Hybrid Manager 2026.4.1 release notes Innovation Release
This documentation covers the current Innovation Release of
EDB Postgres AI. See also:
- Hybrid Manager dual release strategy
- Documentation for the current Long-term support release
This Innovation Release includes features and enhancements.
Release Date: April 23, 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
Automations engine expansion: New Automations tasks let you auto-apply recommended indexes, auto-increase CPU and memory, and automate minor and patch database updates — all with configurable maintenance windows and human-in-the-loop approvals.
Expanded AIDB coverage in Langflow: New Langflow components expand the AIDB integration, including hybrid search for the EDB Knowledge Base component and new EDB Airman MCP Server, EDB Embedded Models, and EDB Vector Index components. You can also now deploy Langflow flows as managed workloads on HM with configurable resource constraints and full observability integration.
Langflow migration template: A new AI-powered Langflow template automates the end-to-end migration process from Oracle to Postgres, including provisioning HM-managed clusters, migrating schemas, and performing snapshot plus streaming data migration.
Unified schema migration: You can now migrate database schemas from Oracle or Postgres sources to Postgres destinations directly through Hybrid Manager, with support for Oracle SQL compatibility assessment and API-driven automation via AI agents.
Proactive query and schema change detection: HM now automatically detects queries running for the first time on a monitored database and identifies schema changes — both within one day — helping you spot unexpected modifications and investigate potential regressions early.
dbaasscenario separation: Database cluster management capabilities are now controlled by a dedicateddbaasscenario, separated from thecorescenario. Existing installations upgrading from 2026.3 must explicitly adddbaasto their scenarios list to retain Postgres cluster management functionality.Hybrid Manager apps: Deploy community database tools — including Apache Airflow, Apache Superset, Metabase, pgAdmin4, and pgBadger — directly to your HM projects from the Asset Library, with configurable parameters, scheduling, and approval workflows.
Release details
Hybrid Manager (HM) platform
| Type | Description |
|---|---|
| Feature | Introduced Hybrid Manager apps, a catalog of community database tools that can be deployed directly to your HM projects. The initial catalog includes Apache Airflow, Apache Superset, Metabase, pgAdmin 4, and pgBadger. Apps are deployed from the Asset Library > Apps section, with configurable parameters, optional deployment scheduling, and human-in-the-loop approval workflows. Deployed apps can be monitored, upgraded, and deleted from Estate > Apps or your project's Apps view. |
| Feature | Added the ability to automatically apply recommended database indexes via the Automations engine. Filter suggestions by databases, schemas or tables and choose to apply them with explicit human approval and/or during your Maintenance Window. |
| Feature | You can now configure HM-managed clusters in the Automations engine to automatically scale up CPU and/or memory resources when usage thresholds are exceeded. |
| Feature | You can now automate minor and patch database updates via the Automations engine. Set your preferred schedule and maintenance rules to ensure your HM-managed clusters stay current with zero manual effort. |
| Enhancement | Added advanced compression to HM backup capabilities through integration with Barman Cloud compression algorithms. WAL compression is now enabled by default to improve backup performance, reduce storage consumption, and prevent WAL storage full incidents. Defaults to Snappy compression for basebackup and zstd compression for WAL files. |
| Enhancement | Added optional read-only connection strings for High Availability (HA) clusters. You can now connect read-only operations to replica instances, reducing the workload on the primary instance and improving overall performance. |
| Enhancement | Added support for node selectors and tolerations in the cluster provisioning and patch API. You can now specify additional pod scheduling constraints to control where PostgreSQL pods are deployed within Kubernetes clusters, including the ability to target specific nodes and tolerate node taints for advanced workload isolation scenarios. |
| Enhancement | Added support for Custom Loadbalancer annotations in EDB Postgres AI agent (beacon-agent) provisioning, allowing you to configure load balancer annotations for database deployments across supported platforms. This functionality can be enabled or disabled at the location capabilities level. |
| Enhancement | Added the ability to provision clusters via API with custom load balancer annotations (metal-lb or any custom annotation), when this capability is supported by the location via EDB Postgres AI agent (beacon-agent) config. This is an API-only feature for this release; UI and CLI support will be added in the upcoming LTS. |
| Feature | Added the ability to automatically assign HM user roles based on IdP group attributes. You can now map LDAP user attributes or group names to roles in a declarative way, eliminating the need to manually assign roles for users in your LDAP directory. |
| Enhancement | The Chat Agent now becomes available automatically after you create an inference service. Previously, you had to refresh your browser to make the Chat Agent available. |
