WordPress 7.1's AI Client Gets Streaming and Semantic Search

Alongside today's WordPress 7.1 Beta 1, the WordPress AI team published what's coming to the AI Client in this release cycle: streaming responses and embeddings-based semantic search, both announced June 19, 2026. Neither changes what a stock WordPress site looks like — this is foundation-laying for plugin developers, the same way 7.0's Abilities API was foundation rather than a user-facing feature on its own.

If you're tracking WordPress core's AI direction because you build custom blocks or plugins for clients — or you're evaluating what a plugin claiming "AI-powered search" is actually built on — this is the layer underneath.

What's shipping in the AI Client for 7.1

Streaming for generation responses. Right now, an AI Client request in WordPress waits for the full model response before returning anything. Streaming changes that to a token-by-token flow, the same pattern you see in ChatGPT-style interfaces where text appears progressively instead of all at once. The AI team's own framing is precise here: streaming will be available within WP 7.1, but not as something core itself surfaces in a built-in interface — it's infrastructure for developers and hosts to build on, not a new admin feature you'll see by default.

Embeddings for semantic/vector search. Embeddings are numerical representations of text that let a system compare meaning rather than just matching literal keywords. This work leans on vector storage capabilities that landed recently in MySQL and MariaDB, which means the semantic-search capability doesn't require bolting on an external vector database — it can live in the same database WordPress already uses.

For agencies, the practical near-term implication is narrow but real: if a plugin you rely on for client search (a "smart search" or "AI search" add-on) starts advertising semantic search built on the core AI Client, this is what it's built against — and it means the plugin doesn't need its own separate vector-database subscription to do it.

The Knowledge proposal (formerly "Guidelines")

Running alongside the AI Client work is a proposal — previously called Guidelines, now referred to as Knowledge — for a wp_knowledge custom post type. The idea: give sites a dedicated, structured place to store information that AI features can read and write, separate from regular post content. Three knowledge types are outlined so far: guideline (editorial rules — house style, tone, things to avoid), memory (persistent context an AI feature should retain across sessions), and note (ad hoc, less structured entries). A filter is planned to let developers register additional custom types beyond those three.

The proposal includes dedicated REST routes (/wp/v2/knowledge) and its own capability structure (*_knowledge_item permissions), separate from standard post-editing capabilities — meaning access to a site's AI-facing knowledge base could be governed independently of who can edit regular content. This is still a proposal rather than a finalized, shipped feature as of this writing, so treat the exact shape as subject to change before it lands.

If it ships close to this description, the agency-relevant use case is fairly obvious: instead of every AI plugin inventing its own way to store "here's our brand voice guidelines" or "here's what this client's business actually does" as context, there'd be one core-native place to put it, with its own permission model, that any compatible plugin could read from.

How this connects to today's Beta 1 release

This is a companion piece to our WordPress 7.1 Beta 1 agency guide — the AI Client work isn't part of what that article's testing checklist covers (page-builder compatibility, the experimental React 19 flag, new blocks), because it's a lower-level API layer most agencies won't interact with directly unless they're building or evaluating AI-powered plugins. If that's not you, there's nothing to test here before August 19 — it's invisible until a plugin built on top of it reaches your stack.

If it is you — if you maintain a custom plugin or work closely with a client's development team on AI-powered features — this is the moment to read the AI team's proposal threads on make.wordpress.org/ai directly rather than relying on secondhand summaries, since both the streaming API surface and the Knowledge post type are still being actively discussed ahead of the August 19 final release.

The pattern across 7.0 and 7.1

Zoom out and the last two major releases tell a consistent story: 7.0 shipped the plumbing (AI Client, Abilities API, Connectors Hub for provider authentication — see our WordPress 7.0 AI features guide), and 7.1 is extending that plumbing (streaming, semantic search, a proposed structured-knowledge store) rather than shipping a new headline AI feature of its own. That's a deliberate sequencing choice by the release squad — infrastructure first, so that when user-facing AI features do arrive, plugins aren't each reinventing incompatible versions of the same capability.

For an agency, the useful takeaway isn't "go enable AI streaming today" — there's nothing to enable yet. It's knowing this is happening under the hood, so that when a plugin update six months from now claims "now with real-time AI streaming" or "AI-powered semantic search," you know exactly what it's built on and can evaluate it accordingly instead of taking the marketing copy at face value.