VERA 4.3 — February 9, 2026

Compute Engine: VERA 4.3

Built-on: Apache Flink® 1.20

Overview

VERA 4.3 introduces major enhancements across AI capabilities, SQL and data management, and data ingestion. This release significantly expands Flink SQL functionality, adds vector search support, and improves AI model configuration and robustness.

Key updates focus on features that impact users directly, including:

• Vector Search: Real-time similarity search using the VECTOR_SEARCH function and Milvus integration.
• AI Model Context Handling: Configurable strategies for handling requests that exceed an AI model’s maximum context window.
• Materialized Tables: Automatically managed stateful representations of data for efficient stream processing.
• Unified Batch and Streaming: Native batch execution on Ververica Cloud for a single, unified platform.
• Enhanced SQL Functions: A significantly expanded library of built-in Flink SQL functions for string, JSON, and regex processing.

New Features

This release introduces capabilities that directly impact how you work with the Ververica Unified Streaming Data Platform, improving efficiency, performance, and usability across AI, batch, and streaming workloads.

Perform Vector Search

The new VECTOR_SEARCH function enables real-time similarity searches directly within your SQL queries. It uses Milvus as the underlying vector database to support AI-driven use cases like semantic search, recommendation engines, and retrieval-augmented generation (RAG).

Benefits:

• Simplifies the integration of vector databases into streaming pipelines.
• Enables low-latency semantic retrieval without external application logic.
• Supports complex AI workflows within a familiar SQL interface.

Supported Deployments: BYOC, Managed Cloud

Learn More: VECTOR_SEARCH (/managed-service/reference/ai-capabilities/vector-search)

Configure AI Model Context Handling

This release adds configurable strategies for handling requests that exceed an AI model’s maximum context window. You can now define how the system behaves when an input is too large, choosing between truncating the data or discarding the record.

Benefits:

• Improves the reliability of AI pipelines by preventing overflow errors.
• Provides better observability through optional logging of truncated or skipped records.
• Allows for finer control over data costs and model performance.

Supported Deployments: BYOC, Managed Cloud

Learn More: Model DDL

Expand SQL Function Library

VERA 4.3 greatly expands the built-in Flink SQL function library, adding powerful new tools for string manipulation, JSON processing, and regular expressions. This includes the JSON_ARRAYAGG function for aggregating rows into JSON arrays and REGEXP_EXTRACT_ALL for advanced pattern matching.

Benefits:

• Reduces the need for custom User-Defined Functions (UDFs).
• Simplifies complex data transformations directly in SQL.
• Improves compatibility with other SQL dialects and processing engines.

Supported Deployments: BYOC, Managed Cloud

Learn More: JSON_ARRAYAGG (/managed-service/reference/json-arrayagg)

Use the VARIANT Data Type

The VARIANT data type provides a flexible container for any supported data type. It is specifically designed to handle semi-structured data or evolving schemas where the exact structure of a column might change over time.

Benefits:

• Simplifies schema management for diverse data sources.
• Enables easier ingestion of semi-structured JSON or NoSQL data.
• Provides a middle ground between rigid schemas and schema-less processing.

Supported Deployments: BYOC, Managed Cloud

Learn More: VARIANT

Use Materialized Tables

Materialized tables are stateful representations of data that are persisted and automatically managed by Flink. They enable efficient stream processing for complex scenarios like temporal joins and windowed aggregations by maintaining intermediate data structures.

Benefits:

• Provides low-latency, fault-tolerant stream computations.
• Supports incremental updates that calculate only changed data.
• Simplifies the development of stateful streaming applications.

Supported Deployments: BYOC, Managed Cloud

Learn More: Materialized Tables

Orchestrate with Workflows

Workflows is a visual orchestration feature that allows you to build, schedule, and manage both batch and streaming deployments. It provides an integrated experience for managing the complete lifecycle of a streaming lakehouse.

Benefits:

• Simplifies the management of complex, multi-stage data pipelines.
• Supports both manual and periodic scheduling using cron expressions.
• Provides a clear graphical view of dependencies and execution status.

Supported Deployments: BYOC, Managed Cloud

Learn More: Workflow Scheduler

Execute Native Batch Jobs

Ververica Cloud now supports native batch execution, allowing you to run large-scale, finite jobs using SQL, Python, or JARs. This unification eliminates the need for separate batch engines and ensures consistent semantics across your entire data platform.

Benefits:

• Reduces operational overhead by using a single engine for all workloads.
• Ensures consistent business logic between real-time and historical processing.
• Simplifies infrastructure by removing the need for a separate Spark cluster for batch.

Supported Deployments: BYOC, Managed Cloud

Learn More: Batch Mode

Manage Multi-Environment CLI Contexts

The vvctl CLI now supports context commands, similar to kubectl. This allows you to define multiple environments (servers, users, and workspaces) and switch between them instantly without manually editing configuration files.

Benefits:

• Reduces the risk of accidental actions against production environments.
• Simplifies onboarding for users managing multiple workspaces.
• Improves security by standardizing authentication and environment handling.

Supported Deployments: BYOC, Managed Cloud

Learn More: CLI Reference

Connectors & Catalogs

This release expands the ecosystem with new catalogs and enhanced format support, focusing on lakehouse integration and MySQL CDC fidelity.

New Integrations

• Delta Lake Connector: Provides a centrally managed metadata layer for writing Flink data into Delta Lake tables, supporting consistent governance and exactly-once semantics. Learn More
• Apache Iceberg Catalog: Enables users to configure Iceberg backends once and reference tables consistently across all deployments. Learn More
• Databricks Unity Catalog: Integrates with the Databricks governance layer to allow discovery and management of lakehouse tables using fully qualified names. Learn More

Updates / Enhancements

• Canal-JSON Format: Now supports parsing native MySQL types when using the MYSQL_TYPE strategy. This improves the correctness of MySQL CDC data ingestion through Kafka by maintaining high schema fidelity. Learn More