Vachr Web Search Connector for Muse AI

Description

Vachr Web Search Connector for Muse AI connects WordPress to the Meta Model API (https://api.meta.ai/v1). It adds Muse AI as a provider to the WordPress AI Client and the native Settings > Connectors screen, so one API key serves the whole site.

  • Plain chat uses the OpenAI-compatible Chat Completions endpoint.
  • Web search uses the Responses API with Meta’s server-side web_search tool and returns the source links the model found.
  • The API key stays on your WordPress server. It is never sent to visitors or to external applications.

What it does

  • Adds the [vachr_ai_muse] shortcode for an optional visitor chat.
  • Adds a protected POST /wp-json/vachr-ai-muse/v1/chat endpoint for external applications.
  • Provides vachr_ai_muse_query() for other plugins and themes on the same site.
  • Registers Muse AI in the native Connectors screen (WordPress 7.0+). The API key, created at https://dev.meta.ai, is managed there.
  • Lets other plugins use Muse AI through the WordPress AI Client, including structured JSON responses.
  • Includes a connection diagnostics tool that shows the real reason when a key cannot be connected.
  • Ships with translations for 13 languages.

How web search works

Web search is performed by Meta’s server-side tool, not by WordPress and not by a crawler included in this plugin. It is on by default and can be turned off in Settings > Vachr Web Search Connector for Muse AI (this switch applies to every request, including those made by other plugins). It can also be disabled for a single request:

  • REST API: send "web_search": false.
  • Visitor chat: clear the “Search the web” checkbox.
  • PHP helper: pass false as the third argument.

When web search is used, Muse AI may run one or more searches before it answers. This can increase latency and API cost. Source links are returned only when the provider includes them (at most 10 are shown).

Models

The model list is read from your key’s /v1/models response, and only text models are offered (image, voice and segmentation models are hidden). The default is muse-ember-1.0, the lowest-cost reasoning model. You can enter any model ID your key may use, for example muse-spark-1.3 or muse-wonder-1.0.

If the API accepts your key but returns no models, the connection is refused with an explanation; the plugin never substitutes a made-up model list. Enable a text model for the key at https://dev.meta.ai.

Languages

Czech, French, German, Italian, Japanese, Korean, Polish, Portuguese (Brazil), Russian, Simplified Chinese, Spanish, Thai and Turkish. The plugin follows the language of your site or of the logged-in user. Translations were prepared by machine; corrections are welcome.

External service and privacy

This plugin connects to the Meta Model API (https://api.meta.ai/v1), operated by Meta Platforms, Inc. The service provides the Muse AI models that generate the answers and the optional web search. Without it the plugin cannot work.

Data sent and when:

  • Every time a chat message, REST API request or PHP helper call is made: the question text, the selected model, the system instructions, whether web search was requested, and your API key (as an Authorization header). Web search can cause Meta to process the question through its search service.
  • When an administrator opens the Connectors screen: a request to https://api.meta.ai/v1/models with the API key to list the available models.
  • When an administrator uses “Test connection” on the settings page: the key typed into the field is sent to https://api.meta.ai/v1/models. The key is not saved.

Your API key is sent only from the WordPress server to Meta, never to visitors. Create and manage keys at https://dev.meta.ai.

  • Meta Terms of Service: https://www.facebook.com/terms
  • Meta Privacy Policy: https://www.facebook.com/privacy/policy/

The plugin does not store questions, responses, source links or conversations in the WordPress database. It does not use analytics or tracking.

If the site owner enables the WordPress AI feature AI Request Logging, requests made through this connector appear in its central AI Request Logs according to that feature’s own settings.

Before enabling public chat, inform visitors about the data transfer and establish a lawful basis appropriate to your jurisdiction. Do not submit sensitive personal data.

Security

  • The external REST endpoint requires an integration bearer token, generated on activation and visible only to WordPress administrators. Treat it like a password and replace it if it may have been exposed.
  • Browser chat requests use a WordPress REST nonce. Public chat is disabled by default.
  • Public browser requests are limited to 12 per five minutes per IP address.
  • Questions are limited to 12,000 characters.

Developer usage

REST API

Send a JSON request to your own WordPress site. Never place the integration token in browser JavaScript.