| Enhancement | Enhanced the database notification system with expanded event sources. The notification system can now notify you of events logged by the HM system, including agentic automated tasks generated by the platform. Additionally, all resource types in the audit log table are now clickable, extending beyond the previously available task-only functionality. |
| Enhancement | Enhanced the image library Clusters Using This Image view to only show clusters you have permission to access, based on your user role and project permissions. |
| Enhancement | Added support for GitLab container registry as a registry provider for image discovery. |
| Enhancement | You can now create HM-managed clusters through the Chat Agent, including High Availability, Advanced High Availability, and Distributed High Availability distributions. |
| Enhancement | Enhanced the activity log to include API key identifiers in audit log entries, improving identity metadata tracking at the gateway for better security monitoring and correlation. |
| Enhancement | The Chat Agent's functionality and security have been enhanced. The Chat Agent now uses a selection list interaction method which increases the certainty of interactive information, and uses a separate password transmission path which ensures passwords are not uploaded to the large language model. |
| Enhancement | Added support for deploying published flow bundles via the upm-flow-proxy service. The HM platform can now fetch flows directly from the internal flow proxy service as an alternative to cloud storage (S3, Azure, GCP), enabling streamlined deployment of published flow configurations. |
| Enhancement | Introduced the consolidated system database (upm-system-db). The operator preflight check now validates that the required database credential secrets (db-upm, db-dex, db-langflow, db-lakekeeper, db-beacon-app, db-transporter, db-superuser) exist before installation or upgrade. These secrets are auto-generated by edbctl setup create-install-secrets when targeting IR 2026.4 and later versions. |
| Enhancement | Database cluster management capabilities — including provisioning, backups, storage locations, and cluster templates — are now controlled by a dedicated dbaas scenario, separated from the core scenario. New installations include dbaas by default when no scenarios are specified. Existing installations upgrading from 2026.3 must explicitly add dbaas to their scenarios list after upgrading the operator to retain Postgres cluster management functionality. Without dbaas, cluster management APIs return 404 and the Clusters section is hidden in the HM console. |
| Change | Changed the default value of the edb_stat_monitor.edbsm_normalized_query GUC from false to true. This change ensures that values used in predicates aren't saved in HM by default, providing better data privacy for customers who don't want sensitive query data to leave their databases. |
| Deprecation | Removed the Apache Spark component bundled with Hybrid Manager. Note that connecting to external Spark clusters remains supported. |
| Deprecation | Deprecated the beaconServer.additionalTrustDomains field in the HM CRD. This field is no longer used by the upm-beacon Helm chart and has been replaced by dedicated Helm values configuration. |
| Security | Enhanced network security policies for the Langflow component to restrict access to AI model server APIs. Langflow can now access only read-only discovery APIs for model server clusters and inference services, while blocking dangerous mutating operations such as create, update, and delete. |
| Security | Fixed a security issue whereby SPIRE server audit logging was disabled, which prevented logging of security-relevant events such as SVID issuance, entry modifications, and federation operations. The SPIRE server now has audit logging enabled to meet SOC2 and PCI DSS compliance requirements. |
| Security | Enhanced project-scoped authorization for published flows. The upm-flow-proxy now enforces project-level permissions, ensuring users can only access and modify published flows within their authorized projects. When creating flows, users must have access to all specified projects. When updating existing flows, users can only add or remove projects they have access to, while projects outside their scope remain unchanged. |
| Bug fix | Fixed an issue whereby ClusterWrapper could become stuck in the soft deleting phase due to a race condition. |
| Bug fix | Fixed an issue whereby component defaults weren't applied in the upm-beaco-ff-base component, which could cause the accm-server to crash loop due to missing configuration values for usage_generator_interval, AES_KEY_ROTATION_INTERVAL, and SESSION_DURATION_SECONDS parameters. |
| Bug fix | Fixed an issue whereby read-only connection tests failed on Google Kubernetes Engine (GKE), Amazon Elastic Kubernetes Service (EKS), and Red Hat OpenShift (RHOS) platforms due to unprotected null reference errors when no PostgreSQL status store existed. |
| Bug fix | Fixed an issue whereby beaconServer.logLevel defaulted to info, overriding user-configured values. The setting now correctly respects the user-configured value. |
| Bug fix | Fixed an issue whereby a pod was crash looping on RHOS environments due to DNS resolution failures when the SPIRE agent attempted to connect to the kubelet via node hostname, which was not resolvable in cluster DNS. |
