Drupal AI 1.5.0-alpha2 Adds Chat Processor Selection to DeepChat
Drupal AI 1.5.0-alpha2 was released on 20 July 2026 with configurable chat processing for the module’s DeepChat block. Drupal AI core maintainer David Galeano Barrado highlighted the release in a LinkedIn post, drawing attention to the new AI Assistant API processor and its integration with DeepChat. The official release notes list 21 resolved issues since version 1.4.3. The alpha works with Drupal ^10.5 || ^11.2, while Drupal.org lists 1.4.5 as the current stable release covered by the Drupal security advisory policy.
The central change removes the DeepChat block’s direct dependence on the AI Assistant API as its only processing path. Site builders can select a registered ChatProcessor plugin in the block configuration, allowing the same chatbot interface to work with different processing logic. The selection is stored at the block-configuration level rather than exposed as a runtime switch for each conversation.
The DeepChat integration issue adds a processor selector to the block configuration and changes the API controller to execute the selected processor through the common interface. A related AI Assistant API issue introduces an AiAssistantApiProcessor plugin that makes configured AI Assistants available through the processor system. Its configuration includes assistant selection and options for streamed or verbose output.
David presented a plain AI Assistant, a FlowDrop workflow, and custom domain logic as examples of processing strategies that could sit behind the interface. The official release notes name the AI Assistant API processor and the DeepChat integration, while the other approaches require compatible processor implementations. The architectural change therefore creates an extension point rather than bundling every processing strategy with the release.
Other additions include an Extractive Question Answering operation type, an AI reranking processor for Search API, and batch embeddings intended to improve indexing performance. The RestrictToTopic guardrail gains semantic topic matching, while a new SensitiveContentStream guardrail can buffer and evaluate content during streamed output. AI Automators receive rich-text image-description support, and plain-text field widgets gain an option to enable MDXEditor.
The release also aligns AI observability output with the gen_ai.* OpenTelemetry semantic conventions and fixes missing final token-usage data in some streamed spans. DeepChat now rejects caller-supplied execution identity fields at its API boundary, and AI Automator queues process one worker item during each cron execution. Additional fixes cover missing built-asset warnings, the “Create new prompt” button in AI CKEditor configuration, and support for ai_json_schema elements inside off-canvas dialogs.


