Sakura AI Engine

Sakura AI Engine (さくらのAI Engine) is an inference service operated by Sakura Internet and hosted in Japan. It exposes an OpenAI-compatible Chat Completions API, so Kurahi talks to it with the standard openai-completions adapter. It is the first profile kurahi init offers and the default for evaluations: no card details are needed to start, and prompts do not leave Japan.

Japanese version: さくらのAI Engine.

Facts (verified 2026-10-13 against the official pages)

Item	Value
Base URL	https://api.ai.sakura.ad.jp/v1
Chat endpoint	POST https://api.ai.sakura.ad.jp/v1/chat/completions
Authentication	Authorization: Bearer <API token> (token issued in the control panel)
Default model in Kurahi	sakura/gpt-oss-120b
Other models in the preset	llm-jp-3.1-8x13b-instruct; Kimi-K2.6 (public preview)
Free tier	3,000 chat requests per month on gpt-oss-120b; resets monthly; beyond that the free plan is rate limited and does not switch to paid billing by itself
Pay-as-you-go (gpt-oss-120b)	JPY 0.15 per 10k input tokens, JPY 0.75 per 10k output tokens (tax included)
Official pages	https://ai.sakura.ad.jp/sakura-ai/ai-engine/ · API reference https://manual.sakura.ad.jp/api/cloud/portal/ai-engine-inference-api/index.html

The model list in the control panel is authoritative; the preset in Kurahi carries the models above and you can add others under models.providers.sakura.models in kolo.json.

Register and get a token

1. Create a Sakura Internet member ID (phone verification required).
2. Create a Sakura Cloud project (a card is registered to the account; the free tier itself is not charged).
3. Open the AI Engine control panel at https://secure.sakura.ad.jp/ai/ and issue an API token.

Configure Kurahi

Interactive, on first run:

kurahi init          # choose "Sakura AI Engine", paste the token

Without prompts:

SAKURA_API_KEY=<token> kurahi init --non-interactive --provider sakura

Existing installation:

kurahi onboard --auth-choice sakura-api-key
# or
kurahi secrets set SAKURA_API_KEY --stdin
kurahi config set agents.defaults.model.primary sakura/gpt-oss-120b

The token is stored in the auth profile sakura:default under KOLO_HOME (file mode 0600), never in kolo.json. The environment variable SAKURA_API_KEY is honoured as an alternative.

Verify:

kurahi models list         # sakura/gpt-oss-120b listed as available
kurahi tui --message "Say hello in one line"

Notes for evaluators

• Requests count against the free tier per request, not per token; a long conversation is still one request per turn.
• The service is in Japan (Sakura Internet data centres); check the provider’s terms for the current data-handling statement before sending regulated data.
• To fall back to another provider when the monthly quota is used up, set agents.defaults.model.fallbacks (see Models).