AI-Kit + execution control

Use AI in WordPress Without Sending Every Draft Through an AI SaaS

An AI plugin is also a data-path decision. AI-Kit can run supported tasks on-device in the browser and use a customer-configured backend only when the workflow needs it.

Short answer Use local browser AI for supported editor and media tasks, then choose an optional customer-configured backend for fallback, frontend AI, DocSearch, chatbot, or stronger processing. A vendor SaaS does not have to be the mandatory runtime for every request.

The hidden architecture question

“AI inside WordPress” does not tell you where the content goes

Two plugins can expose a similar button in Gutenberg while using very different execution paths, provider accounts, and operational boundaries.

Data path

A familiar editor button can hide a remote vendor runtime

If every rewrite or metadata request is routed through a vendor application backend, the plugin provider becomes part of the content path even when the UI lives entirely inside WordPress.

Availability

Local AI support varies by browser and device

On-device processing can keep supported tasks in the browser, but it should not be described as universally available. Browser support, OS, hardware, and feature availability matter.

Ownership

Backend AI introduces a separate infrastructure decision

When remote processing is useful, teams should know which endpoint receives the request, who controls that endpoint, and which account owns the model runtime and logs.

Buyer question Ask where each AI feature executes, not only which model or button the plugin provides.

Execution paths

Separate on-device AI from optional backend processing

AI-Kit supports different modes. The exact data path depends on the feature and configuration, so privacy claims should be tied to the mode actually in use.

WordPress editor / Media Library / frontend
      |
      +--> supported on-device AI
      |      processing stays in the browser
      |
      +--> backend-only or backend-fallback
             |
             v
      configured HTTPS API endpoint
             |
             v
      customer-configured / customer-owned AWS backend
             |
             v
      model / retrieval / business logic

Accuracy boundary Do not describe AI-Kit as universally local. Free local-only features depend on supported Chrome on-device AI. Pro backend modes send the feature input to the API endpoint configured for that site.

Evaluation path

Choose the execution mode feature by feature

Start from the workflow and data boundary instead of selecting one privacy label for the whole plugin.

Identify the AI action — List the actual tasks: rewrite, proofread, translate, metadata generation, frontend AI, chatbot, or DocSearch.
Test local support — For supported editor and media tasks, test whether the target Chrome environment provides the required on-device AI capability.
Configure backend use deliberately — If local AI is unavailable or insufficient, choose backend-fallback or backend-only behavior and configure the API endpoint that should receive the request.
Document the resulting data path — Record which content leaves the browser, which endpoint receives it, who owns the backend, and which authentication or rate-limit controls apply.

When this architecture fits

Good fit

Use AI-Kit when execution ownership matters

You want supported editor or Media Library AI tasks to run on-device where the browser supports them.
You need optional backend processing but want to configure the endpoint and commonly deploy it into your own AWS account.
Your static WordPress frontend still needs AI features without relying on WordPress PHP as a proxy.

A SaaS AI plugin may be simpler

Use a vendor-hosted AI service when

You value the simplest setup more than controlling the execution path.
Routing content through the vendor service is acceptable for the project.
You do not need local processing, a customer-controlled backend, or static-friendly direct API calls.

Buyer questions

WordPress AI data-path FAQ

How can I use AI in WordPress without sending all content through an external AI SaaS?

Use features that can execute locally for the tasks and environments that support on-device AI. AI-Kit Free uses Chrome on-device capabilities where available, while Pro can optionally use a configured backend endpoint for features that need remote processing.

Does AI-Kit run everything locally?

No. Local-only mode applies to supported features when Chrome on-device AI is available. Pro can use backend-only or backend-fallback modes, and those requests are sent to the API endpoint configured for the site.

Can WordPress AI use my own AWS account?

Yes. The typical AI-Kit Pro backend pattern is an AWS API deployed into the customer’s own account, for example through the WP Suite AI-Kit deployment template. The WordPress site then calls the configured API URL.

Will backend AI still work after static export?

Yes, if the exported frontend can reach the configured endpoint. AI-Kit sends backend requests from the browser rather than requiring WordPress PHP to proxy them.

WP Suite AI-Kit

Choose where WordPress AI processing runs

Start with on-device AI where it fits, then add a customer-configured backend only for the capabilities that need it.

Explore AI-Kit

Compare AI execution models