Hello everyone,
3Chat.ai v5.2 is expected to be released at the end of June.
In v5.0, we officially launched Agent Builder. In v5.1, we continued to enhance the version management capabilities of Agent Builder, allowing for clearer draft, publish, and historical version management for Agent configurations, instructions, and skills.
In the upcoming v5.2, Agent Builder will continue to evolve into a complete closed loop:
From Agent configuration generation to automated testing, performance analysis, continuous optimization, and version management, forming a more comprehensive Agent building and optimization workflow.
Additionally, v5.2 will introduce new “E-commerce Channel” integration capabilities. Through the 3ChatClaw desktop assistant, it will help merchants integrate seller consultation scenarios from more domestic e-commerce platforms.
Below is the preview for this version 
I. Agent Builder: From “Generating Agents” to “Continuously Optimizing Agents”
The core direction of v5.2 is to make Agent Builder not only help you generate an Agent but also help you continuously verify, analyze, and optimize it.
In the past, after users completed Agent configuration, they often had to test repeatedly on their own:
- Can this Agent handle typical customer inquiries?
- When encountering high-intent customers, does it guide them correctly?
- When facing after-sales issues, does it reply correctly or transfer to human agents?
- After modifying instructions or skills, did the performance improve?
- In real online conversations, where does the Agent perform well, and where does it need optimization?
v5.2 will continue to enhance Agent Builder’s capabilities around these questions.
1. Agent Builder Supports Simulated Testing
v5.2 will add Simulated Testing capabilities.
Once the Builder completes Agent generation, the system will guide you to use two testing methods:
- Simulated Testing
- Sandbox Testing
Their purposes differ.
Simulated Testing is more suitable for verifying core and typical scenarios. The Builder will automatically generate recommended test scenarios and simulate multi-turn conversations between the customer and the Main Agent, finally providing test results and evaluations.
Sandbox Testing is more suitable for users to role-play as customers and manually experience the actual conversational effects of the Agent. It is especially suitable for verifying edge cases, temporary issues, or unpredictable situations.
Simply put:
Simulated testing is used for systematic verification, while sandbox testing is used for manual exploration of details.
2. The Builder Automatically Recommends Core Test Scenarios
After the first Agent generation, the Builder will generate a set of recommended simulated test scenarios for you.
These test scenarios will prioritize covering typical paths that need the most verification before the Agent goes live, such as:
- High-intent customer inquiries
- Customers who only want to learn more and are not ready to buy immediately
- Customers asking about prices, shipping, after-sales service, materials, etc.
- Customers who need to be guided to leave contact information
- Customers triggering the transfer-to-human-agent scenario
After generation, the system will prioritize displaying 5 recommended test scenarios.
Each test scenario will include:
- Scenario Name
- Opening message from the simulated customer
- Customer persona and background information
- Goals to be verified
- Pass criteria
- Execution entry point
Users can choose to run a single scenario or run all recommended scenarios at once.
3. Automatic Execution of Multi-Turn Conversations and Output of Test Results
When you run a simulated test scenario, the system will automatically simulate multi-turn conversations between the customer and the Main Agent.
During execution, you can see:
- Messages sent by the simulated customer
- Replies from the Main Agent
- Status of the Main Agent’s processing
- Agent Logs
- Execution process of the current test scenario
After execution, the system will provide test results.
Results will be displayed as “Passed” or “Failed”:
- If all evaluation criteria are met, it is marked as Passed
- If any evaluation criterion is not met, it is marked as Failed
In case of failure, the system will explain the difference between the Agent’s actual performance and the pass criteria, and provide optimization suggestions.
This means users do not need to judge the Agent’s quality based solely on intuition but can determine whether the Agent meets expectations based on test scenarios and evaluation criteria.
4. Support for Custom Test Scenarios
In addition to using the test scenarios recommended by the Builder, users can also create or edit test scenarios themselves.
A test scenario can be configured with:
- Scenario Name
- Customer Opening Message
- Customer Persona
- Customer Background Information
- Initialized Contact Fields
- Evaluation Criteria
- Final Conversation Outcome Requirements
Evaluation criteria can cover various types, such as:
- Whether the Agent fulfilled a specific reply requirement
- Whether the Agent correctly collected customer information
- Whether the Agent correctly updated custom fields
- Whether the Agent should close the conversation
- Whether the Agent should transfer to a human agent
For test scenarios requiring images or attachments, users can also supplement corresponding materials to verify the Agent’s handling effectiveness in more realistic business scenarios.
5. Test Scenarios Support Copy, Edit, and Delete
In v5.2, simulated test scenarios will also support common management operations.
You can:
- Create a new test scenario
- Edit an existing test scenario
- Copy a test scenario
- Delete a test scenario
- View the most recent execution result
- View historical execution records
Please note:
Deleting a test scenario only deletes the test case itself, not the conversation records generated from previous executions. The generated simulated conversations can still be viewed in the inbox.
This keeps the test scenario list clean while preserving historical test records.
6. Support for Analyzing Online Execution Results and Logs
In addition to pre-launch simulated testing, v5.2 will further enhance online performance analysis capabilities.
Users can selectively choose online execution results and logs, allowing the Builder to analyze real customer conversations.
