AI Automators Agent

Route AI Automators fields through a configured AI Agent instead of calling an AI provider directly.

Project Overview

AI Automators lets a site builder attach an automator to an entity field so its value is generated automatically by sending a prompt straight to an AI provider's chat endpoint.
AI Agents is a different, richer building block: a configured agent with its own system prompt, tool-calling loop, and choice of provider/model. Until now, there has been no way to point a field automator at an agent. The two lived in separate plugin systems with no bridge between them.

AI Automators Agent adds that bridge: a new LLM: AI Agent automator type. Instead of calling a provider directly, it hands the automator's rendered prompt to a configured AI Agent and stores whatever the agent's own reasoning/tool-calling loop returns. Everything the agent can already do - multi-step tool use, retrieval-augmented search, a dedicated system prompt - becomes available to any field automator, with no change to the AI Agents
module itself.

Core Functionality

• LLM: AI Agent automator type, selectable on any field an AI Automator can already target, alongside the standard provider-calling rules.
• An AI Agent picker replaces the usual AI Provider / Model picker in the automator's advanced settings. The agent's own provider configuration governs how it runs, falling back to the site's default provider when the agent doesn't pin one.
• Works on plain and formatted text fields (string, string_long, text, text_long): the agent's answer is stored as-is, with a text format applied for the formatted variants.
• No changes to the AI Automators or AI Agents modules. This is a single additive plugin (with one derivative per supported field type) and is safe to enable alongside either module's normal usage.

Example use case

A site could configure an AI Agent with a retrieval-augmented-search tool (for example, the AI Search module's RAG search function) and a system prompt instructing it to find content related to whatever it's given, then use this module to automatically populate a "Related content" link field whenever an article is saved, the agent does the searching and reasoning, this module turns its answer into real field values so make sure the agent output is in correct format, sharing a output template with the model is a great way to ensure that. This is one example among many: anything an AI Agent can already do is now available to a field automator.

Technical Architecture

The module ships a base plugin class and a deriver, following the same base-class-plus-deriver pattern used elsewhere in the AI Automators ecosystem:

• A base AiAutomatorType implementation that resolves the configured AI Agent, runs it against the automator's rendered prompt, and stores the result according to the target field's type.
• A deriver that produces one plugin definition per supported field type (string, string_long, text, text_long), so each shows up as its own selectable automator type, scoped to the fields it's valid for.

Requirements

• Drupal core ^10.5 || ^11.2
• AI (ai), with its bundled AI Automators submodule enabled
• AI Agents (ai_agents)

Installation

Install as you would normally install a contributed Drupal module:

composer require drupal/ai_automators_agent
drush en ai_automators_agent

Usage

1. Configure at least one AI Agent (provided by the AI Agents module) with the system prompt, tools, and behaviour you want.
2. On the field you want automated, add an AI Automator and choose LLM: AI Agent as the automator type.
3. Fill in the usual base field and prompt, as for any other automator type.
4. Under advanced settings, pick the AI Agent to run. On text/text_long targets, a text format picker is also available.
5. Save. When the automator runs, it sends the rendered prompt to the agent as a single message, and the agent's final answer is stored in the field.

Similar Projects

• AI Automators is the base module this project extends with a new automator type.
• AI Agents provides the agentconfiguration this project routes automator prompts through.

Neither project currently offers a way to drive a field automator through a configured agent; this project exists specifically to close that gap.