| Bug fix | Fixed an issue whereby the EDB Postgres AI agent (beacon-agent) produced confusing error messages when re-enrolling servers with newer agent versions against older HM servers. The agent now handles version compatibility more gracefully and provides clearer error reporting when server APIs are unavailable. |
| Bug fix | Fixed a bug where AI Pipelines didn't correctly detect deleted source records on volume sources. This bug would cause orphaned destination records to be left behind, and is now fixed. |
| Bug fix | The upm-beacon-server enters a crashloop when the spire component is broken or unavailable. Because spire is required for all HM deployments — which are configured to support secondary locations — a broken spire component prevents the deployment of Postgres clusters. |
| Bug fix | Fixed an issue whereby the credentials creation process in the edbctl setup create-install-secrets command had non-deterministic ordering, causing namespaces and secrets to be created in different orders across runs. |
AI Factory
| Type | Description |
|---|---|
| Feature | Expanded AIDB coverage in Langflow with hybrid search support added to the EDB Knowledge Base component, combining text matching with vector similarity search. Three new components were added: EDB Airman MCP Server; EDB Embedded Models, for CPU-based text and image inference using AIDB's embedded models; and EDB Vector Index, for creating vector indexes on EDB knowledge bases. |
| Feature | Added a Langflow template for AI-powered, end-to-end database migrations. The template automates the complete migration process from Oracle to Postgres, including provisioning HM-managed clusters, migrating schemas, and performing a snapshot plus streaming data migration. You can customize the migration flows to integrate your own solutions as needed. This initial version supports single database migrations with fully compatible source schemas and requires GPT-4o or better reasoning-class models. |
| Feature | The AI Pipelines component of the AIDB extension for Postgres now supports multiple compatible pipelines writing into a single shared knowledge base, enabling hybrid knowledge bases from multiple sources and formats — including multi-model (image and text) — each with independent processing steps. The vector table includes a new pipeline_id column to track source pipelines per embedding, and retrieval functions return the originating pipeline name. |
| Feature | Added support for deploying Langflow flows to Kubernetes environments managed by Hybrid Manager. You can now deploy your Langflow flows with configurable quality-of-service constraints including memory, CPU allocation, autoscaling, and restart policies. Deployed flows include full observability and alerting integration with Hybrid Manager and can be managed through the HM interface with deploy, undeploy, and status monitoring capabilities. |
| Feature | Added hybrid search capabilities to the AI Pipelines component of the AIDB extension, combining full-text search, vector/semantic search, and predicate filtering. Previously, only vector/semantic search was supported. You can now perform advanced retrieval queries from AI Pipelines Knowledge Bases using custom SQL queries. These queries may include arbitrary relational filtering or sub-queries and can also perform full text search using a bm25 index. AI Pipelines includes convenience functions to help build those queries, and documentation contains examples for common use cases. |
| Enhancement | Added an error log and dead letters queue to improve AI pipeline error handling. You can now track which files or records are being processed successfully and which files encounter errors during pipeline execution. |
| Enhancement | Added non-blocking error handling to AI pipelines. Pipelines now continue processing when errors affect only individual records (such as malformed PDF files or invalid data), while still halting on pipeline-level errors like missing models or configuration issues. You can query and manage pipeline errors using the new aidb.get_error_logs() and aidb.clear_error_logs() functions. |
| Enhancement | Added support for creating AI data pipelines for on-premises databases and external CloudNativePG clusters. You can now create control plane pipelines for unmanaged estates, expanding pipeline capabilities beyond HM-managed database instances. |
| Enhancement | Added AIDB pipeline support for external Postgres clusters managed by the Hybrid Manager (HM) platform. You can now build AIDB pipelines on these external clusters, including clusters running on secondary locations of HM and clusters running outside of the HM platform, with access to model and inference services from the primary location. |
| Enhancement | The AI Pipeline Designer now retrieves available models dynamically instead of using static default settings. This change enables model selection based on specific project, resource, and database combinations while maintaining the same user experience. |
| Enhancement | Added support for configuring the PDF to Image pipeline step directly in the visual AI Pipeline Designer. You can now set conversion options such as format, DPI, page ranges, and annotation rendering through the designer interface. |
| Enhancement | Added support for EDB AIDB pipeline extension v6.1.0 in Hybrid Manager, bringing the latest pipeline capabilities and enhancements to your AIDB workflows. |
| Enhancement | You can build your AI workflow using LangFlow, and publish your flow as assets on the Hybrid Manager platform. You can then run these flows as standalone jobs on selected locations within your projects. |
