SlyTranslate – AI Translation Abilities

Description

SlyTranslate brings practical AI translation to WordPress. It is built for teams that need translation directly in editing workflows and also want the same workflows available through REST and MCP automation.

Why this plugin?

Use SlyTranslate when you need one consistent translation workflow for:

• page/post translation in wp-admin
• TranslatePress visual-editor translation on the current page
• inline selected-text translation in Gutenberg
• Gutenberg block translation
• bulk translation from list-table actions
• SEO title/description translation in the same process

Internal flow (short):

• Uses native WordPress AI connectors through wp_ai_client_prompt().
• Registers translation workflows as WordPress Abilities.
• Exposes abilities over REST (/wp-abilities/v1/) and MCP discovery.
• Supports long/structured content with chunking and output validation.
• Optional direct_api_url supports OpenAI-compatible endpoints for model-specific payload needs.
• In WP Multilang mode, translation state is detected from language-specific content so placeholder titles do not count as completed translations.
• List-table translation now includes an explicit overwrite option with a confirmation step.
• TranslatePress editor pages get a SlyTranslate sidebar panel that can translate the current singular page with model selection, overwrite, progress, and cancel controls.

Abilities:

ai-translate/get-languages:  List languages exposed by the active language plugin |
ai-translate/get-translation-status: Show translation status for a content item, including `source_language` and `single_entry_mode` |
ai-translate/set-post-language: Change the language assignment of an existing content item (only exposed when supported, e.g. Polylang) |
ai-translate/get-untranslated: Find content still missing a target translation |
ai-translate/translate-text: Translate arbitrary text |
ai-translate/translate-blocks: Translate serialized Gutenberg blocks |
ai-translate/translate-content: Create or update one translated post/page/CPT entry (call `get-translation-status` first; optional `source_language` + `overwrite`) |
ai-translate/translate-content-bulk: Bulk-translate multiple entries (supports optional `source_language` and `overwrite`) |
ai-translate/get-progress: Return live progress for a running translation |
ai-translate/cancel-translation: Cancel a running translation |
ai-translate/get-available-models: List models from configured connectors |
ai-translate/save-additional-prompt: Save per-user additional instructions |
ai-translate/configure: Read or update persistent plugin settings |

MCP Call Flow:

• Call ai-translate/get-languages first when the correct target language code is unknown.
• Call ai-translate/get-translation-status before ai-translate/translate-content to inspect source_language, single_entry_mode, and whether the target language already exists.
• Omit source_language unless you intentionally pin a source variant. In single-entry adapters, reuse get-translation-status.source_language when you do pin it.
• Set overwrite=true only when status or prior context shows that the target language already exists.
• Expect translated_post_id to equal source_post_id in single-entry adapters such as WP Multilang, WPGlobus, and TranslatePress. In multi-post adapters such as Polylang, the translated item uses a sibling post ID.

Requirements:

• WordPress 6.9+
• PHP 8.1+
• An AI connector configured in WordPress (Settings > Connectors)
• A supported language plugin (Polylang, WP Multilang, WPGlobus, or TranslatePress Multilingual) for content-translation workflows across posts/pages/CPTs
• WordPress MCP Adapter if you want MCP discovery

Supported plugins:

• Language plugin: Polylang, WP Multilang, WPGlobus, TranslatePress Multilingual
• SEO plugins: Genesis SEO, Yoast SEO, Rank Math, All in One SEO, The SEO Framework, SEOpress, Slim SEO

Supported model profiles:

Any LLM available through a WordPress AI connector works out of the box — no special configuration needed. The following model families additionally have dedicated built-in profiles that tune prompt style, chunking, and retry behavior for better results:

• TranslateGemma: dedicated runtime with chat_template_kwargs support through direct_api_url
• TowerInstruct / Salamandra: bilingual framing, conservative chunking, stricter retry behavior
• Nvidia Nemotron: system-prompt-aware with reasoning-disable and provider-parameter forwarding
• Qwen 3.x / GLM-4.6v / Gemma 4 / Phi-4: thinking-aware profiles
• EuroLLM / Llama 3.1-8B / SauerkrautLM: conservative chunking tuned for European languages
• Ministral-3 / Ministral-8B: optimized for the Ministral model family