AI Translate BYOK (WordPress) changelog

1.5.14

Rate limits (429/503) are retried after the provider's Retry-After instead of failing at once; a rejected API key now says so clearly. Your API key is no longer printed back into the settings page.
Full version: the bulk queue no longer drops posts on a timeout or rate limit (retried up to 3 times); a rejected key or exhausted credit pauses it with a Resume button, and every failed post is listed instead of only the last one.
Full version: the site limit is rechecked every hour (with a "Check again" link), and uninstalling removes all plugin data.

When your AI provider account runs out of credit (or hits its daily quota) the plugin now says so clearly and pauses the bulk and re-sync queues instead of failing every remaining post: top up, then run again.

Long posts into Hindi, Thai, Russian, Arabic, Chinese, Japanese and other non-Latin languages are sent in smaller parts sized for the target script, so they no longer hit the AI output limit; a part that is still cut off is split again and retried instead of failing.
WordPress AI Client (Settings > Connectors): a response cut off by the token limit is now rejected instead of being saved half-translated.

Full version: menus, WooCommerce attributes, Elementor pages and SEO/excerpt fields with many texts are translated in batches, so long lists into languages like Hindi or Japanese no longer hit the AI output limit; texts the AI skips are asked again, and line breaks in excerpts are kept.

Image alt and title attributes inside the content are now translated too.
Full version: ACF text, textarea and WYSIWYG fields are translated (formatting kept); other ACF field types, repeaters and flexible content are copied unchanged.

License: "Deactivate on this site" button frees a seat for another domain without emailing support.
Multisite: network-activated installs take one license key in Network Admin > Settings, and the network counts as one site.

A license key now counts the site it unlocks the full version on (3 sites per license); past the limit the full-version features turn off on that site with a clear notice and upgrade links.

New: on WordPress 7.0+ you can pick "WordPress AI" as the provider: translations go through the AI client built into WordPress, using the connectors already set up in Settings > Connectors, so no API key is pasted into this plugin.
Fix: uninstall cleanup on multisite no longer uses a global variable name without the plugin prefix.

New: DeepL as a provider (Free and Pro keys, detected automatically). Block markup and shortcodes are kept out of the translation and restored unchanged.

After 5 and 25 posts translated one by one, a dismissible notice points to the full version (whole site in one click).
Settings: optional, unticked "Send me setup tips" email opt-in; the email is sent to shop.lumnika.com only if you tick it.

Settings page: links to the other BYOK plugins from Lumnika (static links, no remote calls).

First run: activating the plugin opens the settings with a 2-step getting-started box and a "Translate your first post" button.
After 10 translations, a one-time dismissible notice asks for a WordPress.org review (dashboard, plugins and plugin settings only).
Full version: Elementor pages are translated too: widget texts (headings, text editor, buttons, tabs and other repeaters) in two requests, layout, ids and page template kept.

Fix: the Google Gemini preset now uses `gemini-3.5-flash-lite` (Google no longer serves `gemini-2.5-flash-lite` to new API keys).
Full version: automatic updates from Plugins > Updates, with the license key included in your download.

Fix: the DeepSeek preset now uses `deepseek-flash` (DeepSeek retired `deepseek-chat`); the Google Gemini preset uses the stable `gemini-2.5-flash-lite`.
The cost estimate knows gpt-5-mini, gpt-5-nano, gemini-2.5-flash-lite, deepseek-flash and deepseek-v4-pro (DeepSeek at peak-hour rates).

New: "Provider" list in the settings with OpenAI, OpenRouter, Anthropic (Claude), Google Gemini, DeepSeek, Groq, Mistral and Ollama (local): picking one fills in the base URL and a recommended model, with a link to get the API key. "Custom" keeps any OpenAI-compatible endpoint.
The cost estimate knows the default models of the new providers.
Developers: new `aitr_provider_presets` filter to add or change providers.

Full version: WooCommerce global attributes (Color, Size...) are translated too: each term is translated once and reused across products, and variations and default selections point to the translated terms, so filters and variation pickers show the translated language.
Full version: navigation menus of block themes (Twenty Twenty-Four, Twenty Twenty-Five...) are translated, with links pointing to the translated pages.

New: cost estimate in the editor box before translating (tokens and USD for the selected language), based on the list price of common models or on your own price per 1M tokens in the settings.
Developers: new `aitr_model_prices` filter to add or update model prices.

New: `hreflang` alternate links between an original and its published translations (x-default = original), when Polylang/WPML are not active.
New: `[aitr_languages]` shortcode and "AI Translate: languages" block, a simple language switcher for readers.
Developers: new `aitr_lang_code` filter to map a custom language name to its hreflang code.
Full version: outdated-translation tracking and "Re-translate with AI" (single, bulk, or all outdated at once).

Translations created in the background (WP-Cron) keep the author of the original post.

Fix: translations of Gutenberg posts could lose the block markup (`` comments), because the sanitization step treated them as unrecognized HTML comments and stripped them. Block structure is now preserved.

First release: BYOK translation from the post editor, draft-by-default workflow, translations list page, free/full-version tier.