David Stone

This plugin extends the WordPress AI Client to support any AI service or server that uses the standard chat completions API format (<code>/v1/chat/completions</code> and <code>/v1/models</code> endpoints).
Supported services include:
<ul>
<li>Ollama – Run open-source models (Llama, Mistral, Gemma, etc.) locally on your own hardware.</li>
<li>LM Studio – Desktop application for local LLM inference with a one-click server.</li>
<li>OpenRouter – Unified API providing access to 100+ models from multiple providers.</li>
<li>vLLM – High-throughput inference server for production deployments.</li>
<li>LocalAI – Drop-in replacement for running models locally.</li>
<li>text-generation-webui – Popular web UI with API server mode.</li>
<li>Any compatible endpoint – Works with any server implementing the standard format.</li>
</ul>
Requirements:
<ul>
<li>WordPress 7.0+ – The AI Client SDK is included in core. This plugin works on its own without any additional dependencies.</li>
<li>WordPress 6.9 – Also supported when the Gutenberg plugin (23.0+) is active, which provides the AI Client SDK.</li>
</ul>
Why it matters:
Other AI-powered plugins that use the WordPress AI Client (such as AI Experiments) can automatically discover and use any model you connect through this plugin. Configure your endpoint once and every AI feature on your site can use it.
How it works:
<ol>
<li>Install and activate the plugin.</li>
<li>Go to Settings > Connectors and configure the connector with your endpoint URL (e.g. <code>http://localhost:11434/v1</code> for Ollama).</li>
<li>Optionally provide an API key for services that require authentication.</li>
<li>The plugin registers a provider with the WordPress AI Client and dynamically discovers all available models from your endpoint.</li>
</ol>
The plugin also handles practical concerns like extended HTTP timeouts for slow local inference and non-standard port support.

Ultimate AI Connector for Compatible Endpoints

Connects the WordPress AI Client to Ollama, LM Studio, or any AI endpoint that uses the standard chat completions API format.

