Hybrid Manager 2026.3.0 release notes Innovation Release
This Innovation Release includes features and enhancements.
Release Date: March 12, 2026
Release cadence update
HM now follows a dual release strategy: LTS and Innovation Releases. For details on this new model, see Hybrid Manager dual release strategy.
Release highlights
Introduction of HM automations: This release debuts the
Automationsengine, a major step toward autonomous database management. The first available task isDisk Auto-Scaling, which allows the system to monitor disk usage and dynamically expand the primary storage based on project-level maintenance windows and human-in-the-loop approvals.New "Summarize" SQL aggregate function in AI Factory: You can now use the
summarize_text_aggregate()function in standard SQLGROUP BYqueries. This allows you to summarize text from multiple rows into a single LLM-generated response. The engine automatically handles large text volumes via a map/reduce approach to ensure you never hit model token limits.Automated daily cluster health reports: Hybrid Manager now automatically generates comprehensive health reports once a day for every cluster in your estate. These reports are retained longer than raw metrics, allowing you to track long-term health trends, spot anomalies, and share performance snapshots. You can even ask the chat agent to retrieve or compare reports for you.
Interactive AI insights via chatbot: The Hybrid Manager chatbot is now more deeply integrated into the operational workflow. It can now provide automated schema design recommendations based on running workloads and retrieve and compare daily-generated cluster health reports via simple natural language queries.
Enhanced migration risk assessment: To ensure higher success rates for data migrations, this release adds granular assessments for source databases. This includes monitoring replication slot health and identifying risky table configurations (Primary Keys, TOAST, and data types) that could impact Debezium-based CDC.
Advanced AI pipeline capabilities: AI Factory now supports multi-step, composable pipelines. Users can now chain complex tasks together, such as converting structured PDFs to images, extracting text via OCR, and summarizing the results via LLM—directly within a visual pipeline designer.
Features and enhancements
Hybrid Manager (HM) platform
| Type | Description |
|---|---|
| Feature | Introducing Automations, a powerful new engine that allows users to automate routine database operations. The initial release includes Disk auto-scaling, the first in a series of planned automated tasks. The new framework features project-level Maintenance Windows to schedule tasks, granular execution controls, optional human-in-the-loop approvals and an integrated Task Manager to monitor all automated activities from a single, centralized interface. |
| Enhancement | Added the option to configure dedicated object storage per project via the CLI using the new edbctl project create-object-store command. When creating a new project, you can now choose between using the default shared object storage or configuring a dedicated object storage, ensuring project-level isolation for Postgres backups. Existing projects are protected and will not be affected unless explicitly overwritten using a force flag. |
| Enhancement | Updated the PGAA version shipped with HM to 1.7. |
| Platform support | Added support for Kubernetes 1.3.4 and OpenShift 4.21. |
| Platform support | Added support for running Hybrid Manager components on ARM-based architectures, enabling deployments on modern ARM platforms (such as ARM-based Kubernetes nodes and cloud instances). |
| Bug fix | Updated the Azure ObjectStore authentication mechanism from Managed Identity-only to DefaultAzureCredential. This enables support for multiple authentication methods, automatically attempting various credentials until a successful connection is established. |
| Bug fix | Fixed an issue whereby Hybrid Manager attempted to create public load balancers when running in a private VPC, causing cluster provisioning failures. The system now correctly creates private load balancers when the environment is in a pure private network. |
| Bug fix | Fixed an issue in GUC (Grand Unified Configuration) validation where the system used unevaluated values instead of the final evaluated values applied by Postgres, ensuring more accurate configuration validation. |
| Bug fix | Fixed an issue whereby adding a replica cluster to a primary cluster failed following a major version upgrade due to database file incompatibilities. |
| Bug fix | Fixed an issue whereby replica cluster provisioning failed due to an error in the replication configuration enrichment logic. The reconciliation process was updated to ensure the replica lifecycle completes successfully. |
| Bug fix | Fixed an issue whereby the chat agent UI lost content when receiving rapid streaming of small chunks. |
| Change | The chat agent now supports nvidia/llama-3.3-nemotron-super-49b-v1.5. |
AI Factory
| Type | Description |
|---|---|
| Feature | Extended the AIDB pipeline framework to support multistep, composable pipelines. You can now chain preparer functions (e.g., PDF to text, HTML to text) so that the output of one step feeds into the input of the next, enabling flexible pipeline composition through the visual pipeline designer. |
