This project is not covered by Drupal’s security advisory policy.

A Drupal module that integrates mittwald AI Hosting with the Drupal AI module, enabling AI-powered features on your Drupal site using mittwald's infrastructure.

Features

This module provides a mittwald provider implementation for the Drupal AI module, allowing you to leverage mittwald's AI hosting platform for various AI operations including chat completions, embeddings, reranking, speech-to-text, text-to-speech, and more.

Credit

This module was forked from the ai_provider_openai module and adapted for mittwald's AI hosting platform, which uses an OpenAI-compatible API.

Post-Installation

  1. Obtain an API Key: Follow the mittwald AI hosting access guide to get your API credentials.
  2. Create a Key entity:
    • Navigate to /admin/config/system/keys/add
    • Create a new key for storing your mittwald API key
    • Choose an appropriate key type (e.g., "Authentication")
  3. Configure the provider:
    • Navigate to /admin/config/ai/providers/mittwald
    • Select your API key from the dropdown
    • Save the configuration

The module will automatically validate your API key and check for rate limits during configuration.

Supported operations and models

Chat completions

✅ Fully supported for conversational AI, content generation, and chat-based interactions. See the list of available models for an up-to-date list of all available models.

Capabilities:

  • Standard text chat
  • Image vision (Ministral, Qwen3.5-122B-A10B-FP8, Qwen3.6, Qwen3.8 and GLM-OCR models; Qwen3.5-0.8B is text-only and is not offered for image input)
  • Reasoning mode with configurable effort (GPT-OSS, Qwen3.5, Qwen3.6 and Qwen3.8 models)
  • Document OCR for PDF, DOCX, PPTX, XLSX, HTML, SVG and images (GLM-OCR)
  • JSON output formatting
  • Tool/function calling
  • Streaming responses

Default model: Ministral-3-14B-Instruct-2512

Embeddings

✅ Fully supported for semantic search, similarity matching, and vector operations.

Default model: Qwen3-Embedding-8B (4096 dimensions)

Reranking

✅ Fully supported for improving RAG retrieval precision by reranking search results.

Default model: Qwen3-VL-Reranker-2B (text and image scoring)

Speech to Text

✅ Fully supported for speech recognition.

Default model: whisper-large-v3-turbo

Text to Speech

✅ Fully supported for generating spoken audio from text.

Nine voices are available: aiden, dylan, eric, ono_anna, ryan, serena, sohee, uncle_fu and vivian. Voice cloning is not available. Audio can be returned as mp3, opus, flac, wav or pcm, and the speaking rate and spoken language are configurable.

Default model: Qwen3-TTS-12Hz-1.7B-CustomVoice

Text to Image

⛔ Currently not supported by mittwald AI hosting.

Moderation

⛔ Currently not supported by mittwald AI hosting.

Resources

Supporting organizations: 
mittwald
Funding development

Project information

Releases

1.2.0 released 2 September 2026
Works with Drupal: ^10.3 || ^11

Adds reranking and text-to-speech support along with the Qwen3.8-27B-NVFP4 chat model, and syncs the model lineup with mittwald's current offering.

Install:
1.1.0 released 5 May 2026
Works with Drupal: ^10.3 || ^11

Added support for Qwen3.5-122B-A10B-FP8 and Qwen3.6-35B-A3B-FP8 models with reasoning and vision capabilities.

Install:
1.0.1 released 16 January 2026
Works with Drupal: ^10.3 || ^11

Adjust to model deprecation, use new default models

Install:
0.1.1 released 3 December 2025
Works with Drupal: ^10.3 || ^11

Initial release

Install:

Using Composer to manage Drupal site dependencies