Andrei Lupu

AI Provider for Osaurus wires the <a href="https://osaurus.ai" rel="nofollow ugc">Osaurus</a> local LLM runtime into the WordPress AI Client that ships with WordPress 7.0.
After activation, any plugin that calls <code>wp_ai_client_prompt()</code> (Gutenberg AI features, official provider plugins, third-party code) can route prompts to Osaurus running on the same machine. Prompts never leave the local network.
<h4>Why use it</h4>
<ul>
<li>Privacy first — prompts stay on your machine, no third-party API receives the data.</li>
<li>Zero cost — no API keys, no metered requests.</li>
<li>Fast — MLX-backed inference on Apple Silicon.</li>
<li>OpenAI-compatible wire format — streaming, tool calls, JSON output, and chat history work out of the box.</li>
</ul>
<h4>Capabilities</h4>
<ul>
<li>Text generation</li>
<li>Chat history</li>
<li>Function / tool calls</li>
<li>Structured JSON output</li>
<li>Streaming</li>
</ul>
Image generation, embeddings, and text-to-speech are not supported because Osaurus does not currently expose those endpoints. Pair this plugin with an image-capable provider such as the official AI Provider for OpenAI when you need images.
<h4>How it works</h4>
The plugin registers a provider with <code>AiClient::defaultRegistry()</code> and points it at the Osaurus HTTP server. Because Osaurus speaks the OpenAI Chat Completions wire format, the implementation extends the SDK’s OpenAI-compatible base classes — so capabilities like streaming and tool calling flow through automatically.
<h4>Configuration</h4>
The plugin resolves the Osaurus base URL in this order (first match wins):
<ol>
<li><code>OSAURUS_BASE_URL</code> PHP constant defined in <code>wp-config.php</code>.</li>
<li><code>osaurus_ai_connector_base_url</code> option, settable from Settings → Connectors in wp-admin.</li>
<li>Default: <code>http://127.0.0.1:1337/v1</code> (works out of the box for bare-metal WordPress on the same Mac as Osaurus; Docker setups must override).</li>
</ol>
<h4>Pick the right URL for your setup</h4>
WordPress can run in many places — the right base URL depends on how it reaches your Mac, where Osaurus is listening.
<ul>
<li>Bare-metal WordPress on the same Mac (MAMP, Laravel Valet, Local, WordPress Studio, native PHP): 
 http://127.0.0.1:1337/v1 (this is the default — no configuration needed)</li>
<li>Docker-based WordPress on the same Mac (DDEV, Lando, Docker Desktop): 
 http://host.docker.internal:1337/v1</li>
<li>WordPress on a different machine on your LAN, Osaurus on your Mac: 
 http://:1337/v1 — make sure Osaurus is bound to a non-loopback interface and your firewall allows port 1337.</li>
<li>Remote Osaurus you operate (e.g. a Mac mini on Tailscale, a colocated server): 
 http://:1337/v1 or <code>https://...</code> if you front it with TLS.</li>
</ul>
The fastest way to confirm a URL works from the WordPress host: run <code>curl <base-url>/models</code> from the same shell environment WordPress runs in. A JSON list back means the connector will work.
Osaurus project home: <a href="https://osaurus.ai" rel="nofollow ugc">osaurus.ai</a> 
Osaurus documentation: <a href="https://docs.osaurus.ai" rel="nofollow ugc">docs.osaurus.ai</a>
<h3>External services</h3>
This plugin connects to an Osaurus HTTP server — a local LLM runtime that you install and run yourself on the same machine as WordPress (or on a host you operate). The plugin is useless without it: every text-generation request from the WordPress AI Client is routed to this server.
Service: Osaurus (self-hosted, local LLM runtime)
Osaurus is an open-source Apple Silicon LLM runtime that exposes an OpenAI-compatible HTTP API. It is installed and operated by the same person running WordPress — there is no third-party service operator. This plugin forwards prompts to it so WordPress can perform text generation, chat, tool calls, and structured JSON output without sending data to any cloud provider.
Project homepage and download (<code>osaurus.ai</code>): https://osaurus.ai 
Project documentation (<code>docs.osaurus.ai</code>): https://docs.osaurus.ai
These URLs are referenced in the plugin’s admin UI as documentation/download links only; the plugin does not make HTTP requests to <code>osaurus.ai</code> or <code>docs.osaurus.ai</code>.
Hosts the plugin actually contacts at runtime
The plugin sends HTTP requests only to the host and port resolved (in order) from:
<ol>
<li>The <code>OSAURUS_BASE_URL</code> PHP constant (if defined in <code>wp-config.php</code>).</li>
<li>The <code>osaurus_ai_connector_base_url</code> WordPress option (set from Settings → Connectors).</li>
<li>The built-in default <code>http://127.0.0.1:1337/v1</code>.</li>
</ol>
Two host targets are referenced by name in the plugin code as quick-pick presets and as the default:
<ul>
<li><code>127.0.0.1:1337</code> — loopback on the same machine as WordPress (default). Used by bare-metal installs (Studio, MAMP, Valet, Local, native PHP).</li>
<li><code>host.docker.internal:1337</code> — Docker’s hostname for the host machine. Used by Docker-based WordPress (DDEV, Lando, Docker Desktop). Resolved by Docker, not by a public DNS server.</li>
</ul>
If the user configures a different URL, the plugin only contacts that URL. The plugin never contacts any third-party endpoint operated by the plugin author, the Osaurus project, or any other party.
What data is sent and when
<ul>
<li>When any plugin (including WordPress core) calls <code>wp_ai_client_prompt()</code> and routes through Osaurus, the plugin sends a <code>POST</code> to <code>{base_url}/chat/completions</code> with: the prompt text, the conversation history you supplied, the model ID, sampling parameters (temperature, top_p, max tokens, etc.), and any tool / function declarations or JSON schemas you supplied.</li>
<li>When an admin opens Settings → Connectors, the plugin sends a <code>GET {base_url}/models</code> to populate the model picker and a <code>GET {base_root}/health</code> to display a connection status indicator. Both run only for users with <code>manage_options</code> and are triggered by user actions in wp-admin.</li>
<li>No telemetry, analytics, or background requests are sent.</li>
</ul>
Terms of service and privacy policy
Osaurus is open-source software you self-host. There is no third-party service operator collecting your data. See the project pages linked above for the project’s terms and source. If you point the plugin at a remote Osaurus server operated by someone else (for example, a Mac mini you reach over Tailscale), the operator of that server is the data recipient and is governed by whatever terms you have with them.

AI Provider for Osaurus

Registers Osaurus, a local Apple Silicon LLM runtime, as a provider for the WordPress AI Client.

Ultimate AI Connector for Compatible Endpoints vs AI Provider for Osaurus