openapi: 3.0.3 info: title: Translate API version: 1.0.0 description: "**Translate API** — production-grade translation, i18n batch, structure-preserving **HTML** and **JSON** translation,\ \ plus a multilingual **dictionary** for language learners. Powered by **Qwen3-14B-AWQ** served on vLLM.\n\n## Why this\ \ API\n- **Five purpose-built endpoints** — text, i18n strings, HTML, JSON object, dictionary. Pick the one that matches\ \ your data shape.\n- **Structure-aware HTML & JSON** — tags, attributes, inline formatting, nested objects and arrays\ \ are preserved 100%. No more broken markup or stripped IDs.\n- **Brand-safe `ignore_terms`** — pin product names, place\ \ names and technical jargon to keep them verbatim inside translated copy.\n- **Smart skip filters for JSON** — auto-skip\ \ URLs, emails, UUIDs, hashes, ISO dates, currency and code/ID patterns. Plus per-path and per-key exclusion rules with\ \ wildcard support.\n- **Idiom-aware prompting + script-purity validation** — output is always in the target language's\ \ native script, never mixed with foreign characters.\n- **High throughput** — vLLM continuous batching gives ~78 strings/s\ \ on JSON, ~40 segments/s on HTML.\n\n## Endpoints\n| Endpoint | Best for |\n|---|---|\n| `POST /v1/translate` | Single\ \ sentences and small batches |\n| `POST /v1/json` | i18n key-string arrays |\n| `POST /v1/translate-html` | Articles,\ \ product descriptions, CMS exports |\n| `POST /v1/translate-json` | Product catalogs, structured documents |\n| `POST\ \ /v1/dict` | Language-learning apps, dictionaries, vocabulary builders |\n\n## Supported languages\nVietnamese (`vi`),\ \ English (`en`), Japanese (`ja`), Chinese Simplified (`zh`) and Traditional (`zh-TW`), Korean (`ko`), French (`fr`),\ \ German (`de`), Spanish (`es`), Russian (`ru`), Thai (`th`), Indonesian (`id`), Portuguese (`pt`), Italian (`it`), Arabic\ \ (`ar`), Hindi (`hi`) and many more ISO 639-1 codes.\n\n## Quick start\n```bash\ncurl -X POST https://translate.spacecloud.fit/v1/translate\ \ \\\n -H 'Content-Type: application/json' \\\n -d '{\"text\":\"Hello world\",\"target_lang\":\"vi\"}'\n```\n\nReturns:\ \ `{\"translations\":[{\"translated_text\":\"Xin chào thế giới\",\"detected_source_lang\":\"en\"}]}`\n\n## Pricing & rate\ \ limits\nManaged by RapidAPI based on your subscription tier." termsOfService: https://translate.spacecloud.fit/terms contact: name: Translate API Support url: https://translate.spacecloud.fit email: support@spacecloud.fit license: name: Proprietary x-logo: url: https://translate.spacecloud.fit/logo.png altText: Translate API servers: - url: https://translate.spacecloud.fit description: Production tags: - name: Translation description: Translate plain text, i18n batches, HTML and JSON. - name: Dictionary description: Multilingual dictionary lookup for language learners. paths: /v1/translate: post: tags: - Translation summary: Translate text — single string or batch. description: '**Use case:** translate a single sentence, paragraph or up to 100 strings in one call. **Behavior:** when `source_lang=auto`, every item carries its own `detected_source_lang` — useful for mixed-language batches. With an explicit `source_lang`, the response is a plain `list[str]` for compact integration. **Supported languages:** Vietnamese, English, Japanese, Chinese (Simplified + Traditional), Korean, French, German, Spanish, Russian, Thai, Indonesian, Portuguese, Italian, Arabic, Hindi and more. **Quality:** powered by Qwen3-14B-AWQ on vLLM with idiom-aware prompting, script-purity post-validation and single-pass retry on mixed-script outputs. **Latency:** ~80–200 ms per item (single text), batches benefit from continuous batching on the GPU.' operationId: translate requestBody: required: true content: application/json: schema: $ref: '#/components/schemas/TranslateRequest' examples: single_auto: summary: Single text + auto detect value: text: Hello world target_lang: vi single_explicit: summary: Single text + explicit source