Backends

yaku supports four translation backends. Each sends your text to a different LLM provider.

Available backends

Backend	Default model	Streaming	Notes
`hosted`	Server-side	No	Zero-config default. Refined per-language prompts.
`gemini`	`gemini-2.5-flash`	Yes	Free API key from Google AI Studio.
`openai`	`gpt-4o-mini`	Yes	Works with any OpenAI-compatible API (Groq, Together, DeepSeek, Ollama).
`anthropic`	`claude-haiku-4-5-20251001`	Yes	Via the Anthropic SDK.

All backends use a temperature of 0.3 for consistent, low-variance translations. With --verbose, yaku reports the actual model version used by the API (which may differ from the alias you specified).

How yaku picks a backend

The backend is resolved in this order:

--backend flag — always wins if specified.
Config backend field — used if no flag is passed.
Default — hosted. Having an API key does not change the default. You must explicitly choose a local backend.

# No config at all → uses hosted (the default)
yaku --to en "Bonjour"

# API key set, but no backend configured → still uses hosted
yaku config set api-key AIza...
yaku --to en "Bonjour"

# Explicitly select a local backend to use your API key
yaku config set backend gemini
yaku --to en "Bonjour"

# Override per-command with --backend flag
yaku --backend openai --to en "Bonjour"

Hosted backend

The default backend. Uses refined, per-language prompts with higher translation quality than local backends. See Hosted Service & Plans for a comparison, plan tiers, and quota limits.

# Use the hosted backend (always the default)
yaku --to en "こんにちは"

Customize the hosted endpoint:

yaku config set hosted-url https://api.staging.yakulang.com

Gemini backend

Uses Google’s Gemini API. Get a free API key from Google AI Studio. yaku disables Gemini’s thinking mode to reduce latency and token cost — translation is instruction-following, not a reasoning task.

# Configure
yaku config set api-key YOUR_GEMINI_API_KEY

# Or use environment variable
export GOOGLE_API_KEY=YOUR_GEMINI_API_KEY

Override the model:

yaku --model gemini-2.5-pro --to en "Bonjour"

OpenAI backend

Works with OpenAI’s API and any OpenAI-compatible provider.

# OpenAI
yaku config set backend openai
yaku config set api-key YOUR_OPENAI_API_KEY

# Or use environment variable
export OPENAI_API_KEY=YOUR_OPENAI_API_KEY
yaku --backend openai --to en "Bonjour"

OpenAI-compatible providers

Use --api-base to point to any compatible endpoint:

# Groq
yaku --backend openai \
  --api-base https://api.groq.com/openai/v1 \
  --model llama-3.3-70b-versatile \
  --to en "Bonjour"

# Together.ai
yaku --backend openai \
  --api-base https://api.together.xyz/v1 \
  --model meta-llama/Llama-3-70b-chat-hf \
  --to en "Bonjour"

# Local Ollama
yaku --backend openai \
  --api-base http://localhost:11434/v1 \
  --model llama3 \
  --to en "Bonjour"

# DeepSeek
yaku --backend openai \
  --api-base https://api.deepseek.com/v1 \
  --model deepseek-chat \
  --to en "Bonjour"

Save as defaults so you don’t repeat flags:

yaku config set backend openai
yaku config set api-base https://api.groq.com/openai/v1
yaku config set model llama-3.3-70b-versatile
yaku config set api-key YOUR_GROQ_API_KEY

Anthropic backend

Uses Claude models via the Anthropic SDK.

yaku config set backend anthropic
yaku config set api-key YOUR_ANTHROPIC_API_KEY

# Or use environment variable
export ANTHROPIC_API_KEY=YOUR_ANTHROPIC_API_KEY
yaku --backend anthropic --to en "Bonjour"

Override the model:

yaku --backend anthropic --model claude-sonnet-4-5-20250514 --to en "Bonjour"

Switching between backends

You can switch backends per-command without changing your config:

# Config says gemini, but use openai for this one
yaku --backend openai --to en "Bonjour"

# Compare outputs from different backends
echo "Bonjour" | yaku --backend gemini --to en
echo "Bonjour" | yaku --backend openai --to en
echo "Bonjour" | yaku --backend anthropic --to en
echo "Bonjour" | yaku --backend hosted --to en