds new recognition channels.

Features Description

Entry point: Bot Management → Edit bot → new Voice settings tab; the customer speech recognition settings formerly under Advanced settings have moved here.

• Recognition channels: Channel Al (trial) and Channel By (trial) have been added.

Note: the exact channel names are those shown on the product configuration page.

Nexus Messaging

Agent Workbench: translation for visitor image-and-text messages

Functional Background

In cross-language reception, visitors often attach a note in a foreign language next to a screenshot. Image-and-text messages sent by visitors did not enter the translation pipeline, so agents never saw a translation of that note. The text inside such messages now takes part in translation, while images and files are left as they are.

Features Description

Entry point: Agent Workbench → Chats → Translate to the right of the visitor's message (applied automatically when auto-translation is on)

• The text part of an image-and-text message from a visitor can be translated automatically, or manually by the agent clicking Translate
• The position and order of images and files are unchanged
• When the original and the translation are shown together, the translation area shows only the translated text and does not repeat images and files
• Only the text counts toward the translated character count

Agent Workbench: a system message when a chat is snoozed

Functional Background

In asynchronous reception mode, after an agent snoozed a chat the message stream only showed the un-snooze notice, with no record of the snooze itself. Agents taking over the chat, and quality monitoring staff reviewing it, could not tell whether the gap was a missed reply or a deliberate snooze. A system message is now left in the chat when an agent snoozes it.

Features Description

Entry point: Agent Workbench → Chats → Snooze at the top of the chat (Always On Mode)

• Once snoozed, the chat message stream shows "HH:mm:ss Agent {agent name} snoozed the chat"
• When snoozing in bulk, one message is shown in each chat

Agent Workbench: search my chats by customer name

Functional Background

The search box in Reception could only search chat content, so to find a particular customer an agent had to recall what was discussed and search backwards, or scroll through the chat list one by one. Chats can now be located directly by customer name.

Features Description

Entry point: Agent Workbench → Chats → Reception → search box dropdown (Chat Content / Customer Name)

• A dropdown has been added to the left of the search box to switch between Chat Content and Customer Name
• Customer names are matched loosely against the name shown on the chat list card
• If the list shows a remark name, searching the remark name also finds the chat

Agent Workbench: when closing chats in bulk, close the summarized ones first

Functional Background

With Service summary required to end chats enabled, a single chat without a submitted service summary blocked the whole batch from closing, and the agent had to open chats one by one to find it. Chats that have been summarized are now closed directly, leaving only the unsummarized ones behind.

Features Description

Entry point: Agent Workbench → Chats → Reception / Passively Ended Chats → select chats → close in bulk

• After clicking close in bulk, chats with a submitted service summary are closed and removed from the list
• Unsummarized chats stay in the list and stay selected
• A dialog reports how many chats were closed and how many remain
• Once the agent adds the missing service summaries, they can click close in bulk again

Agent Workbench records: chat content and message content become separate filters

Functional Background

In chat records and message records, "chat content / message content" sat in the same dropdown as customer nickname, integration ID and other conditions, so only one could be chosen at a time and there was no way to search by customer and by content together. The content condition has been split into its own input box so that conditions can be combined.

Features Description

Entry point: Agent Workbench → Record → Livechat Record (Filter Chats) / Submission Record

• Chat content and message content are now separate input boxes and can be filled in alongside customer nickname, customer phone, integration ID and other conditions; results are the intersection of all of them
• The matching behaviour for the content itself is unchanged

Chat records: Max Response Time filter and list metric

Functional Background

Chat records could only be filtered by average response time and first response time. An average hides individual slow replies, and first response only covers the first turn, so administrators could not find chats where "some turn took longer than N seconds." A maximum response time metric has been added for filtering and review.

Features Description

Entry point: Nexus Messaging → Record → Chat Record → Filter (Chat Info)

• A Max Response Time filter (greater than or equal to / less than or equal to N seconds) and list column have been added
• The value is the longest human reply time across the turns in the chat, using the same turn definition as average response time
• The column is included in exported files

Chat settings: multiple receiving addresses for message push

Functional Background

Agent offline messages and customer unread messages each allowed only one receiving address, so customers running primary/standby or multi-region receiving services had to forward messages again on their own end. Each message type can now have multiple receiving addresses.

