Daily Digest

Daily Digest – 5 August 2026

2026 Capabilities Progress

  1. Page generation — The AI-powered Canvas page builder refactored its request processing to pause after each AI decision and stream progress to the browser, improving responsiveness during page builds [109]. The system also received a fix to prevent discarding entire page builds when the AI references non-existent regions [108].
  2. Context management — The Context Control Center in progress UI design work for subscription limit indicators and advanced toward its 1.0 release with documentation, QA scripts, and configuration improvements underway [143][144][145][146][147][148][149][152][153][156][159]. Token budget persistence and access control hardening are in active development.
  3. Background agents — Agent reliability improved with fixes for hostname filtering during unattended execution and expanded test coverage for externally-driven agent runs [84][83]. The ECA agent was migrated to modern tools architecture [61][66], and work continues on agent status streaming and tool execution limits [87][92].
  4. Design system integration — Canvas in progress support for pushing code components from headless codebases and enabling field additions to page entities [103][101][122]. Brand kit color support and component metadata handling remain in progress [102][124][127].
  5. Content creation and discovery — Interactive refinement capability is in progress for Field Widget Actions, enabling iterative content improvement through conversational prompts [2][3][32][34]. A chat memory plugin type was introduced for persistent conversation history [10][38][39]. Search indexing stability work continues with a data integrity issue under investigation [164].
  6. Advanced governance — AI translation added documentation and fixed pipeline infrastructure, with batch translation of referenced entities under review [165][166][167][168][169][171]. Work on guardrails configuration and automator status handling is in progress [4][13][18].
  7. Intelligent website improvements — No direct progress today.
  8. Multi-channel campaigns — The ECA integration added nine new operation actions covering text-to-image, translation, summarization, and video generation, enabling workflow automation across content formats [50][51][52][53][54][55][56][57][58][59][60][65].

Shipped

  1. Interactive refinement for AI-generated content — Field Widget Actions allows editors to iteratively improve AI outputs through conversational prompts before accepting them, giving more control over automated content creation, with work continuing on the 2.x branch [2][3][32][34].
  2. Chat memory foundation for persistent conversations — A new plugin type enables chatbots to store conversation history across sessions using server-side backends, supporting more sophisticated conversational experiences [10][35][36][38][39].
  3. Nine new AI workflow building blocks for ECA — Operations covering text-to-image, classification, translation, summarization, and video creation are now available as visual workflow actions without custom code [49][50][51][52][53][54][55][56][57][58][59][60][64][65].
  4. ECA agent modernization — The agent that generates workflow automations from plain language descriptions was migrated to structured tools and reasoning loops architecture [61][66].
  5. Postgres vector search performance fixes — Cosine search now properly utilizes embedding indexes, attribute columns are preserved during reindexing, and collection deletion queries were corrected [71][72][73].

Ongoing

  1. AI module releases 1.4.6 and 1.3.11 in quality assurance — Both versions are undergoing testing before release to ensure stability [8][9].
  2. Model Context Protocol adoption proposal pending approval — A substantial architectural change for rich structured chat in version 1.6.x awaits maintainer review before work can proceed [20][21][22][23][24][25][26][27][28][29].
  3. Context Control Center approaching first release candidate — Documentation, QA scripts, configuration video, and interface refinements are in progress toward the 1.0 milestone [143][144][145][146][147][148][149][160][161][162][163].
  4. Canvas headless architecture improvements under review — Brand kit color support, additional entity type compatibility, and AI chat tools integration are actively being developed [102][103][114][115][122][124][126][127][128][141].
  5. AI Search indexing completeness bug identified — Progressive indexing can drop content chunks when miscounting document splits, affecting search accuracy and requiring engineering attention [164].

AI (Artificial Intelligence)

The project delivered several important capabilities during this period. A major advancement was completed to enable interactive refinement in Field Widget Actions, allowing editors to iteratively improve AI-generated content through conversational prompts before accepting it. This work enhances content creation workflows by giving editors more control over automated outputs.

A new chat memory plugin type was introduced, providing chatbots with standardized mechanisms to persist conversation history across requests. Each chatbot can now opt into server-side memory backends or continue managing history entirely in the browser. This foundation supports more sophisticated conversational experiences.