This type of analysis can help you discover:
- Which scenarios the Agent handled well
- Which problems the Agent did not understand accurately
- Which conversations failed to guide customers to continue inquiring
- Which scenarios need additional knowledge base content
- Which scenarios need adjustments to Guidance
- Which scenarios need new or optimized Skills
- Which scenarios require fallback rules for human agent transfer
In other words, the Builder will not only help you build the Agent but also gradually participate in reviewing the Agent’s performance and providing optimization suggestions.
7. Weekly Push of Main Agent Performance and Insight Reports
v5.2 will also support weekly pushes of Main Agent performance and insight reports.
These reports will be sent to super administrator accounts within the workspace.
Based on account information, the system will push reports via:
- SMS
The weekly reports will help workspace administrators regularly understand the overall performance of the Agent, such as:
- This week’s Agent reception status
- Common customer questions
- High-value consultation scenarios
- Conversations that need optimization
- Knowledge that may need to be supplemented
- Business insights that require focused attention
This allows the team to continuously monitor the Agent’s performance and optimization direction even without entering the backend daily.
II. E-commerce Channels Coming Soon
v5.2 will introduce new “E-commerce Channel” capabilities.
These channels are primarily aimed at seller consultation scenarios on domestic e-commerce platforms and require integration via the 3ChatClaw Desktop Assistant.
3ChatClaw will assist in connecting your platform account on the desktop and help the 3Chat Agent handle consultation messages from buyers.
1. E-commerce Channels to Be Supported
v5.2 is expected to support the following e-commerce platform-related channels:
| Platform | Skill ID | Skill Name |
|---|---|---|
| Douyin Shop | douyin-shop-im-reply |
Douyin Shop - Top Sales Champion |
| Taobao | qianniu-android-im-reply |
Taobao Qianniu - Top Sales Champion |
| Pinduoduo | pdd-android-im-reply |
Pinduoduo - Top Sales Champion |
| JD.com | jingmai-android-im-reply |
JD.com Jingmai - Top Sales Champion |
| 1688 | 1688-android-im-reply |
1688 Merchant End - Top Sales Champion |
| Xianyu | xianyu-web-replying |
Xianyu - Top Sales Champion |
| Meituan | meituan-comment-replying |
Meituan Comments - Super Front Desk |
These channels will help merchants integrate buyer consultations from more platforms into 3Chat, allowing the Agent to participate in pre-sales reception, comment replies, consultation guidance, and customer conversion.
2. Integration via the Omnichannel Page
Integration entry points for all e-commerce channels will be uniformly displayed on 3Chat’s Omnichannel page.
You can find the corresponding platform on the Omnichannel page and complete the integration following the on-screen instructions.
The integration process generally includes:
- Enter the Omnichannel page
- Find the e-commerce platform to integrate
- Click the integration entry point for the corresponding channel
- Download and install the 3ChatClaw Desktop Assistant
- Open 3ChatClaw and enter the Skill Market
- Install the skill for the corresponding platform
- Follow the guidance in the 3ChatClaw conversation to complete environment configuration
- After connection is complete, the 3Chat Agent will assist in handling buyer consultations
3. Why is the 3ChatClaw Desktop Assistant Needed?
Some e-commerce platforms do not currently provide standard open APIs, or their open capabilities cannot fully cover the instant consultation scenarios in the seller backend.
Therefore, 3Chat completes integration through the 3ChatClaw Desktop Assistant.
Under your authorization and configuration, 3ChatClaw assists in connecting to the platform account, allowing the 3Chat Agent to participate in handling buyer consultations within the platform.
This method is suitable for the following scenarios:
- The platform does not provide a standard open API
- Merchants still rely on the platform’s native customer service tools
- You want AI to assist in handling high-frequency pre-sales consultations
- You want to quickly integrate multiple domestic e-commerce platforms
- You want to reduce the repetitive reply workload of human customer service agents
4. Integration Boundary Description
Integration of e-commerce channels relies on the 3ChatClaw Desktop Assistant, so please note:
- You need to download and install 3ChatClaw
- You need to install the corresponding platform skills in 3ChatClaw
- You need to complete environment configuration following the skill conversation guidance
- The platform account needs to remain available
- Specific supported capabilities may vary depending on the platform environment and skill capabilities
If you are already using platforms such as Taobao, Douyin Shop, Pinduoduo, JD.com, 1688, Xianyu, or Meituan Comments, v5.2 will make it easier for you to integrate consultations from these platforms into the 3Chat Agent.
III. Summary
3Chat.ai v5.2 will continue to upgrade around the theme of “making Agents easier to build, test, optimize, and connect to real business.”
Key highlights of this preview include:
- Agent Builder supports simulated testing
- Builder automatically recommends core test scenarios
- Support for running multi-turn simulated conversations and viewing pass/fail results
- Support for creating, copying, editing, and deleting test scenarios
- Support for performance analysis based on online execution results and logs
- Support for weekly pushes of Main Agent performance and insight reports
- Launch of e-commerce channels, integrating multiple domestic e-commerce platforms via 3ChatClaw
- Support for scenarios such as Douyin Shop, Taobao Qianniu, Pinduoduo, JD.com Jingmai, 1688, Xianyu, and Meituan Comments
v5.2 will take 3Chat Agents from “can be built” to “verifiable, reviewable, and continuously optimizable.”
More functional details will be synchronized when v5.2 is officially released.
—
The 3Chat.ai Team