Drupal AI 1.4.7 and 1.3.12 Fix Agent Memory and Automator Defects

The Drupal AI project released versions 1.4.7 and 1.3.12 on 19 August 2026, addressing defects in agent memory, AI Automators, agent tool configuration and DeepChat positioning. The 1.4.7 release notes list 11 resolved items: eight bugs, two tasks and one new feature. The six bug items listed for 1.3.12 also appear in 1.4.7. Neither release note identifies a security fix or advisory.

One shared fix addresses malformed agent-memory history that could cause providers to reject requests. The issue record says AgentMemorySummarizer could trim conversation history in the middle of a tool loop, leaving a tool-result message without the assistant message containing its matching tool call. Providers that validate tool-call sequencing can reject that history, with OpenAI cited in the issue as an example.

Another shared fix changes Field Widget Actions so each selected action targets only its assigned automator. The affected setup requires multiple automators on the same field, which can be configured by duplicating an automator configuration but is not exposed through the default UI. Before the fix, selecting one action could run all automators or a different automator and overwrite the field. Both releases also fix an agent tool configuration defect that could prevent removal of the only configured tool and produce an Undefined array key warning.

Both releases also correct a DeepChat positioning bug on Drupal 11 sites using the Navigation admin toolbar. When the floating chatbot was positioned at bottom left, the expanded Navigation sidebar could cover its header or input because the chatbot did not account for the sidebar's width. The issue describes a frontend positioning problem rather than a provider or chatbot-processing failure.

Version 1.4.7 includes five changes not listed for 1.3.12. These comprise fixes for Translate Text Explorer and an Interactive-section display problem after adding a tool, moving AI External Moderation under Safety & Compliance, documentation for TokenUsageDto, and thumbs-up and thumbs-down icons added to the shared AI icon set.