Multiple bug fixes were addressed, including issues affecting guardrail configuration forms and automator status settings. Work continued on improving chatbot stability, particularly around interactions with modals and the Drupal toolbar.

Looking ahead, the team is preparing releases for versions 1.4.6 and 1.3.11 following quality assurance testing. A substantial architectural proposal for rich structured chat—adopting Model Context Protocol shapes—has been filed but remains postponed pending maintainer approval. This proposal would affect the 1.6.x branch once created.

Several issues remain under review, including concerns around reranking logic for chunked search results, request tag registration for better visibility, and proper handling of unconfigured AI provider states. These reflect ongoing refinements as the module matures toward stable releases.

How can I help on this project?

Prioritize maintainer review of the Model Context Protocol adoption proposal to unblock architectural planning for version 1.6. Ensure adequate testing resources are allocated for the pending 1.4.6 and 1.3.11 releases to maintain quality standards. Consider whether additional resources are needed to address the backlog of open issues currently under review.

Contributors: Abhishek Dhariwal (adaptative), Ahmad Khader (Ahmad-Khader), Ahmad Khalil (ahmad-khalil-imagex), Ann Mary Sruthy (annmarysruthy), Artem Dmitriiev (a.dmitriiev) [1], Joshua Fernandes (joshua1234511), Levente Besenyei (LeventeB), Miguel Guerreiro (gueguerreiro), Prabhavathi Vanipenta (prabha1997), Ricardo Castañeda (cadence96), Rob Loach (robloach) [1]

View project on GitLab: ai

Drupal AI Initiative

Activity this period focused on planning and quality assurance. A roadmap was established targeting DrupalCon Rotterdam in late September, with the goal of demonstrating a fully integrated AI-powered Drupal site. The planned demonstration will showcase AI search, content review, translation, editor chat, observability, and guardrails working together on multilingual content. Visitors will be able to launch the demonstration site themselves within seconds. According to the [META] 2026 AI Inside Roadmap to DrupalCon Rotterdam 2026 issue, most individual features already function; the remaining work involves packaging them together with recipes and demonstration content.

The quality assurance team held its bi-weekly meeting, documented in QA Bi-Weekly Meeting - 5th August 2026. Testing is underway on Drupal CCC and regular releases. The team noted that developers are not consistently flagging issues as needing QA review, and plans are being made to onboard new QA team members.

How can I help on this project?

  • Ensure the team has adequate resourcing and budget to complete the DrupalCon Rotterdam demonstration on schedule, including recipe development and demonstration content creation.
  • Support QA capacity expansion by approving onboarding time for new quality assurance volunteers joining the initiative.

View project on GitLab: ai_initiative

AI Integration - ECA

This module bridges Drupal's Event-Condition-Action (ECA) workflow system with the AI module's capabilities. Over the past 24 hours, two major deliverables were merged that significantly expand what users can do with the integration and prepare it for its 1.0 release.

Nine new workflow building blocks were added, covering operations like text-to-image generation, image and text classification, translation, summarization, and video creation. Users can now trigger these AI operations through visual ECA workflows without writing custom code. This completes the coverage of all supported AI operation types except one (image-and-audio-to-video), which remains niche. A final operation—Rerank—is in development but not yet merged.

The ECA agent—a tool that lets users describe workflow automations in plain language and have the system generate them—was migrated to a new architecture that uses structured tools and reasoning loops. This modernizes the agent's implementation and brings it in line with the upstream AI module's current design patterns.

Work also began on improving how users configure structured responses in chat actions, and on a new feature that will allow workflows to intercept and modify AI requests and responses in real time.

How can I help on this project?

  • Ensure testing resources are available to validate the new operation actions before the stable 1.0 release, particularly for media-heavy operations like video and image generation.
  • Prioritize a decision on whether the AI Interceptor feature warrants release alongside 1.0 or should follow in a later version.

Contributors: Ahmad Khader (Ahmad-Khader), Marcus Johansson (Marcus_Johansson) [2]

View project on GitLab: ai_integration_eca

Anthropic Provider

Activity this period focused on project infrastructure and new feature development. Issue templates for bug reports and feature requests were added to the repository to streamline how contributors and users report problems and suggest improvements. This will help the team triage and respond to feedback more efficiently as the module matures.

