TIO AUTOPLAT · Data layer · DataHub 2.7

DataHubAI-oriented data governance platform

Assets, standards, quality, security and open services run in one governed chain, and governance results are operated continuously. Governed results are then supplied to agents through knowledge and MCP / Skill contracts, so AI reads governed semantics instead of guessing fields; AI also takes part in discovery, governance, modeling and data development.

1
governed chain: assets, standards, quality, security, open services
3 roles
separated; masking, audit and export control throughout
4
kinds of AI agents for discovery, governance, modeling and development
MCP
/ Skill contracts; AI reads governed semantics

* Lakehouse and development efficiency figures are typical-scenario estimates; actual results depend on project evaluation and deployment.

Data governance is turning toward AI

From storing data to using it, and from reports for people to consumption by systems and agents: governance is no longer a one-off project but the precondition for supplying AI.

Stitched tools, broken chains
Ingestion, storage, governance and development each run in a separate system, and data is copied between them.
Models, master data and knowledge in silos
The same customer means different things in different systems, and governance work cannot be reused.
AI has to guess
Without shared semantics and contracts, agents see table names, not the business.

One pipeline, five promises

Ingest → store → govern → serve → make it consumable by AI.

  • Ingest
  • Databases, files, CDC, streams and multimodal data flow into the lake through visual tasks.
  • Store
  • An Iceberg lakehouse plus an acceleration layer: one copy serves both detail storage and fast analytics.
  • Govern
  • Standards, models, metric definitions, quality checks and classified security become reusable assets.
  • Serve
  • Data services, open catalogs, subscriptions and event routes supply business systems securely.
  • AI-consumable
  • Knowledge outputs, training datasets and MCP/Skill interfaces let agents read governed data by contract.
IngestDatabases, files, CDC, streams and multimodal data flow into the lake through visual tasks.

AI takes part in governance and consumes its results

Not only is data used by AI; AI takes part in the data work itself.

AI discovery
Recognizes source structures and business meaning
AI governance
Suggests standards, finds sensitive fields, drafts quality rules
AI modeling
Turns business descriptions into subject models and metrics
AI development
Builds and debugs pipelines from natural language
Cursor
Claude
Trae
Qoder
Kimi Code
Lingma
WorkBuddy
DataHub
MCP · Skill

Popular AI coding assistants and agents connect through MCP / Skill and read governed data contracts within their permissions.

  • $ mcp call datahub.search_assets --q "receivables"
  • → 3 governed assets · definition: finance domain v2
  • $ mcp call datahub.describe fin_receivable
  • → 18 fields · 12 standards · quality 96 · sensitivity L2
  • $ mcp call datahub.query --metric overdue_amount --by customer
  • → row-level permissions applied · 25 rows
datahub · mcp
 

Traditional build vs DataHub

AreaTraditionalDataHub
GovernanceStandards, models and metrics maintained separatelyModel-native: modeling becomes assets, assets become knowledge
SupplyPoint-to-point interfaces, separate permissionsData services, subscriptions and unified approval
AIAgents guess field meaningsKnowledge + MCP contracts; AI reads governed semantics
StorageTwo copies in lake and warehouseIceberg lakehouse: one copy, many workloads
DevelopmentScripts scattered across toolsVisual DAG + AI data development, stream and batch
Master dataSeparate interfaces and reconciliationWorks with Maindata, sharing quality, lineage and open services

Five advantages

01
Lakehouse-native
One copy of data in one store for low-cost retention and high-performance analysis; in typical scenarios runtime efficiency improves about 2–20×, analysis feedback 10×+, and compute use drops 30%+.
02
Model-native
Modeling is the asset, and the asset is knowledge: standards, models and metrics are defined once and consumed directly.
03
Master data linkage
Works with Maindata: core business entities join one quality, lineage and open-service chain, maintained once and used consistently everywhere.
04
Built for AI
Agents for discovery, governance, modeling and development work together, raising productivity by roughly an order of magnitude.
05
Visual operation
Unified batch and streaming DAGs: design, publish and run the whole chain from ingestion to service on a canvas.

Eight capability domains

Ingestion and lakehouse ops
Visual tasks, CDC, file and multimodal ingestion, with the Iceberg lakehouse and acceleration layer run together.
Catalog and lineage
Metadata, inventory, and table- and column-level lineage for clear impact analysis.
Standards, models and metrics
Field standards, subject models and metric semantics registered in one place and materialized on publish.
Visual pipelines
Batch and streaming DAGs combining ingestion, SQL, quality checks, scripts and sinks.
Quality
Quality rules, trends and asset health scores, with issues fed back into pipelines.
Security
Sensitive-data discovery and classification, masking, encryption, separation of duties and access audit.
Services and sharing
Data services, open catalogs, subscriptions and event routes, supplied after approval.
Knowledge and training data
Knowledge-base sync, processing policies, recall debugging and training datasets for AIBase.

Real console screens

From cockpit to pipelines, security and training data — one workbench for the whole chain.

  • Processing workbench
  • Organized as ingest → design → run and debug → governance checks → publish services.
  • Cockpit
  • Assets, lakehouse health, ingestion tasks and acceleration at a glance.
  • Security overview
  • Sensitive-data discovery, classification and access audit in one place.
  • Training datasets
  • Multimodal, time-series and document datasets supplied to AIBase.
Processing workbench

Organized as ingest → design → run and debug → governance checks → publish services.

Layered architecture

Role portals
Workbenches for data development, governance, security and business consumers
Capability layer
Ingestion · Lakehouse · Processing · Catalog & lineage · Standards & metrics · Quality · Security · Open services · Knowledge · Master data linkage
Unified foundation
Iceberg lakehouse · Acceleration layer · Compute · Unified permissions and audit

Scales from a single node to small clusters and full clusters; supports domestic operating systems (Kylin, UOS, openEuler) and databases (Dameng, Kingbase, DWS).

Within AUTOPLAT

DataHub supplies governed data and semantics: AIDA asks data on it, AIHub reads catalog contracts, AIBase uses training datasets, and Maindata links master data.

Let AI use governed results safely

Book a demo to see one dataset go from ingestion and governance to secure use by agents.