3Chat.ai v5.3 Release Notes: Enhanced Agent Builder Capabilities, Contextual Knowledge Retrieval, and Service Language Support
Hello everyone,
3Chat.ai v5.3 is officially released.
Starting from v5.0, the Agent Builder has helped users create their own 3Chat Agents using natural language; v5.1 added version management; and v5.2 further completed simulation testing, allowing Agents to move from “buildable” to “testable and verifiable.”
In v5.3, we have taken another step forward:
The Builder no longer just generates configurations based on requirements. Instead, it first understands your business, plans the overall solution, completes the build, and then reviews the final result.
At the same time, 3Chat Agents have undergone multiple upgrades in areas such as knowledge understanding, knowledge retrieval, customer profile collection, group chat configuration, language recognition, and IM reply experience.
Below are the detailed update notes for this version 
1. Significant Upgrade to Agent Builder Capabilities
1. Plan First, Then Build
In previous versions, the Builder could already understand business requirements through natural language and generate Agent configurations such as Guidance and Skills.
In v5.3, we have further upgraded the Builder’s building approach.
Now, before actually modifying the Agent, the Builder starts from the overall business objectives and plans your business scenarios, marketing strategies, and agent architecture.
You can think of this as adding a “Chief Architect” to the Builder.
When you tell the Builder:
- Who your customers are
- What products or services you sell
- What your core business objectives are
- What consultation stages customers typically go through
- Which questions the AI should handle
- Which scenarios should be transferred to human agents
- What actions you want customers to complete eventually
The Builder will synthesize this information to form a more complete business strategy and Agent architecture blueprint, and then complete the specific build based on this overall plan.
This means the Builder is no longer just “receiving a requirement and changing a configuration.” Instead, it first determines:
How this requirement should be designed to be most reasonable within the entire customer conversion process.
2. Re-evaluate the Overall Plan When Business Changes Occur
“Plan first, then build” does not only happen during the initial creation of an Agent.
When an already running Agent undergoes significant business changes, such as:
- Changing business objectives from lead capture to direct appointment booking
- Adding a new sales pipeline
- Switching to a main product
- Significant changes in the customer base
- Overall adjustment of the original conversion process
- Changes in the Agent’s positioning
The Builder will first assess whether these changes affect the existing overall plan.
If it is only a local adjustment, it will only modify the corresponding content.
If it involves the overall direction of the Agent, the Builder will clearly inform you of which parts might be affected and, after obtaining your confirmation, proceed with broader adjustments.
This reduces situations where “changing one thing unexpectedly affects other processes.”
3. Add Checks and Reviews to Build Results
Starting from v5.3, every Builder build result will undergo an additional check and review.
The Builder will re-check the newly generated or modified content, such as:
- Whether it aligns with current business objectives
- Whether there are obvious conflicts between different Skills
- Whether the design of Guidance and Skills matches
- Whether important business scenarios were missed
- Whether there are contradictory rules
- Whether the modification accidentally damaged content that did not need adjustment
If obvious problems are found, they will be corrected before the build is completed.
This adds a quality check after “generation,” making every version of Agent configuration more stable.
4. Precise Modifications with Minimal Interference
v5.3 also enhances the Builder’s understanding of “fine-grained modification intentions.”
Now, you can more explicitly tell the Builder to modify only a specific part.
For example:
- “Only adjust the communication style in the guidance”
- “Tighten the conditions for transferring to human agents a bit more”
- “Only change the Inquiry Skill”
- “Add a refund scenario to the After-sales Processing Skill”
- “Do not change other skills, only adjust the opening stage”
- “Make the second step of this Skill more proactive”
The Builder will try to limit modifications to the scope you specify, avoiding unnecessarily redoing the entire Agent.
However, “precise modification” does not mean completely ignoring upstream and downstream impacts.
