⚡️ JSON Translate

Localize i18n JSON into 120+ languages — values only, keys and schema untouched

English · 简体中文

Feed a locale file to a general-purpose translator and it happily translates your keys too — "submit" becomes "soumettre", your nesting collapses, and the file no longer loads. JSON Translate walks the tree locally and sends only the string values; keys, hierarchy, arrays, and non-string types are never in the request to begin with.

JSON Translate is a free, browser-based JSON localization tool for next-intl, i18next, vue-i18n, react-intl, and any other key-value structure. Five modes decide what gets translated — everything, a JSONPath subtree, named keys, one field across every entry, or all languages merged into one file — then connect 9 traditional translation APIs (DeepL, Google, Azure, DeepLX, Qwen-MT, TranslateGemma, MiLMMT, GTX, Edge) or 26 LLM providers and gateways and translate into 120+ languages, several at once. Everything runs locally in your browser; source content and API keys never touch a server. A CLI drives the same engine headlessly when you'd rather script it.

👉 Try it online: https://tools.newzone.top/en/json-translate

Translation Modes

Pick the mode first — it decides which values are collected. Everything else (engine, languages, glossary) is shared across all five.

Mode	What it translates	Use it for
Global translation	Every string value, at any depth	Translating a whole locale file in one go
Keys under a specific node	Values matched by JSONPath ($.products[*].name, $..title; comma-separated for several)	Large files where only a subtree needs translating
Specify keys	Only the named keys, optionally written to different output keys (name → name_zh)	Translating title / description while keeping the originals
One field across all entries	The same field inside every top-level entry, from an optional start key onward	Error / log catalogs where only message should change, code and level stay
i18n mode	Adds a field per target language beside the source-language field	Merging every language into one file

i18n mode is the one worth an example. With en as source:

{ "title": { "en": "Settings" } }

translating to zh and fr gives:

{
  "title": {
    "en": "Settings",
    "zh": "设置",
    "fr": "Paramètres"
  }
}

Existing target-language fields are skipped rather than overwritten, so re-running only fills the gaps. Combined with multi-language output, one pass produces a single file holding the source and every target language.

Structural requirements. One field across all entries needs a two-level shape (entry → object); if a listed field is missing everywhere, the run aborts rather than silently skipping it. Specify keys is case-sensitive, and keys containing dots (share.owner) collide with JSONPath's nesting syntax — rename them or use another mode.

Key Features

• Schema-Preserving: Only string values are collected and sent; keys, key order, nesting, arrays, numbers, booleans, and null are never part of the request.
• Mapped Translation: Write results to different output keys (name → name_zh) so the source values survive alongside the translation.
• Multi-Language Output: Translate into several target languages in one pass — each exported as its own file, or merged inline in i18n mode.
• Forgiving Input: JSON5 is accepted — trailing commas, comments, unquoted keys, single quotes — and a bare fragment without its outer {} / [] is wrapped for you.
• Precision Warnings: Integers beyond the IEEE-754 safe range (snowflake IDs, order numbers) and integer-like keys that a JSON round-trip re-sorts are flagged before you ship the file.
• Glossary: Lock brand names, SKUs, and API terms to a fixed translation — enforced in the LLM prompt and re-checked against the output, so the same term renders identically in every locale.
• Context-Aware Translation (LLM only): Surrounding values sent as context for better coherence and terminology consistency.
• Unlimited Caching (IndexedDB): Every translated value is cached locally with no browser-storage size limit — only changed strings cost an API call on the next run.
• Command Line: yarn cli runs the same engine and cache from a terminal — see Command Line.
• Multi-Locale UI: Powered by next-intl, with full UI translation across 18 languages.
• Private by Design: Fully client-side — source content and API keys stay in your browser; LLM requests go directly from your browser to the API endpoint you configure.

Translation APIs

Supports 9 traditional MT APIs and 26 LLM providers and gateways.

Traditional APIs

API	Quality	Stability	Free Tier
DeepL	★★★★★	★★★★☆	500K chars/month
Google Translate	★★★★☆	★★★★★	500K chars/month
Azure Translate	★★★★☆	★★★★★	2M chars/month (first 12 months)
DeepLX (Free)	★★★★☆	★★★☆☆	Self-host or free public endpoints
Qwen-MT	★★★★☆	★★★★☆	Alibaba DashScope quota
TranslateGemma	★★★★☆	★★★★☆	Self-host (LM Studio / llama.cpp / etc.)
MiLMMT	★★★★☆	★★★★☆	Self-host (LM Studio / llama.cpp / etc.)
GTX API (Free)	★★★☆☆	★★★☆☆	Free (rate-limited)
Edge API (Free)	★★★★☆	★★★☆☆	Free (rate-limited)

GTX and Edge need no configuration at all — they are the zero-setup defaults, and each is the other's fallback.

LLM Providers

DeepSeek, OpenAI, Claude, Gemini, Qwen, Kimi (Moonshot), Doubao (Volcengine), Xiaomi MiMo, Zhipu GLM, MiniMax, StepFun, Baidu ERNIE (Qianfan), Mistral, xAI (Grok), Cohere, and YandexGPT.

Gateways

OpenRouter, OpenCode Zen, TokenHub (Tencent), Groq, Cerebras, SiliconFlow, Atlas Cloud, Nvidia NIM, Azure OpenAI, plus any Custom (OpenAI-compatible) endpoint (Ollama / LM Studio / vLLM / LiteLLM / Together AI / Fireworks AI etc.).

