Dalia

A power tool for Drupal developers.
Ship code you truly own — faster than you thought possible.

How It Works Get Started
dalia.drupal-contrib.ddev.site
Dalia's new generation form with base branch selector, example prompts, and task description

LLM training data for Drupal is a garble of versions — Drupal 7 patterns, deprecated APIs, mixed conventions. Without deliberate intervention, AI‑generated Drupal code lacks the specificity that professional work demands.

Dalia addresses this through a structured pipeline that puts the human developer at the center of every critical decision.

Zero additional cost.

Dalia runs on your existing Claude Code subscription. Predictable billing — no API keys, no surprise charges.

How It Works

Six phases. Two crucial moments.

Each phase improves the quality of the final output. Your expertise shapes the result at the moments that matter most.

01

Specification Chat

The AI Task Manager interviews you to understand requirements. It asks clarifying questions, reveals hidden information, and suggests improvements based on your existing codebase. Your prompt improves through structured dialogue.

Dalia's spec generation chat where the LLM asks clarifying questions about implementation details
Crucial Moment 02

Plan Review

Before any code is written, you review a detailed plan with inline comments — categorized as bug, security, style, question, task, or nit. A separate validation document locks acceptance criteria before execution begins.dashboard

Dalia's plan preview with architectural diagram and inline review comments
03

Task Breakdown

The approved plan is broken into atomic tasks, each categorized by the skill necessary to build it. This semantic categorization selects the most appropriate agents for each task and orchestrates sequential and parallel execution.

Dalia's execution logs showing blueprint execution with tool calls and thinking
04

Code Generation

Tasks execute according to the orchestrated plan, with Drupal-specific RAG knowledge injected at runtime. You can send messages to the agent during execution to provide guidance or course-correct.

Dalia's logs showing RAG-injected Drupal knowledge documents during spec generation
05

Self-Validation

Dalia validates features against your curated requirements using Playwright scripts for browser-based testing and Drush commands for CLI verification. When validation fails, it enters a fix-and-retry loop.

Dalia's self-validation modal with CLI and browser validation options
Crucial Moment 06

Code Review

All code edits are presented in a diff view — similar to a GitHub PR, but private. You review locally before sending anything public. By reviewing the code, you intellectually own it. You understand what goes out, and you endorse it.

Dalia's code review diff view with file tree, inline commenting, and SDC schema

Key Components

RAG for Drupal Knowledge

Injects semantically relevant Drupal documentation at runtime, ensuring generated code follows current best practices rather than outdated or incorrect patterns from training data.

AI Task Manager

Runs the specification interview, generates plans, breaks work into atomic skill-categorized tasks, and orchestrates execution order.

GitHub ↗

Self-Review

Powers the inline commenting UIs for plan review, validation document review, and the final code review — all with categorized feedback labels.

GitHub ↗

Setup

One command to setup.
Use your existing Claude Code subscription.

Distributed as a DDEV container. No Anthropic API key, no surprise billing. It runs on the Claude subscription you already have. 

$ ddev add-on get /path/to/ddev-dalia.tar.gz

Dalia is currently in closed beta. Getting Started Guide

dalia.drupal-contrib.ddev.site/settings
Dalia's settings page showing RAG status, model selection, and environment checks

Why Dalia

RAG-Augmented Accuracy
Every generation is anchored to current, correct Drupal documentation — not the garbled training data LLMs have by default.
Human at the Center
Plan review and code review put your expertise where it matters most. You shape the output at two crucial moments.
Structured Pipeline
Specification, planning, task breakdown, execution, validation, review. Each phase improves quality.
Code You Own
Review everything locally before it goes public. No noise for your peers, no code you don't endorse.