Features Description

Entry point: Nexus Messaging → Setting → System Integration → Push Message → Unread Message Push

• The page is now a list of receiving addresses; click Add to choose a message type and enter an address
• Up to 5 addresses per message type, effective within 10 minutes of being added
• A message is pushed to every address configured for that type; a failure at one address does not affect the others

Asynchronous reception: chats assigned by queue order across an agent's skill groups

Functional Background

In asynchronous reception mode, when an agent belonged to several skill groups, queued chats were assigned per group, so a customer who queued earlier could be served later than one who queued afterwards in a different group. Assignment now follows the customer's queue time across all of the agent's groups.

Features Description

Entry point: Nexus Messaging → Setting → Business Settings → Mode → Always On Mode (chat assignment)

• Across the skill groups an agent belongs to, the system assigns chats by the order in which customers entered the queue
• Customers who queued earlier are served first

Visitor input box: send images, files and text together

Functional Background

On the visitor client, selecting an image sent it immediately on its own, so a visitor wanting to send a screenshot and describe the problem had to split it into several messages, breaking the link between image and text. When a bot was handling the chat, an image sent on its own was treated as a message it could not understand. Visitors can now combine images, files and text into a single message, and the LLM-powered bot can read the images and files.

Features Description

Entry point: Chat Widget (Desktop Site / Mobile Site / App SDK) → input box

• Images and files the visitor selects, pastes or drags in first go to a staging area above the input box, where they can be removed or added to
• They are sent together with the text as one message, with up to 10 attachments per message
• Agent Workbench, chat records and real-time monitoring all show it as a single message
• When an LLM-powered bot is handling the chat, the text, images and files are all used to understand and answer
• When an NLP bot is handling the chat, it replies based on the text part and no longer triggers the fallback script

Visitor settings: voice messages on desktop sites

Functional Background

Voice messages could only be enabled on mobile sites, leaving desktop visitors to type, which is slow when describing a complex problem. Voice messages are now available on desktop sites so visitors can record and send audio.

Features Description

Entry point: Nexus Messaging → Visitor Settings → Chat Widget Style → Edit → Input Box Setting → Desktop Site → new Voice Message

• Can be enabled separately for bot reception and human reception; off by default
• Once enabled, a microphone icon appears in the desktop input box; the visitor clicks to start recording and clicks to send, up to 60 seconds per message
• Agent Workbench plays the audio and shows the speech-to-text result as it does today
• Supported on the new visitor client only; the site must be served over HTTPS

Nexus Ticketing

Satisfaction survey: automatic survey email for email tickets

Functional Background

The satisfaction survey invitation for email tickets was only attached to the agent's next reply email. Once a ticket was resolved the agent usually stopped replying, so customers never received the invitation and email tickets collected very few ratings. The system now sends a separate satisfaction survey email once the ticket is resolved.

Features Description

Entry point: Nexus Ticketing → Ticket Processing → Satisfaction Survey → Email → new Automatically send a satisfaction survey email

• Once the option is selected and a sending mailbox is chosen, the system sends a satisfaction survey email to the customer as soon as a ticket created through the email channel reaches the status set in Invitation Condition (resolved or closed)
• No further email is sent once the customer has rated; the rating link is valid for 28 days
• The email content is edited in the new Satisfaction Survey template under email template settings
• The existing Include the satisfaction survey link in reply emails becomes optional and remains enabled by default

Ticket plug-in: pass integration fields via token

Functional Background

Values for integration fields were passed as plain-text parameters in the ticket plug-in page link, which exposed them in the browser address bar and in access logs; if a visitor edited the link parameters, the submitted ticket recorded the edited values. Integration fields can now be passed via token to prevent leakage and tampering.

Features Description

Entry point: Nexus Ticketing → Ticket Channel Setting → Ticket Plug-in → new Security Key

• With the security key enabled, your server signs with the key to obtain a token, the page link carries only the token, and the system restores the field values from the token when the ticket is created
• The feature is off by default; once enabled, links without a token still resolve values the old way, so you can switch over after completing the integration work
• See the ticket plug-in integration documentation for details

Ticket plug-in: Persian, Khmer and Hindi added to the submission list and ticket plug-in

Functional Background

Some companies have visitors who use Persian, Khmer or Hindi, and neither the submission list nor the ticket plug-in offered those languages. All three interface languages have been added.

