Release Notes
Last updated:2025-09-15 18:02:59
September 2025
Update | Description | Release Date | Documentation |
Application Mode adds Multi-Agent Mode | Multi-Agent Mode supports creating single-Agent applications, allows adding multiple Agents, and supports switching between different multi-Agent collaboration methods to satisfy needs in various scenarios. Free transfer: Agents collaborate with each other to handle complex requirements by flexibly transferring tasks. Workflow Orchestration: Implement process orchestration for multi-Agent execution via workflow mode to ensure stable and controllable task execution. | 2025-09 | |
New Nodes of Workflow | Database Node: support add, delete, modify, check and other SQL operations for already accessed databases in the application, making it user-friendly to perform data operations in the workflow. Agent node: support for references to Agents within application, reuse and orchestrate their capacity in the workflow. Variable Transform node: support JSON string serialization and deserialization, list data value extraction and sorting, as well as string variable concatenation and separation operations. Variable Assignment node: support assigning values to global type variables (such as application-level variables, workflow-level variables). | 2025-09 | |
Workflow Upgrade | LLM node separate system prompt and user prompt, support customizable output formats, and support prompt templates. Parameter extractor node supports extracting complex data structures such as object and array < object >. Option node supports customizable prompt content. LLM used nodes (LLM node, LLM Knowledge Q&A node, LLM Intent Recognizer node, LLM label Extrctor node,) add the "intermediate message" feature. Tool node supports dynamic parameter passing for path, adds calling methods, and automatically parses output variables. Knowledge Retrieval node and LLM Knowledge Q&A node support displaying "reference". Adjust the workflow status, remove the "Pending Debug" status, no longer enforcing restrictions such as requiring "debug" before testing in a test environment or being referenced as a sub-workflow. Added logic for "execution exception process" in the processing feature. | 2025-09 | |
Model Marketplace | Add Model Marketplace, supporting centralized discovery, configuration, and model management for use in application development. In addition to preset models such as Youtu LLM and DeepSeek, it also supports users to flexibly access another LLM services via API Key configuration. | 2025-09 | |
Application Template | Support Application Templates, offering multiple sample applications, supporting displaying key configuration points, making it easy to quickly reuse and reference. | 2025-09 | |
New Prompt Templates | Support custom templates, and during fill-in prompt, enable direct selection of existing templates or submission to the template library. | 2025-09 | |
Enterprise Management | The product provides three features: enterprise management, space management, and platform-side user permissions. It meets the hierarchical permission requirements for corporate groups and multi-project parallel operations, enabling team collaboration, data isolation between teams, and autonomous permission management. Learn the basic concepts in Business, Workspace and Permissions Overview. | 2025-09 | |
Workspace | | | |
Platform-Side User Permissions | | | |
Variables and Memory | Comprehensively integrates various variables in the Intelligent Agent development platform, maintaining them uniformly in "Variables and Memory", divided into "system variable", "environment variable", "API parameter" and "application variable" four types. Add new Long-Term Memory feature to implement personalized dialogue experience, supporting long-term memory content retrieval via SYS.Memory call. | 2025-09 | |
Model Comparison | Application development in standard mode and agent scenarios in Multi-agent mode supports "Multi-model comparison debugging", allowing simultaneous comparison of different models' response effectiveness for the same issue. | 2025-09 | |
New Document Segmentation Rules | Document segmentation supports default splitting rules and custom splitting rules: 1. Default splitting rules: The product uses model capability for segmentation. Default slicing rules do not support user intervention. 2. Custom splitting rules: Provide different splitting rule options, support user selection, including common identifier split, parent-child identifier split, and line-by-line split. | 2025-09 | |
Application Evaluation Upgrade | Comprehensive upgrade of application evaluation: Evaluation Set Enhancement: Supports evaluation with multiple evaluation sets, and custom columns can be added to flexibly adapt to different test requirements. Diverse Scoring Methods: Newly-added judge model scoring, rule scoring, and code scoring provide three ways to meet diverse evaluation scenarios. Comparative Evaluation: Supports effect comparison for multiple models or prompt content, enabling quick assessment of performance differences between different configurations on the same task. Evaluation Report Generation: Generate comprehensive effect and performance evaluation reports. | 2025-09 | |
Application Analytics - Dialogue History Supports Intelligent Classification | Application Analytics support intelligent classification of unclassified dialogue records. Based on actual interaction logs from the live production environment after application release. | 2025-09 | |
August 2025
Update | Description | Release Date | Documentation |
Tencent Cloud Agent Development Platform Update English Abbreviation | Tencent Cloud Agent Development Platform renamed English abbreviation from "TCADP" to "Tencent Cloud ADP". | 2025-08 | - |
Support user-customized models | Tencent Cloud Agent Development Platform allows users to customize their experience by adding LLM model APIs that comply with the OpenAI protocol. Users can integrate these custom LLMs as generative models, with supported options including ChatGPT, Claude, Gemini, LlaMa, Qwen, Doubao, and Kimi. | 2025-08 | |
