Product Overview
Last updated:2026-02-13 17:19:50
Tencent Cloud Agent Development Platform (Tencent Cloud ADP) is an AI application development platform for businesses. Regardless of whether you come from a business team or technical department, you can quickly build business-oriented smart customer service, knowledge assistant, operation analysis and other Intelligent Agent applications on the platform based on advanced LLM capabilities.
Built Intelligent Agents support various release channels and provide standard APIs/SDKs as well as integration methods with office platforms, making it easy for developers to seamlessly connect them to existing systems and work scenarios, empowering real business and helping enterprises build controllable, operable production-level AI capabilities.
Product Capabilities
Based on different scenarios, Tencent Cloud ADP offers three application modes:
Standard Mode: Use it to build one Intelligent Agent application, starting with simple QA or task processing scenarios, and support by gradually adding knowledge and plugin capabilities as needed. For details, refer to Standard Mode - Quick Start.
Multi-Agent Mode: Suitable for complex tasks where "a single Intelligent Agent cannot complete the task and requires small team coordination," such as comprehensively analyzing a batch of data to form decision-making recommendations. You can configure sub-agents for different stages like "problem analysis, data lookup, result calculation, risk control gatekeeping, and conclusion writing." The system automatically coordinates tasks in the background and finally provides a result after multiple rounds of checks. For details, refer to Multi-Agent Mode Quick Start.
Single Workflow Mode: Suitable for standardized business processes that have already been established, such as ticket routing, after-sales handling, and approval process. You can drag and drop steps like "determine whether it complies with conditions → call API to query data → provide handling suggestions based on results" to arrange a reusable process, allowing the Intelligent Agent to repeatedly execute the same pipeline, handling repetitive work with stability and efficiency. For details, refer to Single Workflow - Quick Start.
Note:
If you are responsible for upgrading a business line and hope to use large models not just for chat but to enhance service efficiency, knowledge service capability, or operational efficiency, Tencent Cloud ADP can serve as a uniform Intelligent Agent base. Based on existing documents and systems, start with a small scenario pilot (such as a knowledge assistant or smart customer service), then gradually expand to more business processes, achieving a closed development loop from PoC to production-level implementation.
Strengths
1. More Accurate Dialogue Effect
In real business, whether the Intelligent Agent's dialogue effect is good depends not only on the large model itself, but also on whether the system can accurately understand enterprise documents, efficiently retrieve related information and provide reliable answers. Tencent Cloud ADP has the following strengths:
Enterprise-level Document Understanding: Relying on more than two years of OCR and document parse technology accumulation from Tencent YouTu Lab, it supports 10+ file types and 200MB single document processing, performing intelligent identification and structured processing for complex image and text mix, tables, and formulas in layout. In scenarios such as manuals and rules and regulations, it can accurately extract text and Image Content from documents, providing a high-quality knowledge base for follow-up QA, while simultaneously supporting direct understanding and response to Image Content.
Self-developed Vector (Embedding) Model: Based on multi-stage training pipelines and refined data engineering, the self-developed 2B-level embedding model achieves SOTA level in C-MTEB benchmark tests. Combined with an LLM-based Reranker model (supporting 8k+ long text and hierarchical knowledge distillation), it delivers high recall and high precision retrieval performance.
Intelligent RAG Retrieval (Agentic RAG) and Structured Data QA: Not only can it "match a piece of text", but it can also answer complex issues across knowledge sources through multi-round retrieval, cross-document reasoning, and multi-tool collaboration. Combined with Text-to-SQL capability, it supports conditional filtering and aggregation QA on tables or databases with thousands of rows, enabling "querying documents, tables, and databases" to be completed in a unified manner within the same application. For the judgment and operation mechanism of Multi-Agent, refer to the self-developed YouTu-Agent open-source project.
2. Fully One-Stop Development and Operation Framework
Multiple Intelligent Agent Architectures: Support mainstream paradigms such as LLM+RAG, Multi-Agent, and Workflow, balancing simple QA and complex task agents.
Application Lifecycle Management: Supporting capability covers the entire chain from configuration, development, debugging, evaluation, to publishing and operation. A single platform handles everything from pilot to scalability.
Enterprise-Level Evaluation System: Built-in multiple assessment methods such as judge models, rules, and code. Support one-click batch comparison of different models, prompt content, and workflow versions to generate quantified effect reports.
Operation and Usage Management: Provides management backend for usage, data analysis, and dialogue logging, making it easy to refine operation and view use cost.
No Code / Low Code First: Most features are done through visual configuration, with a small amount of code added for complex scenarios, balancing business and technical team needs.
Rich templates and best practice: preset industry templates for education, media, health care, finance and more, as well as multi-scenario demos and "key points of application development" to reduce trial and error cost.
Explainable and Continuous Optimization: Each response comes with hit document fragments, similarity score, and retrieval path. In conjunction with application log analysis and evaluation capacity, help business and operators understand why the large model "answers this way", and continuously adjust knowledge organization and retrieval strategy based on actual effect.
3. Flexible Model and Open Ecosystem
Multi-model Unified Access: Support mainstream large models and self-built models with one-click import to Tencent Cloud ADP.
Enterprise-level Plug-ins and MCP Ecosystem: Built-in 150+ high-quality plug-ins cover knowledge q&a, retrieval, image/audio and video processing, and analysis scenarios, fully support MCP protocol and custom plugin expansion.
Multiple Integration Methods: The Intelligent Agent API provides RESTful API, SDK, and other integration methods, supports publishing to frequently used channels such as WeCom and mini program, and adapts to different technology stacks and access scenarios.
4. Enterprise-Level Security, Performance and Observability
High-performance architecture: Streaming response with first-byte time under 1 second, supporting high-concurrency and scaling.
Secure and reliable: Strictly isolated tenant data, content moderation, fine-grained permission system and CloudAudit.
Fine-grained permission system: Multi-level permissions for platform/space/application/knowledge base, with decoupled data permission and feature permission to fit complex organizational structure of corporate groups.
Comprehensive monitoring and Alarm: Monitor core metrics such as request volume, success rate, delay, and Token consumption in real time.
Intelligent Agent data insight capability: Intelligent Agent dialogue history analysis, hotspot issues mining, Q&A effect annotation, effect trend analysis.
Using Tikit
You can read the following tutorial to learn and master Tencent Cloud ADP usage.
Was this page helpful?
You can also Contact Sales or Submit a Ticket for help.
Yes
No
Feedback