Translation Engines

LocaleKit supports multiple translation engines. Choose based on your needs:

Comparison

Engine	Type	European	Asian	Context	Speed	API Key
DeepL	Cloud	★★★★★	★★★☆☆	★★★☆☆	2.5 req/s	DEEPL_API_KEY
OpenAI	Cloud	★★★★★	★★★★☆	★★★★★	20 req/s	OPENAI_API_KEY
MLX	On-device	★★★★☆	★★★★☆	★★★★☆	0.5 req/s	—
Apple Intelligence	🖥 macOS app only	★★★★☆	★★★★☆	★★★★★	1 req/s	—

DeepL

Best for European languages. Fast batch processing (50 keys/batch). Requires a paid API key.

export DEEPL_API_KEY="your-key"
localekit translate --engine deepl

OpenAI

Best context awareness — understands UI strings, placeholders, and tone. Fast batch processing (50 keys/batch). Supports all languages GPT supports.

export OPENAI_API_KEY="your-key"
localekit translate --engine openai

MLX

Runs entirely on your Mac using Apple Silicon. No API key, no cloud, no cost. Models download automatically on first use and are cached at ~/Library/Caches/huggingface/.

localekit translate --engine mlx
# Or specify a model:
localekit translate --engine mlx --mlx-model mlx-community/Qwen3-8B-4bit

Available Models

Model	Parameters	Size	RAM	Languages
Qwen3 4B default	4B	2.5 GB	8 GB	119
Mistral 7B	7B	4.1 GB	8 GB	15
Qwen3 8B	8B	5 GB	16 GB	119
Gemma 3 12B	12B	8 GB	16 GB	35
Qwen3 30B-A3B	30B (3B (MoE) MoE)	16 GB	16 GB	119
Qwen3 32B	32B	18 GB	32 GB	119

Recommendation

8 GB Mac → Qwen3 4B (default). 16 GB+ Mac → Qwen3 30B-A3B (best quality-to-speed with MoE).

Apple Intelligence

🖥 macOS App Only

Apple Intelligence is available in the LocaleKit macOS app but not in the CLI. Use DeepL, OpenAI, or MLX for CLI workflows.

Requires macOS 26+ with Apple Intelligence enabled. Best context awareness (5/5), processes one key at a time.