Application mode upgrade | Application center newly upgraded, launch two application modes, and display in application list: Standard mode: The platform is built-in with best practice process, achieving more stable and precise Q&A effects. Suitable for enterprise knowledge service and product consultation scenarios. Single Workflow Mode: Use the assigned workflow to respond to all user dialogues. Suitable for relatively formal corporate service scenarios, customizing business processes by orchestrating nodes like knowledge Q&A and large models. | 2025-08 | |
Add Node to Workflow | Add batch processing node for parallel task execution to improve operation efficiency. Add Message Queue node to support interfacing with two message queues, Kafka and RocketMQ, and send a message. Add variable aggregate node. The variable aggregate node supports combining multiple workflow branches into one variable and transmitting the first not null variable value to subsequent nodes. ● Add plugin node, support users to select official plug-ins or specific tools in custom plugins in the canvas, and reuse capacity. | 2025-08 | |
Workflow supports single-node debug | Workflow supports debugging for specific nodes, suitable for complex content and a large number of nodes that require debugging one by one. Nodes supporting single-node debug: parameter extraction node, tab node, large model node, large model intent recognition node, large model knowledge Q&A node, large model tag extraction node, knowledge retrieval node, plug-in node, tool node, code node, workflow node. | 2025-08 | |
Plugin Center | The plugin center feature is now live, supporting users to use official plug-ins or custom plugins. Official plug-ins: Officially provided plug-ins such as "weather" and "online query" can be directly used in plug-in nodes. Custom plugin: Support registering self-owned APIs as plugins for use in workflows or Agents. | 2025-08 | |
Added support for MCP plug-in integration | Plugin center added support for quick integration with deployed MCP Server via MCP sse method. | 2025-08 | |
Add application evaluation feature | The application evaluation feature supports batch evaluation of the QA effect in knowledge base applications to ensure actual usage effect. | 2025-08 | |
Add application analytics | The application analytics feature supports performing data statistics on the application, viewing conversation history, and optimizing QA effect. Data statistics: Support viewing the message count of dialogues after application release, the number of users performing QA with the application (same user accessing via different channels cannot be identified for deduplication), the number of likes on application replies, and the ratio of liked messages to all evaluated messages. Support time filtering, distinguishing callers by filter criteria, and exporting statistics details. Dialogue history: Supports viewing conversation history after application release. Supports exporting conversation history by time. ● Unsatisfactory issues: When a user is not satisfied with the application's response and clicks the dissatisfaction icon, the Q&A record will be automatically uploaded to the unsatisfactory issue list in application analytics. The list supports search, filtering, batch ignore, and batch export. ● Rejected issues: Supports adding rejected issues. The model will judge similar semantic questions and reject them, replying with an unknown question reply message. | 2025-08 | |
Add new release channel feature | Application release supports interfacing with WeChat service account, WeChat Subscription Account, WeCom app. | 2025-08 | |
Add new knowledge base module | Add new "knowledge base" module, support creating and managing knowledge bases. One knowledge base can be used by multiple applications, and one application can refer to multiple knowledge bases. The existing knowledge base in the application serves as the default knowledge base within the application. Each application has one default knowledge base, which can only be used in this application and cannot be referenced by other applications. | 2025-08 | |
Support by minute-level settings for knowledge expiration time | Document/QA knowledge allows custom expiration time settings. After expiration, the knowledge will automatically expire. Custom expiration time supports minute-level configuration. | 2025-08 | |
Knowledge supports access to third-party databases | In the knowledge base, you can use the database connection method to integrate third-party databases. This is suitable for scenarios where data synchronization or real-time updates with an existing database system are required. | 2025-08 | |
Add Resolution Split Intervention | The default knowledge base and shared knowledge base in the application provide the parsing and splitting intervention feature, supporting manual intervention on the parsed and split slices. | 2025-08 | |
Add Excel retrieval enhancement configuration | Advanced configuration in the knowledge base supports enabling Excel search enhancement. Once enabled, it allows querying and calculating Excel data based on natural language, but may increase reply duration. Suitable for enhancing table retrieval effect when structured spreadsheet data is available. | 2025-08 | |
Enable knowledge status control | Knowledge management supports control over the enable status of documents and QA. | 2025-08 | |
Added support for document QA | Support importing documents in a dialog box and performing QA based on the documents. | 2025-08 | |
Add Role Command AI One-Click Optimization | After writing role commands, click one-click optimization to enrich role information. | 2025-08 | |
Welcome message settings New example issue | Welcome message settings support adding example questions, with a maximum of three, and allow selecting example questions during the first-time dialogue. | 2025-08 | |
Advanced configuration new intent attain priority | The intent model supports configuration of intent attain priority and selects the attain method based on priority. Sortable attain methods include: QA reply, workflow reply, document reply, large model reply. | 2025-08 | |
Add debug parameter feature | Debugging parameters support customization of API parameters, including parameter name, parameter description and parameter type. Debugging parameters can control modules such as search in the knowledge library and workflow. | 2025-08 |
Was this page helpful?
You can also Contact Sales or Submit a Ticket for help.
Yes
No
Feedback