- Home
- Managed API
One API key. Compute inside Japan.
An OpenAI-compatible managed endpoint. We run the models, the scaling and the hardware; you pay for the tokens you use.
OpenAI-compatible
Point your existing SDK or tooling at a different base URL. Chat completions, embeddings and streaming are supported.
Per token
Billed on input and output tokens, with no minimum commitment. Monthly invoicing in JPY, with usage visible in the dashboard.
Japanese region
Inference runs in a Nachster AI region inside Japan. Requests are not forwarded outside the region.
What we serve.
Models are offered by capability class. The specific models and the provenance of their weights are set out in a service catalogue provided under NDA during qualification. The catalogue changes as generations turn over, and deprecations are announced in advance.
- Text generation — General instruction following and writing, in Japanese and English.
- Reasoning — For tasks that need step-by-step working.
- Long context — Contracts, specifications, transcripts and other long documents.
- Code — Generation, completion and review.
- Embeddings & rerank — The base layer for search and RAG.
- Bring your own weights — We host and run your model — that sits in the private endpoints tier.
We do not publish individual model names. The service catalogue is provided under NDA after qualification.
How your data is handled.
We do not train on your data
Prompts sent to the API and the completions returned are never used to train or improve models. This is written into the agreement.
Zero retention by default
Request content is held only as long as it takes to produce the response. Operational logs are metadata only — timestamp, token counts, latency, origin. Where you need retention for audit, it can be enabled for a defined period.
MaaS is a shared service and therefore includes automated abuse filtering. Private endpoints and dedicated clusters carry no content inspection — see Security & isolation.
General availability follows the first region going live, in Q2 2027. Before that, early access is allocated individually to design partners. Talk to us.
Tell us what you need. We reply within 48 hours.
Qualified enquiries receive proposed times for a technical meeting within 48 hours, and an allocation offer after that meeting. Information is handled under NDA.