Let AI Agents See the Business as It Happens — Alibaba Cloud EventHouse Now Generally Available

By Ken Meng

AI Agents now touch production—service, ops, risk, O&M. Yet models that reason still can't see the business as it happens.

Orders, payments, behavior, alerts, telemetry—all change by the second, scattered across queues, databases, and object storage with mismatched schemas. An Agent stuck on a static knowledge base can't answer "what just happened," "who's affected," "what next."

Turning live events into data Agents can understand, query, and invoke is now table stakes. EventHouse is built for this: EventBridge's cloud-native event lakehouse, unifying ingestion, governance, analysis, and Agent invocation in one Serverless service—closing the loop from Data-to-Data to Data-to-Agent.

It abstracts structured, semi-structured, and unstructured data from Kafka, RocketMQ, MySQL, and OSS into a standard event model, then puts it to work through three layers: Catalog, Analysis, Luma.

Catalog: Unified Event Data Governance

Most enterprises don't lack data—they lack a map of where it lives, what it means, and how it connects.

Catalog auto-discovers metadata across sources and manages schemas, permissions, and lineage in one place. Upstream schema changes are versioned automatically; when anomalies hit, engineers trace an event along its lineage through production, storage, and analysis.

It speaks open table formats—Iceberg, Hudi, Delta Lake—with access control down to the column. Keep your storage and compute; no rip-and-replace.

Analysis: Unified Analysis of Real-Time and Historical Data

Once governed, Analysis queries live events and history in one place—Kafka and RocketMQ streams, plus federated queries over MySQL, PostgreSQL, and Elasticsearch.

One SQL serves live streams and archived data alike. Federated joins span RDS dimension tables and OSS logs—no copying, filters pushed to the source.

Materialized views precompute hot queries for millisecond answers. Columnar compression cuts JSON and CloudEvents storage over 50% versus traditional databases. Storage and compute scale independently.

The old contradiction ends: events arrive in milliseconds; analysis no longer waits on batch sync.

Luma: Natural-Language Data Analysis for AI Agents

Luma is the AI-native analysis layer. Ask in plain language; it finds the right tables and fields, generates and validates the SQL, runs it, returns the answer.

General models can write SQL but don't know your fields. Is status_code = 4 a failed payment, a closed order, or a manual cancellation? No way to know.

The semantic layer lets you attach business descriptions, synonyms, and computed logic to fields. The Data Agent uses them to explore, generate, self-repair, and execute. Missing a time range or metric? Luma asks before it guesses.

An operator asks, "Any order-brushing in the last 30 minutes?" Luma finds the transaction tables and behavior logs, builds a correlated query across time, IP, and device, and returns suspect accounts with evidence.

Luma also exposes data over MCP, so your own Agents can query live business data within scope.

Scenario Example: E-Commerce Transaction Monitoring

Picture e-commerce monitoring: orders in RocketMQ, profiles in MySQL, gateway logs in OSS. The old way queries each separately and joins by hand—ad-hoc questions queue behind the data team.

With EventHouse, Catalog registers all three. Analysis computes success rate and latency over live windows; the rule engine alerts when failed orders spike. Analysts correlate orders, users, and logs to pin the cause to a channel, a region, or a bad deploy.

A business user asks, "Which channel's payment failures grew fastest in the last ten minutes?" No tables or joins to memorize.

Generally Available from July 23

EventHouse goes generally available on July 23, 2026, pay-as-you-go.

Preview users: review your regions, data volume, query frequency, and budget now. See the EventBridge billing docs for pricing and rules.

Sound familiar? EventHouse is the real-time, unified, governable data foundation your Agents need:

• Dormant data: piled up at scale, still hard to analyze;

• Lagging pipelines: cross-source queries on brittle ETL; answers trail the business;

• Intelligence gap: Agents that need direct access to business data;

• High adoption barrier: unified, real-time, governed context—without the heavy lift.

Community

Let AI Agents See the Business as It Happens — Alibaba Cloud EventHouse Now Generally Available

Catalog: Unified Event Data Governance

Analysis: Unified Analysis of Real-Time and Historical Data

Luma: Natural-Language Data Analysis for AI Agents

Scenario Example: E-Commerce Transaction Monitoring

Generally Available from July 23

Read previous post:

Read next post:

Alibaba Cloud Native Community

You may also like

Comments

Alibaba Cloud Native Community

Related Products

EventBridge

ApsaraMQ for RocketMQ

Function Compute

Serverless Workflow