AI Provider for mittwald

This plugin integrates mittwald AI Hosting with WordPress AI features, enabling AI-powered features on your WordPress site using mittwald’s infrastructure.

This plugin requires WordPress 7.0 or newer.

Supported Operations & Models

Chat Completions

Fully supported for conversational AI, content generation, and chat-based interactions.

Available Models:
– GPT-OSS: gpt-oss-120b
– Qwen: Qwen3.5-122B-A10B-FP8, Qwen3.6-35B-A3B-FP8 and Qwen3.8-27B-NVFP4 (support vision/image input), Qwen3.5-0.8B
– Ministral: Ministral-3-14B-Instruct-2512 (supports vision/image input)
– GLM OCR: GLM-OCR, for extracting text from an image; it takes an image with a text prompt and supports none of the capabilities below

Capabilities:
– Standard text chat
– Image vision (Ministral and the larger Qwen models)
– JSON output formatting
– Tool/function calling

Embeddings

Available Models:
– Qwen Embedding: Qwen3-Embedding-8B

Capabilities:
– Text to vector, one embedding per input
– 4096-dimensional vectors by default, up to 32,768 tokens of context per input
– A narrower vector on request, via the dimensions option
– Several inputs per request; embeddings are returned in input order

The dimensions option accepts 256, 512, 768, 1024, 1536, 2048, 3072 and 4096;
256 is the documented floor, as retrieval quality degrades below it. Any other
width is refused before a request is sent: automatic model selection finds no
suitable model, and an explicitly chosen model rejects the width and lists the
accepted ones. Pass the width through the dimensions option itself; as a
custom option it is refused. The endpoint L2-normalises a
projected vector, so you can use it directly for dot-product similarity without
normalising it yourself.

Vector storage is out of scope for this plugin — it returns the vectors and
never touches a database.

Embeddings need version 1.4.0 or newer of the WordPress AI client. WordPress
7.0 and 7.1 bundle an older version, but the AI Experiments
plugin adds embedding support to it from version 1.3.0 on. Without either, the
embedding model is not offered, and every other model keeps working.

Text to Speech

Available Models:
– Qwen TTS: Qwen3-TTS-12Hz-1.7B-CustomVoice

Capabilities:
– Voices: aiden, dylan, eric, ono_anna, ryan, serena, sohee, uncle_fu, vivian (defaults to aiden)
– Output formats: audio/mpeg (default), audio/wav, audio/flac, audio/ogg
– Parameters such as speed, language and instructions