| Enhancement | Added support for Hugging Face as a new model provider. You can now configure AI models using Hugging Face Model Name, Object Storage path, or both when adding new models to the platform. |
| Enhancement | Enhanced error handling in the AI model creation form to display detailed error messages, improving user feedback when model creation requirements aren't met. |
| Security | Fixed a remote code execution vulnerability in the Langflow component whereby an attacker could execute arbitrary commands through the data parameter in the build_public_tmp endpoint. |
| Bug fix | Fixed an issue whereby the AI model service couldn't be listed from the Visual pipeline builder on RKE, GKE, and AKS platforms due to network policy restrictions preventing Postgres pods from accessing the model server API. |
| Bug fix | Fixed an issue whereby the AIDB Pipeline Designer provisioning page displayed clusters from all projects instead of filtering clusters to show only those from the specific project. |
Observability and monitoring
| Type | Description |
|---|---|
| Feature | Added detection of queries that haven't been executed on a monitored database before. HM now identifies new queries within one day, helping you verify they aren't causing issues or easily find the root cause if they are. The feature includes mechanisms to disable detection and suppress excessive notifications, such as when first adding a database where all queries would initially appear as new. |
| Feature | Added schema change detection that automatically identifies changes in monitored database schemas within one day. You can now spot unexpected schema modifications and compare database states before and after changes to identify potential regressions. |
| Feature | Added a new comparison tool in the HM console that lets you compare database metrics and query statistics between two user-defined time periods. You can now easily identify performance differences with side-by-side or overlaid chart views, apply filters and zoom controls, and see highlighted differences to understand the impact of changes over time. |
| Enhancement | Extended the monitoring dashboard annotation capability to support user-inserted custom events. You can now manually mark interesting platform and database activities alongside system-generated events, making it easier to correlate changes with performance trends. |
| Enhancement | Extended observability support for external PGD clusters to include CNPG-GC clusters. Building on the monitoring capabilities introduced in previous releases, you can now monitor metrics from both self-managed PGD and CNPG-GC clusters through HM observability features. |
| Enhancement | Added a new EnableEDBUsageReporting configuration option for the beacon server to provide more explicit control over usage reporting. The beacon server now includes retry logic that attempts to reconnect to IT links every 24 hours when they're unreachable, improving operation reliability in various environments including air-gapped setups. |
| Enhancement | When a cluster spec is updated — whether by a user change, autoscaling, or an agentic automation — an annotation is automatically generated and displayed in the observability view. |
| Bug fix | Fixed an issue whereby the EDB Postgres AI agent (beacon-agent) failed to start due to an invalid Postgres DSN error when the port was specified as 0, which is outside the valid port range. |
| Bug fix | Fixed an issue whereby metric time and values showed discrepancies between cluster-level and node-level views in the HM console. The backend now returns consistent data across both views and aligns timestamps correctly for node-level metrics. |
| Bug fix | Enhanced the robustness and performance of the internal metrics statistics database, optimizing its storage and memory usage. |
| Bug fix | Fixed an issue whereby the WAL storage percent metric displayed incorrect values above 100% in monitoring charts. The metric now accurately aggregates WAL usage by primary nodes only, providing proper percentage calculations. |
HM console
| Type | Description |
|---|---|
| Feature | Added AI-powered query generation to the Query Editor. You can now select Generate with AI to create SQL queries using natural language prompts. The feature generates valid SQL targeting data across your estate that you're authorized to query, making it easier to explore and analyze your data without writing complex SQL syntax. |
| Enhancement | Added the ability to view database health reports directly in the UI without downloading the PDF. You can now preview report contents inline before deciding to download. |
| Enhancement | Added a configuration interface in the HM console for HM authentication policies in the Settings section. Organization owners can now configure authentication policy settings, including idle timeout for web login sessions. |
| Enhancement | Added an Invalidate User Sessions action to the Identity Providers management interface. You can now invalidate all active user sessions for a specific identity provider through the HM console, providing administrators with better control over user access management. |
| Enhancement | Added filters in the role assignment interface to display only manually assigned roles when showing current user permissions. Roles assigned via IdP group mappings are now excluded from this view to reduce noise. |
