Skip to content

multi-provider-support.md

Latest commit

 

History

History
106 lines (71 loc) · 3.76 KB

multi-provider-support.md

File metadata and controls

106 lines (71 loc) · 3.76 KB
Raw

Multi-Provider Support

Summary

The plugin supports multiple AI providers through the WordPress AI Client and connector system. Features and experiments can run against a prioritized provider/model list, automatically fall back when a preferred model is unavailable, and validate capability support before execution.

Overview

For End Users

You can configure one or more AI connectors (for example OpenAI, Google, Anthropic) in WordPress settings. Once configured:

  • Experiments can use any compatible connected provider.
  • Capability checks prevent running requests on unsupported connectors.
  • If a preferred model is unavailable, another configured model/provider can be used.

This allows flexibility in cost, performance, and reliability across provider ecosystems.

For Developers

Provider selection is driven by model preference lists and capability checks:

  • Text workflows use get_preferred_models_for_text_generation().
  • Image workflows use get_preferred_image_models().
  • Vision workflows use get_preferred_vision_models().
  • Features can override preferred ordering using filters.

Most experiment abilities build prompts with using_model_preference( ...$models ), then check runtime support with is_supported_for_text_generation() or is_supported_for_image_generation().

Architecture & Behavior

Connector Detection

Credential detection is connector-aware:

  • has_ai_credentials() inspects wp_get_connectors() for ai_provider connectors with configured authentication.
  • wpai_has_ai_credentials filter allows custom connector implementations to report configured status.
  • has_valid_ai_credentials() performs a runtime support probe using AI client prompt checks.

Model Selection and Fallback

Default model preference arrays are ordered. The first supported and available provider/model pair is used by the AI Client prompt builder.

Example default text preference sequence includes:

  • Anthropic claude-sonnet-4-6
  • Google gemini-3-flash-preview, gemini-2.5-flash
  • OpenAI gpt-5.4-mini, gpt-4.1-mini

When earlier preferences are unavailable, lower-priority entries act as fallback candidates.

Capability Validation

Before execution, abilities commonly call support checks to avoid hard failures and return clear errors:

  • ensure_text_generation_supported() in Abstract_Ability.
  • ensure_image_generation_supported() in Abstract_Ability.

This ensures a feature does not run if no configured provider supports the required capability.

Common Extension Points

Override Preferred Text Models

add_filter( 'wpai_preferred_text_models', function( $models ) {
    return array(
        array( 'openai', 'gpt-5.4-mini' ),
        array( 'anthropic', 'claude-sonnet-4-6' ),
    );
} );

Override Preferred Image Models

add_filter( 'wpai_preferred_image_models', function( $models ) {
    return array(
        array( 'openai', 'gpt-image-2' ),
    );
} );

Report Non-API-Key Connector Credentials

add_filter( 'wpai_has_ai_credentials', function( $has_credentials, $connectors ) {
    // Custom credential validation for alternate auth methods.
    return $has_credentials;
}, 10, 2 );

Operational Notes

  • Configure at least one connector that supports the feature's required capability.
  • Multi-provider setups can improve resilience when individual providers are unavailable.
  • Keep model preference filters aligned with currently available provider model IDs.
  • If no provider supports the requested capability, abilities should return explicit WP_Error responses.

Related Files

  • includes/helpers.php
  • includes/Abstracts/Abstract_Ability.php
  • includes/Services/AI_Service.php
  • includes/Admin/Dashboard/AI_Status_Widget.php
  • includes/Admin/Dashboard/AI_Capabilities_Widget.php