value: text: Tôi yêu Việt Nam source_lang: vi target_lang: en batch_explicit: summary: Batch + explicit source (en→vi) value: text: - Good morning - Thank you - How are you? source_lang: en target_lang: vi batch_auto_mixed: summary: Batch + auto detect on mixed input value: text: - こんにちは - 안녕하세요 - Bonjour target_lang: vi responses: '200': description: Translation result. content: application/json: schema: $ref: '#/components/schemas/TranslateResponse' examples: auto_with_detected: summary: 'Auto: each item carries detected_source_lang' value: request_id: 03f0ece3-9ad8-4b52-ad0c-40da09487722 processing_time_ms: 126 translations: - translated_text: Xin chào thế giới detected_source_lang: en explicit_strings: summary: 'Explicit source: list of strings' value: request_id: f4bfc370-747e-40c7-a3c0-eb1cb86683af processing_time_ms: 86 translations: - Chào buổi sáng - Cảm ơn - Bạn khỏe không? auto_mixed_batch: summary: Auto mixed batch value: request_id: 263aafcd-ea31-423e-bd4b-1591d2f6cbe3 processing_time_ms: 192 translations: - translated_text: Xin chào detected_source_lang: ja - translated_text: Xin chào detected_source_lang: ko - translated_text: Xin chào detected_source_lang: fr '422': description: Validation error. content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' '502': description: Upstream LLM error. content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' /v1/json: post: tags: - Translation summary: Translate an array of strings (i18n batch). description: '**Use case:** translate localization key-strings for an app''s i18n bundle. Same response shape as `/v1/translate`, but the request always uses an array, matching how i18n libraries (i18next, FormatJS, ICU, ...) store keys. **Order is preserved** — `translations[i]` corresponds to `texts[i]`.' operationId: translateJson requestBody: required: true content: application/json: schema: $ref: '#/components/schemas/JsonTranslateRequest' examples: i18n_explicit: summary: i18n keys (en→ja) value: texts: - Welcome - Sign in - Sign up - Forgot password? source_lang: en target_lang: ja i18n_auto: summary: i18n auto detect value: texts: - Welcome - ようこそ - 환영합니다 target_lang: vi responses: '200': description: Translations in input order. content: application/json: schema: $ref: '#/components/schemas/JsonTranslateResponse' examples: explicit_strings: summary: Explicit output (en→ja) value: request_id: 8a24186d-df3b-4843-8caf-63702cc1e84c processing_time_ms: 71 translations: - ようこそ - サインイン - サインアップ - パスワードをお忘れですか？ '422': description: Validation error. content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' '502': description: Upstream LLM error. content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' /v1/translate-html: post: tags: - Translation summary: Translate HTML — preserve structure, ignore brand names. description: '**Use case:** translate articles, product descriptions, emails, CMS pages or any HTML where tags, attributes and inline formatting must be preserved verbatim. **How it works:** 1. Parse with `lxml` (fast path) → fall back to `html5lib` (browser-grade) when errors are heavy. 2. Score health into `clean` / `minor` / `moderate` / `severe`. Severely malformed HTML returns **422** asking the caller to fix the HTML first. 3. Walk the DOM. Skip ` source_lang: en target_lang: vi article_full: summary: Full article with nested inline +

The Future of AI

Artificial intelligence is changing the world. Machine learning models like GPT and Claude are revolutionizing how we interact with computers.

However, responsible AI requires careful consideration of ethics and safety.

Item one
Item two

Chào xe đỏ!

Apple tạo ra MacBook Pro mới và iPhone. Mua ngay!

Tương lai của trí tuệ nhân tạo

Trí tuệ nhân tạo đang thay đổi thế giới. Các mô hình machine learning như GPT và Claude đang cách mạng hóa cách chúng ta tương tác với máy tính.

Tuy nhiên, trí tuệ nhân tạo có trách nhiệm đòi hỏi phải cân nhắc cẩn trọng đến đạo đức và an toàn.

Mục một
Mục hai

Hello world!