Felix Arntz @flixos90 (co-Team RepTeam RepA Team Rep is a person who represents the Make WordPress team to the rest of the project, make sure issues are raised and addressed as needed, and coordinates cross-team efforts.) – Vercel
James LePage @isotropic (co-Team Rep) – Automattic
The WordPress AI Team is excited to invite the community to test a growing collection of community-built AI connector plugins. These plugins extend WordPress by connecting it to additional AI services through the PHP AI Client — the provider-agnosticProvider-AgnosticSoftware design that works with multiple service providers without being tied to one. Recommended for WordPress AI integrations. SDK that is part of the upcoming WordPress 7.0 release (read more about it here).
WordPress 7.0 will ship with the PHPPHPPHP (recursive acronym for PHP: Hypertext Preprocessor) is a widely-used open source general-purpose scripting language that is especially suited for web development and can be embedded into HTML. https://www.php.net/manual/en/preface.php AI Client baked into coreCoreCore is the set of software required to run WordPress. The Core Development Team builds WordPress., and connectors for Anthropic, Google, and OpenAI appear as default options in the connector screen — ready to install with a single click. But the real power of the PHP AI Client’s open architecture is that anyone can build a connector for any AI service. The community has already stepped up — and these plugins are ready for testing.
If you’ve been curious about WordPress’s evolving AI capabilities, this is a great opportunity to try them out firsthand and help shape the experience.
What are AI connector plugins?
AI connector plugins act as bridges between WordPress and external AI services. They implement the PHP AI Client’s providerProviderAn AI service offering models for generation, embeddings, or other capabilities (e.g., Anthropic, Google, OpenAI). interface, allowing any WordPress pluginPluginA plugin is a piece of software containing a group of functions that can be added to a WordPress website. They can extend functionality or add new features to your WordPress websites. WordPress plugins are written in the PHP programming language and integrate seamlessly with WordPress. These can be free in the WordPress.org Plugin Directory https://wordpress.org/plugins/ or can be cost-based plugin from a third-party. or theme to use AI capabilities — text generation, image generation, function calling, and more — through a unified, provider-agnostic APIAPIAn API or Application Programming Interface is a software intermediary that allows programs to interact with each other and share data in limited, clearly defined ways..
Connectors for Anthropic, Google, and OpenAI are featured prominently in the WordPress 7.0 connector screen and can be installed directly from there. The community connector plugins featured in this article expand the ecosystem further, giving users and developers more choice in which AI services power their WordPress sites.
Each connector plugin handles the specifics of communicating with its respective AI service, so developers and users can switch between providers without changing their code or workflow.
Screenshot of the connector screen having a community plugin connected.
Community connector plugins to test
AI Provider for Grok (xAI)
Grok is xAI’s conversational AI, known for its real-time knowledge and distinctive personality.
Capabilities: Text generation, image generation, function calling, vision input, structured output (JSONJSONJSON, or JavaScript Object Notation, is a minimal, readable format for structuring data. It is used primarily to transmit data between a server and web application, as an alternative to XML. schema)
Models: Dynamically discovered from the Mistral API — always up to date with the latest available models, including mistral-medium-2505 for image generation
Ollama lets you run open-source AI models locally on your own hardware — no API key, no cloud, no cost. Perfect for privacy-conscious sites and local development.
Capabilities: Text generation using locally hosted models, automatic model detection, support for both local and Ollama Cloud deployments
Models: Any model available through Ollama’s registry (Llama 3.2, Mistral, Phi, Gemma, and more)
API key: No API key required for local mode — just have Ollama running on your machine. Cloud mode requires an Ollama Cloud API key.
Install: Install from WordPress.org or download from GitHub. For local mode, install Ollama and pull a model (e.g., ollama pull llama3.2).
AI Provider for OpenRouter
OpenRouter is a unified API gateway providing access to 400+ AI models from dozens of providers — including Anthropic, OpenAI, MetaMetaMeta is a term that refers to the inside workings of a group. For us, this is the team that works on internal WordPress sites like WordCamp Central and Make WordPress., Google, Mistral, and many more — through a single API key.
Capabilities: Text generation with access to 400+ models, searchable model selection with provider filtering, free-tier model toggle, context length and pricing display
Models: Full OpenRouter catalog — Claude, GPT-4, Llama 3, Gemini, Mistral, and hundreds more
API key: Create an account at openrouter.ai and retrieve your key from Settings > Keys. Configure under Settings > OpenRouter AI in the WordPress admin.