A significant feature addition is in review that would add three new capabilities to the module: token counting (to track usage and costs), citation support (to reference sources in AI-generated responses), and context management tools (to handle long conversations more effectively). These features are designed to be optional and disabled by default, meaning existing functionality remains unchanged for current users. The work represents a major expansion of the module's capabilities for organizations that need tighter control over AI interactions and associated costs.

How can I help on this project?

  • Ensure the team conducting the feature review has sufficient time allocated to thoroughly evaluate the token counting and context management additions before they reach production.
  • Clarify organizational priorities around AI cost visibility and citation requirements to help guide which optional features should be recommended for broader adoption.

Contributors: Carlos Ospina (camoa), Mykola Podpriatov (nickolaj)

View project on GitLab: ai_provider_anthropic

Microsoft Azure AI

A user interface improvement was merged during this period to clarify how model identifiers are used when connecting to Azure services. The model ID field description update helps users understand that the value they enter will be sent directly to Azure when making requests. This reduces potential confusion during configuration and should help prevent setup errors.

The change represents incremental progress in making the module more accessible to site builders configuring Azure AI integrations.

How can I help on this project?

With no outstanding issues reported in this period, consider checking whether the module has sufficient user feedback to guide its roadmap toward a stable release, or whether additional testing resources would help validate the configuration experience improvements now in place.

Contributors: Peter Törnstrand (peter_tornstrand)

View project on GitLab: ai_provider_azure

Postgres VDB Provider

The Postgres VDB Provider module saw three improvements merged during this period, addressing core reliability and performance issues. A fix was applied to resolve a problem where the cosine vector search was not properly utilizing the embedding index it creates, which improves search performance. Another change ensures that attribute columns are correctly preserved when collections are cleared and reindexed, preventing data loss during maintenance operations. Finally, a query issue affecting the deletion of items from collections was corrected, ensuring data can be removed reliably.

These fixes strengthen the foundation of the module's vector database capabilities, making it more dependable for production use in AI-powered search and content retrieval features.

How can I help on this project?

  • Support allocation of testing resources to validate these fixes in realistic production scenarios before wider deployment.
  • Ensure stakeholder alignment on prioritizing stability work as the module matures toward a stable release.

Contributors: calmforce [1], Ignacio Pérez Puertas (djnash2002) [1], Mykola Podpriatov (nickolaj), Sergiu Nagailic (Nikro)

View project on GitLab: ai_vdb_provider_postgres

AI Agents

The AI Agents module made significant progress over the past 24 hours, with version 1.3.3 released and several quality improvements merged.

Two key fixes were completed. First, a missing close button in the tool configuration popup has been resolved, improving the usability of the agent editing interface. Second, a technical error that prevented agents from correctly restoring conversation histories containing images has been fixed, ensuring agents can reliably continue conversations that include visual content.

The team also addressed issues affecting agent reliability: a security filtering inconsistency was resolved to ensure hostname restrictions work correctly in all execution modes, and additional test coverage was added to protect the external execution mode used by interactive interfaces. A new per-tool setting now allows administrators to prevent certain tools from being called multiple times in a single response, helping maintain control over agent behavior.

Open work includes a minor deletion bug affecting single-tool configurations, ongoing research into multi-agent communication patterns, and several proposals for advanced features such as streaming responses and improved progress reporting.

How can I help on this project?

Consider whether the project needs additional testing resources to address the CI pipeline stability issue. Review and approve or defer the three postponed feature proposals awaiting maintainer decision, as they represent significant architectural direction. Clarify requirements for the file size limit feature request to help contributors move forward.

Contributors: AKHIL BABU (AkhilBabu) [2], Alexis Martínez (alexismmd), Bharat Kelotra (bharatkelotra), Marcus Johansson (Marcus_Johansson)

View project on GitLab: ai_agents

Drupal Canvas

Activity this period focused on improving the headless architecture, developer tooling, and the AI-powered page builder. Two significant improvements were merged that streamline workflows for teams building decoupled frontends.