| Bug fix | Fixed an issue whereby machine users couldn't be filtered correctly in the HM console. The filtering now uses the proper identity provider parameter instead of a general query parameter. |
Analytics
| Type | Description |
|---|---|
| Enhancement | Updated the PGAA version shipped with HM to 1.8. |
| Enhancement | Enhanced Lakekeeper warehouse creation to support object storage environments that don't support AWS Security Token Service (STS). The system now automatically retries warehouse creation with STS disabled if the initial attempt with STS enabled fails on S3-compatible storage, improving compatibility with providers like Vultr's Ceph Object Gateway. |
Migrations
| Type | Description |
|---|---|
| Feature | Added the ability to enter and display metadata about applications that use each database and schema. You can now associate applications with databases and schemas in many-to-many relationships, providing better context for database migration planning and prioritization. Application metadata can be entered manually through the Hybrid Manager interface or programmatically via the Hybrid Manager APIs. |
| Feature | Added the ability for the EDB Postgres AI agent (beacon-agent) to automatically configure Oracle and Postgres database connections using established sources of connection information, including Oracle TNSNAMES.ORA files and CSV files. The option to auto-discover RDS Postgres databases and configure the agent connections to them is also available. |
| Feature | Added unified schema migration capabilities to migrate database schemas from Oracle or Postgres sources to Postgres destinations directly through Hybrid Manager, eliminating the need for external tools. The service supports both HM-managed and externally managed Postgres destinations and integrates with data migration workflows for schema-only, data-only, or combined migrations, with API access for AI agent automation. When migrating from Oracle to EDB Postgres Advanced Server, compatible schemas are retrieved from the Migration Portal. |
| Feature | Added the ability to assess Oracle application SQL statements for compatibility with EDB Postgres Advanced Server. You can now assess SQL statements extracted from Oracle trace files and system views, as well as those provided directly through user interfaces or programmatic APIs. This capability includes support for MCP tools and introduces a new application information object within the Hybrid Manager migration architecture to serve as a container for SQL statements and assessment reports. |
| Enhancement | Added schema parsing, storage, and migration capabilities to the Migration Portal. When a Postgres database is registered with the Hybrid Manager for migration purposes, Postgres database schemas are now automatically ingested, parsed, and stored in the Migration Portal backend database. Those schemas are then available for migration with new API behaviors. |
| Enhancement | Extended the Migration Portal assessment reports API to support bulk downloads. The new endpoint accepts multiple database resource IDs and returns assessment reports for all specified resources in a single JSONL file, making it easier to download reports at scale. |
| Enhancement | The migration agent now honors upstream cancellation signals when extracting Oracle schemas. Previously, when the EDB Postgres AI agent (beacon-agent) received a termination signal during Oracle schema extraction, the sqlplus process would continue running until completion, wasting resources on work that couldn't be delivered. The agent now properly terminates the sqlplus subprocess when receiving cancellation signals, enabling cleaner shutdown behavior. |
| Enhancement | Added support for replicating PostgreSQL DOMAIN types based on DATE, TIME, TIMESTAMP, INTERVAL, UUID, JSON, XML, and POINT types. DOMAIN types are user-defined types that wrap existing PostgreSQL types with additional constraints. Both snapshot and CDC replication modes now correctly handle these DOMAIN types, whether the target uses the same DOMAIN or just the underlying base type. |
| Bug fix | Fixed an issue whereby Oracle schema collection reported incorrect table counts when the EDB Postgres AI agent (beacon-agent) connected as a non-privileged user. |
| Bug fix | Fixed an issue whereby a null string exception occurred in the GetMigrationDatabaseAssessmentData function when DDL extraction for PostgreSQL databases failed during assessment, causing API requests to return a 500 error. |
| Bug fix | Fixed an issue whereby database features weren't being collected during Oracle assessments in certain environments. |
| Bug fix | Fixed an issue whereby the gRPC UI encountered errors when schemas had dependencies. |
| Bug fix | Fixed an issue whereby nested button elements in the Migration Databases empty state caused invalid DOM structure warnings. |
| Bug fix | Fixed an issue whereby the Create OLTP Migration form didn't show error messages when source table loading failed, leaving users without feedback on what went wrong. |
| Bug fix | Fixed an issue whereby CDC replication failed when PostgreSQL table or column names contained multi-byte UTF-8 characters such as Japanese, Chinese, or Korean characters. |
| Bug fix | Fixed an issue which caused the migration of wide tables to fail with no recovery. |
| Bug fix | Fixed an issue whereby the reader would throw an error when uploading objects to S3 after bumping the Netty dependency. The error occurred due to connection pool timeout issues with the async S3 client, which has been resolved by switching to a synchronous S3 client. |