| Feature | Added the summarize_text_aggregate() function to AIDB, enabling you to summarize text from multiple rows into a single LLM-generated response within a standard SQL GROUP BY query. The function supports configurable token limits, custom prompts, and automatically employs a map-reduce approach for large text volumes that exceed the model's token limit. |
| Feature | Added a PDF-to-image conversion pipeline step to AIDB. You can now process PDF documents — including those with complex table layouts — through a pipeline that converts pages to images and applies OCR (PaddleOCR) to extract text suitable for downstream embedding and retrieval. |
| Change | Starting with this version, GenAI Builder uses LangFlow as its default orchestration engine. |
| Enhancement | Extended the invalid model name error message — previously only available on the Internal Model Cluster page — to the external inference service configuration page. |
| Bug fix | Fixed an issue whereby AIDB volumes with a dash in the name couldn't be deleted due to double-quoting of identifiers. |
Observability and monitoring
| Type | Description |
|---|---|
| Feature | Added cluster health reports. These are automatically generated once a day for each cluster in your estate. You can examine health reports to spot anomalies or share a snapshot of your database performance with others. Since they are retained for longer than raw metrics and query statistics, they can also be used to track health over longer periods. The chat agent can also retrieve and compare reports - just ask it! |
| Feature | Added schema design recommendations via the chat agent. You can now ask the chat agent to give you schema design recommendations and it will assess how well-suited your schema is to your running workload and provide actionable recommendations to improve performance. |
| Feature | Added support for managing multiple external PGD groups across different Kubernetes clusters as a single logical cluster, enabling unified visibility and topology management in the Estate section. |
| Enhancement | The Disk Throughput chart now displays read and write throughput separately for more accurate monitoring. |
| Enhancement | EDB Postgres AI agent container images and Helm charts are now published to the community360 and standard repositories in addition to the enterprise repository, simplifying access for broader deployment scenarios. |
| Bug fix | Fixed an issue whereby the database filter was incorrectly required in the recommendation service API. |
| Bug fix | Fixed an issue whereby the Nodes Status total in monitoring included initdb and join pods, resulting in incorrect counts. |
| Bug fix | Fixed an issue whereby the breadcrumb path for External CNP Clusters was incorrectly displayed as /Projects/<project-name>/Detail/Details instead of the expected /Projects/<project-name>/Postgres Clusters/<cluster-name>. |
| Bug fix | Fixed an issue whereby API requests to the Estate endpoint failed with a 431 error due to excessively large request headers when processing estates with numerous projects. |
Migrations
| Type | Description |
|---|---|
| Feature | Added a replication slot health monitor that tracks the state of logical replication slots on the source Postgres database, reporting health status, replay lag, and disk retention for each slot in the downloadable metrics report. |
| Feature | Added a dedicated metrics fetcher that runs independently of schema change detection, enabling continuous monitoring of database metrics such as WAL generation rate at regular poll intervals. |
| Feature | Added a per-table tuple mutation rate assessment that estimates each table's contribution to WAL generation, enabling more granular migration planning for source Postgres databases. |
| Feature | Added a global WAL generation rate assessment that monitors the rate of WAL accumulation on the source Postgres database, helping estimate storage capacity requirements during migration snapshots. |
| Feature | Added a migration assessment that checks for risky combinations of primary key, REPLICA IDENTITY, and TOAST configurations in Postgres tables, identifying scenarios where Debezium CDC might silently miss changes. |
| Feature | Added a migration assessment that identifies Postgres data types unsupported or risky for Debezium-based CDC, providing early warnings about potential compatibility issues before starting a migration. |
| Feature | Added a public API endpoint for downloading Migration Portal assessment reports in JSON format, removing the need to use the GUI. |
| Enhancement | Improved Data Migration Service (DMS) snapshot performance by deferring primary key application until after the snapshot phase. Composite primary keys are now collected and applied in order post-snapshot, achieving performance on par with pg_dump and pg_restore. Also added support for deduplication between snapshot and streaming phases. |
| Bug fix | Fixed an issue whereby dismissing a Migration Portal project creation failure notification didn't remove the failed project entry from the backend, preventing you from creating a new Migration Portal project for the same schema. |
| Bug fix | Fixed an issue whereby schemas weren't listed for AWS RDS PostgreSQL and Aurora PostgreSQL databases in the Database Details page. The EDB Postgres AI agent was attempting to connect to the protected rdsadmin system database, causing schema collection to fail for all databases on the instance. |
- On this page
- Release highlights
- Features and enhancements