Drupal AI 1.2.0 Adds Agent Progress Tracking, PDF Ingestion, and Streaming Tool Calls

The Drupal AI initiative has delivered substantial progress during weeks 36 and 37, with key updates aimed at refining the AI Agents architecture, enhancing user experience, and expanding content integration capabilities. These developments mark a significant milestone in the lead-up to version 1.2.0. The latest updates address longstanding beta blockers and introduce new features to support more complex, real-world use cases.

A major highlight is the introduction of a Progress Service for AI Agents, designed to provide real-time visibility into agent workflows. This service enables granular tracking of an agent’s status, including tools used, context management, and planned actions. It addresses challenges like long agent load times, where users might otherwise assume failure, and offers developers advanced debugging and logging capabilities.

The OpenAI provider in Drupal now supports PDF ingestion. While OpenAI has long processed PDFs, Drupal lacked native integration. With the 1.2.x release, users can upload PDF files in the Chat Explorer or via custom code and use them as context for AI tasks—greatly expanding input flexibility for agents.

Agents can now return structured JSON output by configuring a json-schema in the UI. This ensures predictable responses across use cases like automation or third-party integration, enabling developers to build more robust workflows atop Drupal AI services.

A new Chat History form element standardizes how modules manage interaction history. This reduces code duplication across tools like AI API Explorer, AI Agents Test, and AI Agents Explorer. It also lays the foundation for integration with workflow systems like ECA.

The Pexels AI module has been expanded to support dynamic image search and import from the Pexels stock photo library. If no relevant media is found locally, agents can now search Pexels, offer image suggestions, and trigger downloads—all from within the chat interface.

Streaming responses now support tool calling. Previously, streaming limited agents’ ability to decide on tool use until after message completion. The new system introduces callbacks triggered at stream end, allowing agents to respond dynamically and maintain a responsive interface for users.

Additional improvements include:

• Translatable first message for chatbots
• Common methods for Field Widget suggestions
• Simplified event handling code
• AI Drush command now supports multiline and paste-ready input
• Custom tool descriptions for agent-specific reliability

These updates, shared by contributor Paul Johnson, reflect Drupal AI’s continuing push toward stable, extensible, and intelligent integrations that can meet modern content and automation demands.

Source: Drupal AI Development Progress Week 36–37 by Paul Johnson (September 22, 2025)