AI Provider for QwenCloud

QwenCloud as a provider for the WordPress AI Client: text, vision and text-to-image generation with Qwen, DeepSeek, Kimi and GLM models through the OpenAI-compatible compatible-mode API.

What it does

• Fetches the model list live from GET /compatible-mode/v1/models, so newly released models show up on their own.
• Text generation and chat history, including tool calling, structured output (JSON Schema and JSON Object) and stop sequences.
• Vision input for the Qwen3.5+ commercial models, the Qwen-VL / Qwen2.5-VL / QVQ families, the Omni models and deepseek-v4.1-flash, so features like Alt Text Generation work.
• Text-to-image generation with the qwen-image-3.0 and qwen-image-3.0-pro models, returned as a URL.
• Reasoning output surfaces as thought parts instead of being mixed into the answer.
• Custom QwenCloud parameters — enable_thinking, thinking_budget, top_k, enable_search and friends — pass through customOptions.

Requirements

• WordPress 7.0 or newer, with the PHP AI Client SDK (bundled in WordPress 7.0, or provided by the AI plugin)
• PHP 7.4 or newer
• A QwenCloud account with API access

Install

Download the zip from Releases and upload it through Plugins → Add New → Upload Plugin, or copy the plugin folder to wp-content/plugins/ai-provider-for-qwencloud/. Activate it, then open Settings → Connectors, open the QwenCloud card and paste your API key.

Configuration

The API key is managed by WordPress: either save it on Settings → Connectors or define QWENCLOUD_API_KEY in the environment. This plugin does not read the connector option directly — whether a credential exists is asked of the AI Client registry.

QwenCloud's own examples use DASHSCOPE_API_KEY, because the API is DashScope-compatible. This plugin follows the WordPress AI Client convention and derives the variable name from its provider ID, so the key it looks for is QWENCLOUD_API_KEY. Define that name, or use the Connectors screen.

Optional settings are environment variables or PHP constants:

There is no database-backed settings page: the base URL and the rest are deployment-level overrides.

Because the plugin has no settings of its own, changing an environment variable does not invalidate the AI Client's cached model list. If a new value does not take effect, run wp cache flush.

To change which model the AI plugin picks for its features, use the standard filter — this plugin already puts its preferred models first:

add_filter( 'wpai_preferred_text_models', function ( $models ) {
    array_unshift( $models, array( 'qwencloud', 'qwen3.8-max' ) );
    return $models;
} );

Model capabilities

The /models endpoint reports no capability data, so the live list is classified through a local catalog (QwenCloudModelCatalog). The rules are:

• Text — the Qwen3.x and Qwen3- families, `qwen2.5-, the legacyqwen-*` families, and the documented DeepSeek, Kimi and GLM models.
• Text + image — the Qwen3.5–3.8 max/plus/flash models, the Qwen-VL / Qwen2.5-VL / QVQ families, the Omni models, and deepseek-v4.1-flash.
• Image generation — only qwen-image-3.0 and qwen-image-3.0-pro.
• Everything else — video, audio, embeddings, reranking, world models, the decision model and image translation — stays in the list with no capability, so it is visible but never selected.

Not implemented

• Video, audio, embeddings, reranking and the World Model APIs. The AI Client has no unified interface for them, so those models are listed without a capability rather than shipping code the SDK never calls.
• Image editing. Only text-to-image (POST /images/generations) is implemented; the image editing endpoints take a different request shape and are not part of the AI Client's reference image flow.
• Streaming and the Responses API. Only synchronous Chat Completions is used.

Data and privacy

Prompts you send to the AI, plus whatever the calling plugin or theme adds to them (system instructions, conversation history, tool definitions, a JSON schema, attached files), are sent to QwenCloud. Your API key is stored on your own site and is only ever sent to the QwenCloud host configured above. Nothing is sent until your site actually asks the AI Client for a generation, and the plugin never logs prompts, tokens, cookies or Authorization headers.

QwenCloud documentation

• First API call
• OpenAI-compatible chat API
• Choose models
• Structured output
• Thinking
• OpenAI-compatible image generation

Development

php scripts/selfcheck.php                                              # logic checks, no WordPress needed
php scripts/selfcheck.php --sdk=/path/to/php-ai-client                 # plus live-SDK checks

The self-check asserts model classification, sort order, the declared capabilities and options, the request shapes for chat and images, and that the declared metadata satisfies the requirements the AI plugin's features actually send. It needs neither WordPress nor an API key.

Release: push a tag matching the plugin version (git tag <version> && git push origin <version>). The release workflow verifies the tag against Version:/Stable tag:, verifies a matching changelog entry exists, builds the plugin zip and publishes it.

License

GPL-2.0-or-later. See LICENSE.

The assets/images/qwencloud.svg mark is a generic icon used only to identify the service this plugin connects to.

The full WordPress plugin readme, including the changelog, lives in readme.txt.