If the Chief Architect determines that this adjustment will significantly impact the overall business process, such as deleting a core conversion Skill or modifying the entire Agent’s business objectives, it will first point out the scope of impact, and then let you decide whether to continue adjusting related content.
This allows the Builder to balance two things:
Small changes are kept precise, and large changes maintain overall consistency.
2. Continued Enhancement of 3Chat Agent’s Knowledge Understanding and Retrieval Capabilities
1. Knowledge Base Now Understands “Which Scenario This Knowledge Is Suitable For”
In v5.3, we have further enhanced the processing method of the knowledge base.
Previously, the focus of the knowledge base was mainly on understanding “what the document says.”
Now, when uploading and processing knowledge, 3Chat further understands:
- What business situation this part of knowledge applies to
- How customers typically ask about it
- Which type of consultation intent it may correspond to
- In what conversational context this content should be used
For example, for “refund policies,” different contents may apply to:
- Asking if a refund is possible before ordering
- Paid but not yet shipped
- Goods already received
- Goods have quality issues
- Past the after-sales period
v5.3 will further understand these scenario differences, rather than just treating them as knowledge containing the keyword “refund.”
2. When Retrieving Knowledge, Combine with Current Conversation Context
Along with enhanced knowledge processing capabilities, the Main Agent will pay more attention to what is currently happening in the conversation when querying the knowledge base.
For example, if a customer says:
“Can I still return it if I’ve used it for two days?”
Looking at this sentence alone, there is no complete information about the product, order, or policy.
However, the Agent will combine the previous conversation content to understand that the customer is currently discussing the “post-use return scenario” and prioritize retrieving knowledge that better matches this scenario.
This evolves the knowledge base from simple “content-related” retrieval to “scenario-related” retrieval.
For businesses with complex product rules and numerous pre-sales and after-sales processes, this approach helps the Agent find more complete and relevant knowledge for the current issue.
3. Skills Support Specifying Specific Knowledge Content
In addition to allowing the Agent to automatically determine which knowledge to search, v5.3 adds a more deterministic usage method.
Now, you can specify specific knowledge content as the scope for knowledge retrieval within a Skill.
This is suitable for scenarios where the answer scope is very clear, such as:
- Installation instructions for a specific product
- After-sales policies for a specific region
- Pricing rules for a specific service
- Franchise policies for specific customer types
- SOPs corresponding to a specific business process
When you clearly know which part of the knowledge a Skill should rely on, you can narrow the retrieval scope, allowing the Agent to answer more stably around the specified knowledge.
Therefore:
- For open-ended Q&A, you can continue to use broader knowledge base retrieval
- For Skills with strong processes, strong rules, and clear answer scopes, prioritize specifying specific knowledge content
3. Customer Profile Collection Capabilities Extended to Human Reception Stage
In v5.3, profile collection Skills added an important capability:
Even if the conversation has been transferred to a human agent, the Agent can still continue to extract customer profiles from the chat content between the human agent and the customer.
For example, if the customer’s budget was not confirmed during the AI reception stage, but after transferring to a human, the agent asks:
“What is your approximate budget?”
The customer replies:
“Around 50,000.”
If the current Skill is configured with profile collection for the budget field, the system can continue to identify this information and update it to the corresponding customer profile.
This way, customer profile collection is no longer limited to the AI reception stage but can span the entire communication process from AI to human.
Information suitable for collection includes:
- Budget
- City
- Product requirements
- Purchase stage
- Intent level
- Appointment time
- Other business custom fields
Note:
This capability currently does not support group chat scenarios.
4. IM Reply Experience Continues to Optimize Towards Human-like Communication
v5.3 further optimized the IM reply rhythm of the Agent.
This update mainly adjusted two directions:
- Time intervals between replies are more natural
- Long and short sentences use different reply rhythms
In instant messaging scenarios, human replies usually do not send a large paragraph of content instantly in succession, nor do they wait a long time for a very short reply.
Now, the 3Chat Agent will more reasonably arrange the message sending rhythm based on the length of the reply content and message structure.