POST https://your-site.example/wp-json/vachr-ai-muse/v1/chat

Headers: Authorization: Bearer YOUR_INTEGRATION_TOKEN and Content-Type: application/json

curl -X POST https://your-site.example/wp-json/vachr-ai-muse/v1/chat -H "Authorization: Bearer YOUR_INTEGRATION_TOKEN" -H "Content-Type: application/json" -d '{"message":"What changed in WordPress this week?","web_search":true}'

Request fields:

  • message (string, required): the question, up to 12,000 characters.
  • web_search (boolean, optional): use web search; defaults to true. Ignored when web search is switched off in the plugin settings.

Response: { "text": "...", "sources": [ { "title": "...", "url": "..." } ], "usage": { ... } }

PHP helper

$result = vachr_ai_muse_query( 'Summarise today\'s WordPress news.', '', true );

Arguments: the question, an optional model ID (empty uses the saved model), and whether to use web search. It returns an array with text, sources and usage, or a WP_Error.

WordPress AI Client

Muse AI is registered as the provider muse, so other plugins can use it with wp_ai_client_prompt()->using_provider( 'muse' ). Request web search with the web_search custom option on the model configuration.

Filters

  • vachr_ai_muse_request_timeout (float, default 600.0): request timeout in seconds.
  • vachr_ai_muse_web_search_enabled (bool): override whether web search is allowed.

Installation

  1. Install and activate the plugin on WordPress 7.0 or newer.
  2. Go to Settings > Connectors and add your Meta Model API key (create one at https://dev.meta.ai).
  3. Go to Settings > Vachr Web Search Connector for Muse AI, choose a model (default muse-ember-1.0) and save.
  4. To show the visitor chat, enable “Allow Chat for Visitors” and add [vachr_ai_muse] to a page or post.
  5. For an external integration, copy the Integration Token from the same settings screen into that application’s secret manager.

If the connector will not connect, open Settings > Vachr Web Search Connector for Muse AI > Connection diagnostics, paste the key and press “Test connection”. The report shows the HTTP result and the exact error.

FAQ

The Connectors screen says “It was not possible to connect to the provider using this key.”

That message is generic. Use Connection diagnostics (see Installation) to see the real cause: a blocked outbound connection, an invalid key, or a key that has no models enabled.

My key is accepted but the connector will not connect.

If /v1/models returns an empty list, no model is enabled for the key. Enable a text model at https://dev.meta.ai and try again.

Do I need to add the API key to this plugin’s settings?

No. The key is stored by WordPress in Settings > Connectors and is passed to the plugin by the AI Client.

Why are the answers slow?

Muse models reason before answering, and web search adds one or more search steps. Lower the cost and latency by using muse-ember-1.0 or by turning web search off.

Can visitors use the chat?

Only if you enable “Allow Chat for Visitors” and place [vachr_ai_muse] on a page. Otherwise only administrators can use it.

Reviews

There are no reviews for this plugin.

Contributors & Developers

“Vachr Web Search Connector for Muse AI” is open source software. The following people have contributed to this plugin.

Contributors

Changelog

1.0.0

  • First release

0.1.2

  • Rename the shortcode to [vachr_ai_muse].
  • Prefix all classes, functions, options, hooks, the REST namespace and the CSS classes with vachr_ai_muse to avoid name collisions. The settings are stored under a new option name, so re-save them after updating.
  • The REST route is now /wp-json/vachr-ai-muse/v1/chat and the PHP helper is vachr_ai_muse_query().
  • Stop declaring the connector setting name; WordPress derives it automatically.
  • Document the external Meta Model API service, the data sent and the terms and privacy links.

0.1.1

  • Add a setting to enable or disable Web Search for all requests; other plugins request it with the web_search custom option.
  • Support structured JSON responses requested through the WordPress AI Client.
  • Stop inventing a fallback model list: when the API returns no models for the key, an explanatory error is shown.
  • Add a Connection diagnostics tool to the settings page.
  • Skip the intermediate progress message that web search placed before the final answer, and limit sources to 10.
  • Add translations for 13 languages and load the bundled translations without load_plugin_textdomain().
  • Use muse-ember-1.0 as the default model.
  • Show the full plugin name in the Settings menu.

0.1.0

  • Initial version.