The headless SDK now uses a dedicated Canvas content endpoint instead of relying on an external Lupus dependency, giving the project full control over how content is delivered to decoupled frontends. The CLI tool also gained the ability to synchronize flexible codebase files (package configuration, shared libraries) when pulling content from Canvas back to local development environments, with backward compatibility for older Canvas installations.

The AI page builder's developer chat underwent internal refactoring to improve how it processes requests. Instead of running the AI agent to completion in a single server request, the system now pauses after each decision and streams progress back to the browser, which should improve responsiveness and transparency during complex page builds.

Several user-facing fixes were merged, including a correction for blank previews when editing content templates in view modes other than "full," and improvements to video thumbnail rendering.

Work is progressing on brand kit color management, support for additional entity types beyond pages, and tools integration in the AI chat interface. Several issues remain open around component metadata handling, link editing in translated content, and data migration for properties created before version 1.8.

How can I help on this project?

  • Prioritize decisions on which entity types beyond pages should support Canvas editing to guide the open entity type work.
  • Review whether the brand kit color management roadmap aligns with your organization's design system requirements before implementation advances further.

Contributors: **** (project_19391_bot_17a61b0edeafd5a3db3e8b2462b6c010), **** (project_19391_bot_cb443e60680723bba08a5351bf595251), Adam G-H (phenaproxima), AKHIL BABU (AkhilBabu) [1], Ben Mullins (bnjmnm), Bálint Kléri (balintbrews), Chandan Singh (chandu7929), Dave Long (longwave), David Bravo (davidbp), Harumi Jang (hooroomoo) [1], Jakub Piasecki (zaporylie) [1], Juan Correa (jucs7), Kunal Sachdev (kunal.sachdev), Matt Glaman (mglaman), Rajab Natshah (RajabNatshah), Rajendar Pothireddypally (RajendarReddy), wotnak [2]

View project on GitLab: canvas

Context Control Center (CCC)

The project continued refinement work toward its first release candidate. Multiple design and quality initiatives advanced during the period.

A UI indicator design for subscription limits was completed and closed. This work will help site builders understand when they have reached or are approaching capacity for context items attached to AI agents, and guide them to appropriate next steps. The completed design work feeds into implementation of token count persistence and budget usage displays.

Preparations for the 1.0 release continued with planning around QA test scripts, a configuration video tutorial, and documentation updates to reflect the module's naming approach. Several open proposals address code cleanup, permission hardening, and interface improvements ahead of the release candidate milestone.

Work in progress includes renaming plugins for better usability, decoupling UI descriptions from internal plugin text, and simplifying configuration by relying solely on token budgets rather than item count limits.

How can I help on this project?

  • Confirm resourcing for the QA testing window; the team has flagged a preference for 14 days lead time before the planned release date.
  • Review and approve the design decisions around subscription limits and token budgets to unblock implementation.

Contributors: Juan Correa (jucs7), Kristen Pol (kepol), Tamas Balog (tbalog), Tekla Aivazashvili (aivazashvilit)

View project on GitLab: ai_context

AI translate

The AI Translate module made progress on foundational quality improvements during this period. Documentation has been added to help users understand how to configure and use the module, addressing a previous gap in onboarding materials. Separately, automated testing infrastructure was repaired to ensure code quality checks run correctly, which supports more reliable future development.

Two proposals are under review to extend the module's translation capabilities. One addresses a bug where translating media entities (such as images) fails due to incorrect language handling. A related proposal would enable automatic translation of linked content, such as taxonomy terms referenced within an article, streamlining workflows where multiple related pieces of content need translation together.

How can I help on this project?

  • Ensure review capacity is available to evaluate the two open translation enhancement proposals, which could materially improve user workflows if validated and merged.
  • Consider user testing or feedback loops with teams translating media-heavy content to validate the proposed fixes meet real-world needs.

Contributors: Carlos Romero (o0Charly0o), Marcus Johansson (Marcus_Johansson) [1], Rob Loach (robloach) [1], szato (odi)

View project on GitLab: ai_translate

Langfuse

Development progress continues on the Langfuse module with work underway to expand its feedback capabilities. A new merge request has been opened to introduce functionality for collecting user feedback, which will be packaged as a dedicated submodule. This addition will enable organizations to capture structured feedback within their Langfuse implementation, supporting better measurement of user sentiment and product quality. The work is currently under review and has not yet been integrated into the main codebase.