Providers walled off from browsers by CORS can be routed through an API relay. The built-in relay works out of the box; API Settings → Relay address points every relayed provider at your own deployment of the relay Worker instead.

LLM modes give you:

• Best for: long values (product copy, tutorials, help text), and anything where terminology has to stay consistent
• Customization: configure system / user prompts to lock in terminology and tone
• Temperature Control: adjust AI creativity (0–1 scale)
• Thinking Mode: per-provider toggle for reasoning-capable models

Context-Aware Translation (LLM only)

LLM modes can send surrounding values as context for each batch, improving coherence and terminology consistency across a locale file.

• Concurrent Lines: max values translated in parallel (default 20). Too high triggers rate limits.
• Context Lines: values included per batch as context (default 50). Higher = better coherence but more tokens.

FAQ

Machine translation or LLM? Pick by value length. Short UI strings — button labels, menu items, error messages — come out fine on machine translation, and GTX / Edge cost nothing. Longer values (product descriptions, onboarding copy, help text) read markedly better on an LLM: DeepSeek is the value pick, Claude Sonnet is the most reliable on terminology.

Does it work with next-intl, i18next, vue-i18n, and react-intl? Yes — nested objects, flat-key namespaces, and arrays are all handled, because the tool works on the parsed tree rather than on a framework-specific schema. Keys are never sent to the engine.

What about ICU placeholders like {name} or {count, plural, ...}? They live inside the value, so they do go to the engine. Most engines leave them alone; if one starts rewriting them, put the placeholder in the glossary as a source → target pair and the leak-through net restores it after translation.

How do I translate only some fields? Use Specify keys for named keys, Keys under a specific node for a JSONPath subtree, or One field across all entries when every entry has the same shape. Mapped translation writes results to new keys so the originals survive.

Will my key order change? Only for integer-like keys ("1", "42"). A JSON round-trip re-sorts those ahead of string keys — that's a language-level rule, not a bug in the tool — and the UI warns you when a file contains them.

Is it private? Yes. Parsing, translation, and caching all run client-side. API keys are stored only in local browser storage, and LLM requests go directly from your browser to your configured endpoint.

See the full FAQ in the docs for more.

Command Line

yarn cli translates files headlessly over the same engine as the web app — same retry and rate-limit handling, same cache keys. Configure the service once in the browser, hit Export settings, and hand the JSON to the CLI; nothing has to be re-entered.

The CLI has no mode picker: it translates every string value in the file, which is what locale-file batches want. Reach for the web app when you need JSONPath targeting, key mapping, or i18n merging.

yarn install   # once

# Several locale files into Chinese on the free GTX API — no key, no config.
yarn cli -i en.json -i errors.json -t zh

# Two targets in one pass, using your exported keys/prompts/glossary.
yarn cli -i messages/en.json -t ja -t ko -s ~/json-settings.json -o messages/

# A local model — nothing leaves the machine.
yarn cli -i en.json -t zh -m llm --url http://localhost:11434/v1 --model qwen3

# One-off override on top of a settings file.
yarn cli -i en.json -t de -m deepseek --api-key sk-xxx

Output lands beside the input (or in -o <dir>) as en.zh.json.

Option	Meaning
-i, --input <file>	Input file. Repeatable.
-t, --to <lang>	Target language. Repeatable. Default zh.
-f, --from <lang>	Source language. Default auto.
-m, --method <id>	Service id. Default gtxFreeAPI; --list-methods prints them all.
-s, --settings <file>	Settings JSON exported from the web UI (keys, prompts, glossary, retry…).
-o, --out-dir <dir>	Output directory. Default: next to each input.
--api-key · --url · --model	One-off overrides for the chosen service.
--no-cache · --cache-file <file>	Cache control. Default ~/.translate-cli-cache.json.
--relay · --no-relay	Route through the API relay. Off by default — Node has no CORS to work around.
--format <fmt>	Force a format instead of inferring it from the extension.

⚠️ The CLI requires strict JSON. The web app additionally accepts JSON5 (trailing commas, comments); a hand-edited locale file that translates fine in the browser will be rejected here rather than silently rewritten.

JSON is not the only input: the same command handles subtitles (.srt, .ass, .vtt, .lrc, .sbv — timecodes stripped locally) and Markdown (.md, .markdown, .mdx — code blocks, links, and LaTeX protected by default). yarn cli --list-formats prints the mapping, yarn cli --help the full option list.

Runs are resumable: every translated value is cached, so re-running after a Ctrl-C, a rate-limit wall, or a handful of failed values only pays for what is still missing.

Exit codes: 0 everything translated · 1 finished but some values soft-failed (kept as source text in the output) or a file failed · 2 bad invocation · 130 cancelled.

Run It Yourself

Node.js >= 20.9.0 and Yarn (or npm / pnpm).

git clone https://github.com/rockbenben/json-translate.git
cd json-translate

yarn install
yarn dev        # http://localhost:3000
yarn build      # production build

Documentation & Deployment

For detailed configuration, API setup, and self-hosting instructions, see the Official Documentation.

Quick Deployment: Deploy Guide

Contributing

Contributions are welcome! Feel free to open issues and pull requests.

1. Fork the repo and create a feature branch
2. Run yarn and yarn dev locally
3. Add tests / docs when applicable
4. Submit a PR with a clear description