We encourage testers to use WordPress 7.0 which includes the PHP AI Client as part of core. This is the closest you can get to the final WordPress 7.0 experience and helps us catch integration issues before the stable release.
You will also need the AI Experiments plugin, which provides the user-facing features that leverage the AI capabilities these connectors provide.
Prerequisites
WordPress 7.0 (Beta Tester Plugin recommended ahead of the final 7.0 release)
PHP 7.4 or higher (PHP 8.0+ for OpenRouter and Ollama)
An API key for at least one of the AI providers listed above (not required for Ollama in local mode)
Setup steps
Install WordPress 7.0.
Install and activate the AI Experiments plugin.
Install and activate one or more community connector plugins from the list above.
Navigate to the connector settings in your WordPress dashboard to verify the provider appears and is available.
What to test
1. Provider setup and connectivity
Does the connector plugin activate without errors?
Does the provider appear in the connector selection screen alongside the default providers (Anthropic, Google, OpenAI)?
Is the API key configuration straightforward?
Does the availability check succeed when the API key is valid?
2. Text generation
Enable AI experiments in the AI settings.
Open the post editor and and use text generation fetures. Do AI-powered suggestions appear?
Try switching between community connectors and the default providers — does the experience remain consistent?
Screenshot of showing AI Editor Experiments being enabled.
3. Image generation
For connectors that support image generation (currently Mistral):
Can you generate images using “Generate featured imageFeatured imageA featured image is the main image used on your blog archive page and is pulled when the post or page is shared on social media. The image can be used to display in widget areas on your site or in a summary list of posts.”- feature in the editor?
How does the generation speed compare to the default providers?
4. Error handling
What happens when an invalid API key is provided?
For Ollama: what happens when the local Ollama service is not running?
Are error messages clear and actionable?
Known issues
Image generation is currently only available on select providers and models.
Questions to consider
When testing, we’d appreciate your thoughts on:
Setup experience: Was the installation and configuration process clear? What could be improved?
Provider switching: Did switching between community connectors and default providers feel seamless?
Response quality: How did the AI responses compare across different community providers and the defaults?
Performance: Did you notice latency or performance differences between providers?
Local AI (Ollama): If you tested local models, how did the experience compare to cloud providers? Was setup straightforward?
Error messages: When something went wrong, were the error messages helpful?
Documentation: Was there enough information to get started? What was missing?
Testing is open from 25 March2026 through 2 April 20268 April 2026.
Want to build your own connector?
The PHP AI Client makes it straightforward to build a connector for any AI service. If your preferred AI provider isn’t listed above, consider building one! The PHP AI Client documentation and the community connector plugins listed here are great starting points. At a high level, connector plugins need to:
Extend AbstractApiProvider from the PHP AI Client.
Implement model metadata discovery and model classes for supported capabilities.
Register the provider with AiClient::defaultRegistry() on the WordPress init hook.
We welcome new community connectors and would love to include them in future testing rounds.
Thank you for helping test and improve WordPress’s AI capabilities. The more providers the community builds and tests, the more choice WordPress users will have when AI features land in WordPress 7.0.
Over the past few weeks, we’ve been experimenting with a handful of AI-powered concepts in the AI Experiments pluginAI Experiments PluginWordPress's AI laboratory bringing all building blocks together. Serves as both a user tool and developer reference implementation. First release (v0.1.0) includes Title Generation experiment.. These concepts are intentionally exploratory. Some are rough, some may not stick, and all of them need feedback from folks actually using WordPress in the real world.
This post is a Call for Testing, with less interest in “is this perfect?” and more interested in:
Does this feel useful or confusing?
Would you expect this to exist where it does?
What would you change about the UXUXUX is an acronym for User Experience - the way the user uses the UI. Think ‘what they are doing’ and less about how they do it. or flow?
Does it behave as expected in your testing?
Below you’ll find a set of user-facing experiments, developer-focused tools, and one hybrid feature that cuts across both. Each section includes a short description, steps to test and a screencast example, a WordPress Playground link to utilize for testing, and a link back to the PR for feedback.
Please test what you’re curious about. You don’t need to try everything.
User-Facing Experiments
These experiments focus on AI features that surface directly in the WordPress admin or content workflows.
Type Ahead Suggestions
This experiment explores AI-powered type-ahead suggestions while writing. The goal is to understand whether inline suggestions help with flow and clarity, or whether they feel distracting or overly prescriptive.
We’re especially interested in feedback on:
Does the timing of suggestions feel right?