No new issues were raised during this period, and the project continues its development phase without critical blockers.

How can I help on this project?

  • Ensure the team has access to subject matter experts or early adopters who can validate the feedback functionality against real-world use cases.
  • Clarify the project's release roadmap and stability goals to help prioritize remaining development work.

Contributors: Abhisek Mazumdar (abhisekmazumdar)

View project on GitLab: langfuse

References

[1] Drupal 11.4 DeepChat Block does not work with Webprofiler enabled (ai, Issue)

[2] Add interactive refinement capability to Field Widget Actions (ai, Issue)

[3] Add interactive refinement capability to Field Widget Actions 2.x (ai, Issue)

[4] AiAutomatorEntityModifier::entityHasConfig() ignores the AI Automator's disabled (status: false) setting (ai, Issue)

[8] QA of 1.4.6 and 1.3.11 (ai, Issue)

[9] Release 1.4.6 and 1.3.11 (ai, Issue)

[10] Add a ChatMemoryInterface for storing ChatBot chat history (ai, Issue)

[13] Undefined array key warnings in RestrictToTopic and RegexpGuardrail buildConfigurationForm() when creating new guardrail (ai, Issue)

[18] Add a setup-guardrails agent skill to configure guardrails, guardrail sets, and global guardrails (ai, Issue)

[20] Approve the Chat Fundamentals (MCP-shaped rich chat) issue set for 1.6.x (ai, Issue)

[21] [Meta] Rich structured chat: adopt MCP content and interaction shapes (ai, Issue)

[22] (Optional) Typed context references on user messages via input-side content blocks (ai, Issue)

[23] Citations/sources and suggested replies via MCP shapes (ai, Issue)

[24] Elicitation round-trip: MCP-shaped request/result objects and pause/resume on ChatProcessor (ai, Issue)

[25] ChatBlockRenderer plugin type with default renderers and entity cards (ai, Issue)

[26] ChatProcessor: structured response channel and block-aware message history (ai, Issue)

[27] Add MCP content blocks and a _meta bag to ChatMessage (ai, Issue)

[28] Add MCP tool annotations to tool definitions and a ContentBlockProducerInterface for rich tool results (ai, Issue)

[29] Add ChatOutput::setMetadata() and stop discarding chunk metadata in streamed reconstruction (ai, Issue)

[32] Issue #3586468: "Add interactive refinement capability to Field Widget Actions Automation" (ai, MR)

[34] Issue #3586468: "Add interactive refinement capability to Field Widget Actions Automation" (ai, MR)

[35] Issue#3586539: Cherry pick d8017c74 (ai, MR)

[36] Issue#3586539: Cherry pick d8017c74 (ai, MR)

[38] Issue #3586539: Add ChatMemory plugin type (ai, MR)

[39] Issue #3586539: Add ChatMemory plugin type (ai, MR)

[49] Simplify structured output configuration in the Chat action (ai_integration_eca, Issue)

[50] [Meta] Cover remaining AI operation types with ECA execute actions (ai_integration_eca, Issue)

[51] Add ECA execute action for Translate Text (ai_integration_eca, Issue)

[52] Add ECA execute action for Text to Image (ai_integration_eca, Issue)

[53] Add ECA execute action for Text Classification (ai_integration_eca, Issue)

[54] Add ECA execute action for Summarization (ai_integration_eca, Issue)

[55] Add ECA execute action for Speech to Speech (ai_integration_eca, Issue)

[56] Add ECA execute action for Object Detection (ai_integration_eca, Issue)

[57] Add ECA execute action for Image to Video (ai_integration_eca, Issue)

[58] Add ECA execute action for Image to Image (ai_integration_eca, Issue)

[59] Add ECA execute action for Image Classification (ai_integration_eca, Issue)

[60] Add ECA execute action for Audio to Audio (ai_integration_eca, Issue)

[61] Migrate the ECA agent to the Tools + ReAct architecture (create-only) (ai_integration_eca, Issue)

[64] Issue #3585014 Simplify structured output configuration in the Chat action (ai_integration_eca, MR)

[65] Resolve "9 different new actions" (ai_integration_eca, MR)

