Agent-Native Architecture
Tencent Cloud DataBuddy is not a plug-and-play product that simply adds an "AI plugin to a traditional platform." Instead, it is redesigned from the ground up for Agent scenarios, encompassing everything from the underlying engine to the top-level interactions:
Three Built-in Agents Ready to Use Out of the Box: The Engineering Agent, Governance Agent, and Analytics Agent are deeply integrated with the Tencent Cloud big data ecosystem. They feature a Skill system, a dedicated knowledge base, context optimization, and persistent memory, all of which are deeply optimized for the big data domain.
Agent Bidirectional Protection: It provides prompt injection detection, strong OBO permission binding, high-risk SQL interception, and precise "three-factor" interception, covering all risks in the OWASP LLM Top 10.
Continuous Evolution Flywheel: Each use accumulates knowledge and memory, leading to continuous improvement. The benefits of the model are consistently applied to the data platform.
Data + AI Integrated Platform
A single set of data and a unified computing framework simultaneously serve both data analytics and AI applications:
Unified DAG (Directed Acyclic Graph) Hybrid Orchestration: It orchestrates big data tasks and AI tasks (model training, inference, feature engineering) within the same workflow, with lineage fully connected.
MLOps (Machine Learning Operations) Full Lifecycle: It covers MLflow tracking, AutoML, large model experimentation, and feature management, spanning the entire machine learning lifecycle: training, evaluation, deployment, and monitoring.
Cost Optimization: It reuses CPU and GPU computing power and prevents data from moving back and forth between two systems.
Deployment Cycle: The end-to-end model deployment cycle is shortened from 30 days to 7 days, feature reuse rate is increased by 1.5 times, and development efficiency is doubled.
Unified Semantic Layer Unity Semantics
In the era of LLMs, the bottleneck lies not in the models themselves, but in achieving a better understanding of business contexts. DataBuddy addresses the three major hallucination pitfalls of NL2SQL (Natural Language to SQL) through a unified semantic layer:
Hallucination Traps | Semantics-Free Layer Manifestations | Semantic Layer Solutions |
Metric Ambiguity | The metric "revenue" has eight different calculation methods, and the Agent selects one at random. | Standardize metrics, entities, and business calibers. |
JOIN Disorder | Join keys are guessed, leading to SQL queries that execute successfully but return incorrect data. | Fix association relationships with semantic models. |
Incorrect Query Condition Writing | Query conditions such as "last quarter" and "east China region" are matched to data fields by guesswork. | Define dimension bindings through ontology modeling. |
Effect Comparison:
Without a Semantic Layer: NL2SQL accuracy is between 60% and 70%.
With a Complete Semantic Layer: NL2SQL accuracy is 90%+.
By continuously updating and validating semantics through the governance Agent (ensuring semantics never expire), DataBuddy addresses both types of issues - "lack of semantics leads to hallucinations, lack of governance leads to obsolescence" - which constitutes its core competitive advantage.
Multimodal Governance Comprehensive Governance Objects
TCCatalog (Tencent Cloud Catalog, the unified data catalog) centrally manages four types of assets through a single entry point:
Table Catalog: Tables, Views, Functions
Volume Catalog: Logs, Images, PDFs, Videos, Audio
Model Catalog: ML Models, Version Management
External Catalog: A mirror of external data sources, enabling querying without ETL (Extract-Transform-Load).
Supporting Capabilities:
End-to-End Data Lineage: Automatically collects SQL and integration tasks, with model lineage covering the entire pipeline from training to inference.
Data Quality: Full-pipeline monitoring of freshness, completeness, and consistency, with automatic anomaly detection.
Sensitive Data Identification: AI automatically tags and classifies data, enabling one-click review and confirmation.
Fine-Grained Permissions: Cascading across Catalog, Schema, and Object levels.
Significant Reduction in Total Cost of Ownership (TCO)
Tencent Cloud DataBuddy's integrated architecture reduces architectural costs, with specific manifestations as follows:
No Edition Fee: The difference between the Professional and Enterprise editions lies solely in the incremental unit price of billing items. Customers can start with "zero barriers," use the service first, and then pay as they scale.
Dual Billing: Pay-as-You-Go and Subscription: Using CU as the unified metering unit, customers can optimize costs independently by using pay-as-you-go for peak loads and subscription for steady-state usage.
AI Capabilities Independently Commercialized: Agent subscription starts from CNY 198 / user / month, with five tiers of credit packages available for on-demand purchase and elastic scaling of usage.
Cloud-Native Integration, Complexity Hidden: Basic resources are packaged into CUs and GUs (GPU Unit, GPU computing unit), with a unified billing outlet, so customers do not need to assemble bills from multiple sources.