Features Description

Entry point: Ticket Plug-in / Submission List → Language

• Persian, Khmer and Hindi have been added to the language list of the submission list and the ticket plug-in
• Once selected, the default interface text is shown in that language

Ticket replies: longer email domain suffixes accepted

Functional Background

Ticket replies only accepted alphabetic email suffixes of 2–8 characters, so when a customer's email used a longer suffix such as .solutions or .consulting, or an internationalized domain suffix, the system reported an invalid address and the reply could not be sent. Suffix validation for ticket replies has been relaxed.

Features Description

Entry point: Nexus Ticketing → Ticket Details → Reply → To / Cc

• The email suffix length limit is raised to 2–63 characters
• Internationalized domain suffixes beginning with xn-- are supported
• Agents can reply to tickets from such addresses normally

Email replies: show the customer's original email in full

Functional Background

When a customer's original email body was long, the issue description inside Ticket Creation Info was truncated in the reply email they received, so they could not see the full content they had sent. Reply emails now show the original email in full.

Features Description

Entry point: Nexus Ticketing → Ticket Channel Setting → Email → Email Template Setting → Reply to Customer → Ticket Creation Info

• The ticket description in the Ticket Creation Info block of reply emails now shows the complete content and is no longer truncated by length
• Images and attachment names are displayed as before

Attachment fields: block submission when a required attachment has not uploaded

Functional Background

When an attachment field was set as required but the upload failed — because of a timeout, for example — the ticket could still be submitted, leaving the required attachment missing. Submission validation for attachment fields has been adjusted.

Features Description

Entry point: Ticket submission page → Attachment field

• When an attachment field is required, a ticket cannot be submitted while any attachment has not uploaded successfully
• Multiple attachment fields are validated separately

Agent App

Agent App phone bar: busy status approval

Functional Background

Busy status approval was only supported on the PC phone bar; it is now supported on the app phone bar as well.

Features Description

Entry point: Sobot App → Phone bar → new Busy status approval.

• Request: when switching to a busy status that requires approval, the agent enters a reason and submits a request, which takes effect once approved.
• Cancel: agents can cancel their own pending requests.

Agent App phone bar: choose from all enabled sign-in statuses at login

Functional Background

When signing in to the app phone bar, agents could not choose a specific busy status. They can now sign in directly into a specific busy status — Meeting, for example — which saves a step.

Agent App phone bar: automatic reconnection when the softphone drops on the call page

Functional Background

Automatic reconnection reduces how often the phone bar drops when the user's network is unstable.

Features Description

When the app or SDK opens the call page, it reconnects automatically if it detects that the softphone connection has dropped.

• Automatic reconnection: a reconnection notice is shown on the page after the first failed attempt.
• Reconnection failure: after repeated failures the agent is set offline and prompted to sign in to the phone bar again.
• Outbound restriction: outbound calls are unavailable while reconnecting.

Note: drops during transfer or consultation cannot be recovered. We recommend also enabling "Set agent offline on SIP connection timeout."

Agent App phone bar: hang-up button shown on the call page for phone-mode outbound calls

Mine: Account & Security, with account and company deletion

Functional Background

The Agent App had no entry point for account deletion, so when an agent left or a company stopped using the product, it had to be handled by an administrator or by email request. The app now offers self-service deletion for both accounts and companies.

Features Description

Entry point: Sobot App → Mine → new Account & Security

• Super administrators can Delete Company here, following the same rules as in Admin Center
• Other roles can Leave Company and Delete Account; after confirmation the account is disabled and signed out, personal information collected by the app is deleted, work records such as tickets and chats are retained by the company, and an administrator can re-enable the account

Quality Monitoring

Result Search: filters for the new service summary

Functional Background

Task assignment and inspection scoring already supported filtering by the new service summary, but Result Search did not, so customers could not find quality monitoring results for it. The new service summary filters have been added to Result Search.