This makes the conversation experience more natural in instant messaging scenarios such as WeChat, WhatsApp, LINE, and LiveChat, and reduces the obvious feeling of “machines spitting out messages continuously.”
5. Group Chat Configuration Capabilities Further Improved
Starting from v5.1, 3Chat Agents officially supported standard group chat scenarios; in subsequent versions, we continuously expanded capabilities related to groups, group members, and group chat context.
In v5.3, this configuration is further completed.
Now, in “Contact Groups” and “Deployments,” you can fully use:
- Standard group chat fields
- Custom group chat fields
for filtering and filtering.
This means you can more flexibly decide:
- Which groups should use a specific Skill
- Which group members are subject to certain reception rules
- Adjust groups that do not require 3Chat Agent reception
Combined with the previously supported group chat Skills and conditional following capabilities, group chat scenarios can now undergo more complete business configuration.
6. Agent Language Recognition Strategy Upgrade
For cross-border businesses and multilingual customer service, language is a seemingly simple but easily impactful basic capability that affects the Agent experience.
v5.3 has comprehensively upgraded the language strategy for 3Chat Agents.
1. New “Service Language” Configuration
A “Service Language” has been added to the basic Agent configuration.
You can configure the range of languages this Agent provides services for.
For example, set:
- Simplified Chinese
- English
- Japanese
When the Agent recognizes that the language used by the customer is within the service language range, it will reply in the customer’s current language.
If the language used by the customer is not within the service language range, it will reply using the default language configured for the Agent.
This avoids the Agent switching to a language that the enterprise does not wish to provide services for due to occasional language content.
2. Local Businesses Are Advised to Configure Only One Service Language
If your business mainly serves local customers, such as only serving users in mainland China, you can:
Service Language = Simplified Chinese
Default Language = Simplified Chinese
At this time, the Agent will stably provide services in Simplified Chinese.
Even if customer messages occasionally contain English product names, URLs, or other language content, it will not easily change the reply language.
3. Multilingual Businesses Are Advised to Clearly Configure Supported Languages
If your business serves multiple countries or regions simultaneously, it is recommended to add all languages actually provided for service to “Service Language.”
For example:
- English
- Spanish
- Portuguese
The Agent will reply based on the language actually used by the customer within these language ranges.
This retains multilingual service capabilities while limiting the Agent from unnecessarily switching to languages not supported by the business.
4. More Stable Language Recognition
v5.3 also optimized the language recognition process.
The Agent will automatically reduce interference from non-natural language content on recognition results, such as:
- Addresses
- URL links
- Emails
- Images
- Videos
- Files
- Other non-natural language content carried with messages
For example, if a Chinese customer sends:
“Can this be sent to the address at https://xxx.com?”
The system will not mistakenly identify the customer as an English user due to the large number of English characters in the URL.
5. Stable Conversation Language Once Determined
During a conversation, once the Agent has clearly determined the language used by the customer, this conversation will confirm the use of that language and remain stable.
It will not constantly switch the reply language just because the customer occasionally sends an English product name, an address, or a URL later.
Therefore, the language strategy in v5.3 can be simply understood as:
Determine customer language → Confirm if it belongs to service language → Determine conversation language for this session → Maintain stability thereafter.
7. Open Platform Adds Group Chat Capabilities
As 3Chat group chat capabilities gradually improve, v5.3 also synchronously expands open platform capabilities.
New additions include:
- APIs related to group chat groups
- Events related to group chat groups
Developers can further process group chat businesses based on the open platform, such as synchronizing group information, obtaining group chat-related data, and listening to group chat changes.
For customers who have already integrated 3Chat into CRM, business backends, or other internal systems via OpenAPI, these capabilities can help group chat data further participate in their own business processes.