[66] Resolve "Migrate the ECA agent to the Tools + ReAct architecture (create-only)" (ai_integration_eca, MR)

[71] fix: #3609274 Cosine vectorSearch() defeats the embedding index it creates (ai_vdb_provider_postgres, MR)

[72] Issue #3586862 by ignacio.perez.puertas: Re-apply Attributes columns after clear+reindex cycle (ai_vdb_provider_postgres, MR)

[73] Issue #3553475: Fix query in deleteFromCollection(). (ai_vdb_provider_postgres, MR)

[83] Add test coverage for externally-driven execution (setLooped(FALSE)) (ai_agents, Issue)

[84] Disable hostname filtering does not work when looping is off (ai_agents, Issue)

[87] Expose the agent status stream as MCP-shaped progress to any chat consumer (ai_agents, Issue)

[92] Add limits on how often a tool can run (ai_agents, Issue)

[101] Allow fields to be added to the page entity type (canvas, Issue)

[102] Brand Kit color support (canvas, Issue)

[103] Support pushing code components from a headless codebase (canvas, Issue)

[108] Canvas AI: page builder discards the whole build when the LLM names a region that does not exist (canvas, Issue)

[109] Canvas AI: Run the component agent in the dev chat with setLooped(FALSE), driven by the browser (canvas, Issue)

[114] Replace Lupus CE API with a dedicated Canvas Headless endpoint (canvas, Issue)

[115] [META] Allow Canvas to be used on any content entity type (bundle), as long as it has certain entity type characteristics (canvas, Issue)

[122] feat(Page builder): #3568228 Add field_ui_base_route support for Canvas Page entity (canvas, MR)

[124] #3591838 brand kit color implementation (canvas, MR)

[126] feat(CLI Tool): Push components as external metadata when the Headless SDK is present (canvas, MR)

[127] Draft: #3591838 [PHASE 2] introduce color props (canvas, MR)

[128] feat: #3518272 support all entity types with configurable displays for content templates (canvas, MR)

[141] #3498525 Allow entity types other than node; add feature flag module for taxonomy terms. (canvas, MR)

[143] Rename entity bundle and target entity scope plugins for better DX (ai_context, Issue)

[144] Clean up CCC inconsistencies and DX issues before rc1 (part 3) (ai_context, Issue)

[145] [Meta] CCC MVP 1.0 roadmap (ai_context, Issue)

[146] Update ai_context project page and docs based on new naming approach (ai_context, Issue)

[147] Create CCC video showing how to configure it (ai_context, Issue)

[148] CCC QA scripts for 1.0 testing (ai_context, Issue)

[149] Create CCC rc1 release (ai_context, Issue)

[152] Persist context-item token counts and show token-budget usage in CCC UI (ai_context, Issue)

[153] Decouple context item form descriptions from scope plugin descriptions (ai_context, Issue)

[156] Harden CCC context item access checks and document the authoring trust model (ai_context, Issue)

[159] Remove max_items context limit and rely on the token budget (ai_context, Issue)

[160] Issue #3586271: Clean up CCC inconsistencies and DX issues before rc1 (part 3) (ai_context, MR)

[161] Issue #3586276: Rename entity bundle and target entity scope plugins for better DX (ai_context, MR)

[162] Issue #3586361: Decouple context item form descriptions from scope plugin descriptions. (ai_context, MR)

[163] Issue #3586356: Harden CCC context item access checks and document the authoring trust model. (ai_context, MR)

[164] Progressive indexing can drop tail chunks (ai_search, Issue)

[165] Add documentation (ai_translate, Issue)

[166] Fix Pipeline (ai_translate, Issue)

[167] InvalidArgumentException: Invalid translation language (und) when translating media entities (ai_translate, Issue)

[168] Resolve "Add documentation" (ai_translate, MR)

[169] Issue #3585537: Fix the Pipelines by adding composer.json (ai_translate, MR)

[171] feat: Add batch translation for referenced translatable entities in AiTranslateController (ai_translate, MR)

Period: 5 August 2026 to 6 August 2026  |  Generated: 6 August 2026 05:00 GMT

To view the full AI prompts used to generate this digest check this file: 24h_2026-08-05_prompts.txt

Looking for older digests? Browse all digests here.