Features Description

Entry point: Quality Monitoring → Chat Inspection → Result Search → Filter

• A Service Summary (New) group has been added to the filter area, containing Service Category and Progress, with the same rules as the inspection scoring page
• The existing service summary filters are grouped under Service Summary
• Visible to companies that have the new service summary enabled, and applies only to quality monitoring data produced after release

Quality monitoring filters: select disabled and deleted agent and inspector accounts

Functional Background

Filters for agents and inspectors across quality monitoring pages only offered enabled accounts, so once an agent or inspector was disabled or deleted, their historical chats, calls and inspection records could no longer be filtered or reviewed. Agents and inspectors can now be selected by account status.

Features Description

Entry point: Quality Monitoring → Chat Inspection / Call Inspection → Task Assignment, Inspection Scoring, Result Search, Appeal, Inspection Statistics → agent / inspector filter

• The filter control is now a selection dialog where accounts can be found by All Statuses / Enabled / Disabled / Deleted and selected across statuses
• Exports and saved sampling plans support this as well
• When assigning inspection tasks, only enabled inspectors can still be selected

Admin Center

Role permissions: new Batch Transfer permission

Functional Background

The Batch Transfer button in Agent Workbench had no corresponding item in role permissions, so every role could use it and administrators could not restrict bulk chat transfers. A dedicated batch transfer permission has been added.

Features Description

Entry point: Admin Center → Agent Management → Role Permission → Agent Workbench → Chat → Current Chat → Function → new Batch Transfer

• For roles without this permission, Batch Transfer in the chat list, Transfer Chat when a single chat is selected, and Batch Transfer Chats in the offline handling panel are all hidden
• Existing and newly created roles have it selected by default, so current usage is unaffected
• This permission is independent of the Transfer Function permission

Company account: self-service company closure for super administrators

Functional Background

When a company stopped using the system there was no self-service way to close it — it had to be handled offline through sales or customer success. Super administrators can now close the company themselves.

Features Description

Entry point: Admin Center → Account Management → Company Account → Product Ordering Info → new Close Enterprise

• Visible to super administrators only
• Takes effect immediately after clicking Close Enterprise and confirming the company name
• All accounts under the company can no longer sign in, and reception stops on every channel
• Company data is deleted 3 months after closure and cannot be recovered, so export anything you need to keep beforehand
• Paid companies cannot close themselves during the current service period; click Online Support at the bottom right of the page to contact your account manager

SSO: entering Admin Center messaging pages without signing in no longer logs the agent out of Agent Workbench

Functional Background

Some companies embed Agent Workbench and Admin Center separately in their own systems. Using SSO to go straight to a messaging page in Admin Center signed the agent out of the Agent Workbench session they already had open. The existing session can now be reused when entering Admin Center pages.

Features Description

Entry point: Admin Center → Information Security → SSO (login_url integration parameter)

• A new optional parameter has been added to login_url. When it is passed, an account already signed in to Agent Workbench in the same browser will not be signed in again when entering Admin Center messaging pages
• Agent Workbench stays online and reception is not interrupted
• Behavior is unchanged when the parameter is not passed; see the login_url page in the developer documentation for details

Custom fields: note that HarmonyOS does not support audio attachments

Functional Background

HarmonyOS does not yet support uploading audio attachments, and administrators had no way of knowing this when configuring an attachment field. A note has been added where the field is configured.

Features Description

Entry point: Admin Center → Business and Rule → Basic Settings → Custom Field → Add Field → Attachment

• When configuring a custom field of the attachment type, the audio line in the supported file types now reads Audio (HarmonyOS is not currently supported)
• This reminds administrators that audio attachments cannot yet be uploaded on HarmonyOS

Admin Center \ Business and Rule: replacement word management

Functional Background

Homophone or near-homophone errors in speech recognition could previously only be reduced by raising recognition probability with hot words. The new replacement words automatically replace a recognized incorrect word with the correct one.

Features Description

Entry point: Admin Center → Business and Rule → new Replacement word management.

• Replacement words: configure a correct word and the incorrect words that map to it; whenever an incorrect word appears in a recognition result it is replaced automatically.
• Import: enter them manually or import a TXT file in bulk.
• Scope: inbound Agents, inbound Agents (standard), outbound Agents, real-time transcription of agent calls, and transcription of recordings and voicemail files.

Note: up to 800 replacement words per company, and up to 10 incorrect words per correct word.