Do suggestions feel helpful or noisy?
Where would you expect controls or settings for this to live?
Navigate to Settings > AI Credentials and add an APIAPIAn API or Application Programming Interface is a software intermediary that allows programs to interact with each other and share data in limited, clearly defined ways. key to one of the AI clients listed, then click Save Changes
Navigate to Settings > AI Experiments, check Enable Experiments, click Save Changes, check Type-ahead Text, and click Save Changes
Navigate to Posts > All Posts and either edit the Hello world! post or click Add Post to open the post editor
Start typing content and observe type-ahead suggestions
Press Tab to accept or Escape to dismiss suggestions
This experiment explores using AI to assist with comment moderation. Rather than fully automating decisions, the focus is on providing helpful context or signals to moderators.
We’re especially interested in feedback on:
How do you feel about the AI Reply and Analyze with AI functionality?
This experiment looks at consuming and working with Markdown-based feeds using AI. The goal is to explore new content ingestion workflows and how AI might help interpret or transform structured text.
Navigate to Settings > AI Experiments, check Enable Experiments, click Save Changes, check Markdown Feeds, and click Save Changes
Navigate to Posts > All Posts and view the Hello world! post
Change the end of the URLURLA specific web address of a website or web page on the Internet, such as a website’s URL www.wordpress.org to /hello-world.md and confirm text/markdown output
Note additional testing options:
Visit /?feed=markdown and confirm text/markdown output
Request a post with Accept: text/markdown and confirm Markdown output
Optional: use the new section filters to inject custom fields into output: npm run test:php -- --filter Markdown_Feeds
This experiment expands the set of available AI providers, making it easier to test and compare different models and services.
We’re especially interested in feedback on:
Is the providerProviderAn AI service offering models for generation, embeddings, or other capabilities (e.g., Anthropic, Google, OpenAI). selection UX understandable?
Do the abstractions make sense from a user point of view?
What information would help you choose between providers?
Navigate to Settings > AI Experiments, check Enable Experiments, click Save Changes, check MCP, and click Save Changes
Navigate to the top level MCP menu item and view the MCP screen to ensure that Enable MCP and Enable Server are toggled on and that the server shows as 🟢 Running
Toggle the Expose via MCP item for various abilities to test
Use the Test connection feature in the Connection test section to verify endpoint
Copy a Client configuration and test in Claude Desktop or Cursor
This experiment adds logging for AI requests (provider, model, tokens, duration, cost estimate) that can be viewed from the Admin Dashboard as well as configurable retention period and automatic cleanup to help with debugging, cost tracking, and usage analysis.
We’re especially interested in feedback on:
Is the level of detail appropriate?
Who do you think this is for: site owners, developers, or both?
How should this data be surfaced or stored?
How would you want to filterFilterFilters are one of the two types of Hooks https://codex.wordpress.org/Plugin_API/Hooks. They provide a way for functions to modify data of other functions. They are the counterpart to Actions. Unlike Actions, filters are meant to work in an isolated manner, and should never have side effects such as affecting global variables and output., hide, or export this data?
Navigate to Settings > AI Experiments, check Enable Experiments, click Save Changes, check AI Request Logging, and click Save Changes
Enable other Experiments (and necessary, related AI Client Credentials via Settings > AI Credentials) and perform actions that trigger AI requests (e.g. Title Generation, ExcerptExcerptAn excerpt is the description of the blog post or page that will by default show on the blog archive page, in search results (SERPs), and on social media. With an SEO plugin, the excerpt may also be in that plugin’s metabox. Generation)
Navigate to Settings > AI Request Logs and view the AI Request Logs screen showing requests, tokens, and other data
Review logged requests and metadata
Test filtering by provider, status, and date range
Verify “Purge logs” clears all entries
Disable the experiment and confirm no new logs appear
Feedback directly on the PRs is ideal, but broad impressions are also valuable (e.g. as comments on this post, messages in #core-ai). You don’t need to review code to participate, high-level product, UX, and architectural feedback is just as valuable here. If you’re short on time, even answers to one or two of these questions help:
Would you use this if it shipped tomorrow?
What’s the first thing you’d change?
Where does this feel surprising, in a good or bad way?
These experiments are about learning in public. Your perspective can help us decide what to refine, what to pause, and what to drop entirely. Thanks for helping shape what AI in WordPress could become.
Props to @dkotter for helping with content for this post and @justlevine for pre-publish review.
You must be logged in to post a comment.