8. Subscription and Benefit Experience Upgrade
1. Optimized Entry Points for Purchasing Certain Benefits
v5.3 optimized the purchase entry points for the following benefits:
- WeChat Interoperability
- Social Platforms
- WeChat Personal Accounts
Users can now find relevant benefits and purchase entry points more directly, reducing the cost of searching back and forth between subscriptions, channels, and benefits.
2. New “Large Model Credits”
More and more customers are starting to use AI to process business data within 3Chat’s automated workflows.
For example:
- Automatically summarize text
- Information classification
- Extract structured fields
- Judge customer intent
- Content generation
- AI analysis of business information
To support such scenarios, v5.3 added the Large Model Credits benefit.
3Chat simultaneously provides official large model gateway interfaces, allowing users to call models in their own automated workflows and use large model credits to offset corresponding consumption.
Additionally, the subscription benefits page has added corresponding large model credit purchase entry points.
This allows automated workflows that need to use AI capabilities outside of Agents to uniformly use the model services provided by 3Chat.
3. Adjustment of Benefit Reset Time for Quarterly and Annual Subscription Users
v5.3 adjusted the benefit reset rules for quarterly and annual subscriptions.
Previously, some benefits were reset at 0:00 on the 1st of each month.
Now, quarterly and annual users will base their reset on the current subscription start time, resetting corresponding cycle benefits every 30 days.
For example:
If your subscription started on August 12th, subsequent benefit cycles will be calculated around this subscription time, rather than waiting for the 1st of the next natural month.
This makes the benefit cycle more consistent with the actual subscription cycle.
9. Other Optimizations and Bug Fixes
1. Fixed Connection Interruption Issues During Builder Build Process
Optimized connection stability during the Agent generation and plan adjustment phases of the Builder.
Fixed issues where page connections were interrupted during long generation or modification processes, affecting the user experience.
2. “Contact Us” Entry Restored
v5.3 restored the “Contact Us” entry in the menu bar.
If you need product consultation, plan communication, or online support while using 3Chat, you can schedule an online meeting with a consultant through this entry.
3. Clearer Channel Type Filtering
We also optimized the channel type filtering in pages such as Contact Groups and Deployments.
Previously, the same channel might appear as multiple similar options in filtering conditions due to different integration methods.
v5.3 aggregated identical channels.
Now, users can filter directly according to the channel types as they understand them, without needing to care about which integration method is used at the underlying level.
10. Summary
3Chat.ai v5.3 continues to advance around a core direction:
Make Agent building more professional, make operation more stable, and make business configuration more controllable.
Key highlights of this version include:
- Agent Builder adds overall design capability of “plan first, then build”
- Every build result undergoes checks and reviews
- Builder supports more precise local modifications and automatically assesses the scope of modification impact
- Knowledge base supports scenario understanding, and knowledge retrieval further combines with current conversation context
- Skills support specifying specific knowledge content for precise retrieval
- Profile collection Skills support continuing to collect customer profiles after transferring to human agents
- IM reply rhythm is further humanized
- Contact Groups and Deployments fully support filtering by group chat, group members, and their custom fields
- Added service language to further improve the stability of multilingual reception
- Open platform adds group chat group APIs and related events
- Added large model credits and official model gateway capabilities
- Benefit cycles for quarterly and annual users adjusted to reset every 30 days based on subscription time
- Fixed Builder connection issues and optimized multiple product usage experiences
From the launch of Agent Builder in v5.0, to simulation testing in v5.2, to “plan → build → review → precise modification” in v5.3, the Builder is gradually growing from an Agent configuration assistant into an AI building partner that truly understands business objectives, can participate in the overall design of Agents, and continuously optimizes them.
At the same time, the 3Chat Agent itself is continuously enhancing its understanding capabilities of knowledge, language, group chats, and real customer contexts.
Our goal remains unchanged:
Allow enterprises to build and continuously operate a 3Chat Agent that truly understands their business, without needing to become AI experts.
If you have any questions or suggestions during use, welcome to communicate with us in the community 
—
3Chat.ai Team
