AIDB 7.2.0 release notes

AIDB 7.2.0 release notes#

Released: 30 Mar 2026

AIDB 7.2.0 includes the following enhancements and bug fixes:

Type

Description

Enhancement

Multi-pipeline knowledge bases: Knowledge bases can now be associated with multiple pipelines, enabling richer and more flexible data ingestion and retrieval architectures.

Enhancement

Hybrid search (BYOSQL): Introduced Tier 1 hybrid search support with a bring-your-own-SQL (BYOSQL) interface, allowing custom retrieval logic to be combined with semantic search.

Enhancement

Retrieval support for multistep pipelines: Retrieval operations are now supported within multistep pipeline configurations, enabling end-to-end RAG workflows across pipeline stages.

Enhancement

HuggingFace API support: Added support for the HuggingFace Inference API as a model provider for embeddings and other AI tasks.

Enhancement

Model options: Model configurations can now include provider-specific options, giving finer control over model behavior such as temperature, max tokens, and other parameters.

Enhancement

Pipeline Dead Letter Queue (DLQ) — Phase 1: Introduced the first phase of Dead Letter Queue support for pipelines. Processing errors are now logged for failed pipeline operations, enabling inspection and reprocessing of failed records.

Enhancement

Configurable CPU limit: Added a configurable CPU limit for background pipeline workers to prevent resource contention in shared environments.

Enhancement

HCP model sync hints: When listing models that require HCP synchronization, AIDB now displays a helpful hint guiding users on how to sync models from HCP.

Enhancement

Optionally versioned interfaces (OVI): Pipeline and knowledge base views now support optionally versioned interfaces, allowing clients to pin to a specific API version.

Enhancement

Improved deprecation UX: Deprecated functions now provide clearer messages with guidance on replacement functions and migration steps.

Enhancement

`functions` tag on `aidb.models` view: The aidb.models view now includes a functions tag, making it easier to identify models that support function-calling capabilities.

Bug fix

Fix volume listing with directories: Corrected an issue where volume listing failed when directories were present alongside files in a volume.

Change

Volume names with quoted identifiers blocked: Creating volumes with names that require quoted identifiers (for example, names containing spaces or special characters) is no longer allowed, to prevent ambiguity in SQL references.