Smart Match for WooCommerce

Changelog

1.2.0

• Added an AI Effort setting (Economy, Balanced, Thorough) that controls how much reasoning the AI spends per analysis on models that support it. The Economy default reduces API cost and speeds up analyses.
• Product analysis and content generation now show an estimated USD cost alongside the token count, so you can judge the real cost of a generation at a glance instead of calculating it from tokens yourself.
• Added support for the site-wide WordPress AI connector (WordPress 7.0+). A new AI Connection setting chooses between the connector and a plugin API key, with automatic mode preferring the connector when one is configured — no plugin API key required in that case.
• Made it clear which AI provider, model, and connection the plugin will use: the settings page now shows a live “Currently active” summary and per-provider connector status, greys out the API key field when the site-wide WordPress AI connector handles authentication, and the analysis results show which connection served each request.
• Refreshed the AI model lists to the current provider lineups: OpenAI GPT-5.6 family, Anthropic Claude 5 family, and Google Gemini 3.x, with previous-generation models kept as legacy options. A previously selected model that is no longer listed stays selected instead of being switched silently.
• The content generator now shows the same “Continue anyway?” confirmation as the category analyzer when the rate-limit warning threshold is reached, instead of a dead-end “please wait” error.
• Fixed “Content applied!” being shown even when saving the selected categories failed. The success message now appears only after categories are actually saved, and a failure shows an error and lets you try applying again.
• Fixed the icon on the “Generate with AI” button sitting slightly too high instead of being vertically centered with the button label.
• Fixed long product descriptions in non-Latin languages (Cyrillic, CJK, Arabic, emoji) potentially being cut mid-character when shortened for the AI request.
• Fixed requests silently routing to a different AI provider than the one selected when the chosen provider had no site-wide connector configured; an explanatory error is now shown instead.
• Fixed OpenAI reasoning models (GPT-5 family) failing with an API parameter error when analyzing products, generating content, or testing the connection.
• Fixed the Quick Settings modal closing with no confirmation after saving, and silently discarding changes when the save failed. Saving now shows a confirmation message, and a failed save keeps the modal open with an error so changes are not lost unnoticed.
• Fixed “Test Connection” testing the previously saved configuration instead of the one currently selected on the settings page. The test now uses the provider, model, and AI Connection mode shown in the form, so switching options and testing before saving gives an accurate result.
• Fixed API keys remaining in the database after plugin uninstall — all per-provider keys (OpenAI, Anthropic, Google Gemini) are now removed when the plugin is deleted.
• Fixed content generation failing with a “Bad Request (400) – Unsupported parameter: ‘temperature’” error when using the WordPress AI connector with newer OpenAI models such as GPT-5.x.
• Fixed the connection test always failing with a “max_output_tokens” error when using the site-wide WordPress AI connector with OpenAI.
• Fixed AI requests through the site-wide WordPress AI connection timing out after 5 seconds regardless of the API Timeout setting, causing “Operation timed out” errors during analysis and content generation.
• API key encryption upgraded to authenticated AES-256-GCM with a unique random IV per key (existing keys migrate automatically). If a saved key can no longer be decrypted — typically after the site’s WordPress salts change — the settings page now clearly asks you to re-enter it instead of failing silently, and keys are never stored unencrypted.
• The Google Gemini API key is now sent via a request header instead of the URL, so it can no longer appear in proxy or server logs. Debug-log redaction also covers all provider key formats (OpenAI, Anthropic, Google).
• AI analysis, content generation, and category application now verify that the user can edit the specific product, not just products in general. On stores with restricted roles (e.g. vendors), users can no longer run these actions on products they are not allowed to edit.

1.1.0

• New: AI Content Generator — generate product title, short description, full description, tags, categories, and variation descriptions from a single modal
• New: Google Gemini provider support (Gemini 2.0 Flash, 2.5 Flash, 2.5 Pro)
• New: Variation descriptions — unique AI-generated descriptions per product variation
• New: Style and length controls for content generation
• New: Persistent form preferences — modal remembers your last used settings
• Updated: OpenAI model list — added GPT-4.1 family (nano, mini, full) and GPT-5.x reasoning models (nano, mini, 5.2)
• Updated: Anthropic model list — updated to Claude Haiku 4.5, Sonnet 4.6, Opus 4.6
• Improved: Metabox renamed to “Smart Match AI” to reflect full feature set
• Fixed: Various code quality improvements from full code review

1.0.0

• Initial release
• Support for OpenAI and Anthropic AI providers
• Product analysis metabox on edit screens
• Category suggestions with confidence scores
• One-click category application
• Customizable settings and prompts
• Rate limiting and usage warnings
• API key encryption
• Full internationalization support