Nothing matches. Email [email protected] — we answer Monday–Saturday, 09:00–18:00 Singapore time.
Getting started
How do I get an API key?
Kurrens is in pre-launch and API access is by invitation. Request access and we’ll issue a key. Keys are shown once, when created — store them in an environment variable such as KURRENS_API_KEY.
Is Kurrens compatible with the OpenAI API?
Yes. Point any OpenAI SDK at https://api.kurrens.ai/v1, use your Kurrens key, and set model to a Kurrens model id. Chat Completions, streaming, tool calling, structured outputs, and vision work as documented. See the Quickstart.
Can I use Kurrens with Cursor, LangChain, LiteLLM, or Dify?
Yes — anything that lets you set an OpenAI-compatible base URL works. See Integrations. Tools that only speak the Anthropic Messages API aren’t supported.
Is Kurrens on OpenRouter?
Our catalog is published in the OpenRouter provider format at kurrens.ai/models.json. We’ll announce here when Kurrens is live as an OpenRouter provider.
Models
Which models do you serve?
Open-weight models from DeepSeek, Moonshot AI (Kimi), Z.ai (GLM), Qwen, MiniMax, Xiaomi, OpenAI (gpt-oss), and Google (Gemma). The full, searchable list is in the Model catalog.
What precision are models served at? Do you quantize them further?
Each model’s precision (for example FP8) is listed in the catalog and in models.json. We never lower the precision behind an existing model id. See Precision & versioning.
Will the model behind an id change?
No. A model id always points to the same weights at the same precision. Updated releases get a new id, and old ids are retired with at least 30 days’ notice.
Can you add a model?
Request it. We add models based on demand and license terms. For a model only you need, a dedicated endpoint is usually faster.
Do you support image or video input?
Yes, on models that support it — for example Kimi K3, MiniMax M3, GLM 5.3 Flash, and Qwen3.8. Filter the Model catalog by “Image / video input” and see Vision.
Data & privacy
Do you store my prompts or completions?
No. Prompts and completions are processed in memory and discarded when the response completes. We keep request metadata — token counts, latency, status code — for billing. See Zero data retention.
Do you train on my data?
No, never — not our own models, not anyone else’s. This is written into our Data Policy, which takes precedence over our other policies.
Where are my requests processed?
In Johor (Malaysia) by default, with Jakarta (Indonesia) and Dublin (Ireland) as failover regions. Dedicated endpoints can be pinned to one region. See Regions & data residency.
Can I sign a DPA?
Our Data Processing Addendum applies automatically when you accept the Terms. For a countersigned copy, email [email protected].
Limits & reliability
What are the rate limits?
Limits are per API key, in requests and tokens per minute, and depend on your account and the model. Contact us for higher limits. See Rate limits.
Why did I get a 429 instead of a slow response?
We fail fast instead of queueing. When no region has capacity, you get 429 immediately so your client can retry or fall back without waiting. The OpenAI SDKs retry automatically.
Why does a reasoning model’s stream go quiet for a while?
The model is thinking. We send SSE comment lines (: keep-alive) during long reasoning so connections don’t time out — your client should ignore them. See Streaming.
Billing
How am I billed?
Per token, at the input, cached-input, and output rates listed for each model on kurrens.ai/pricing. During pre-launch, prices are estimates and may change.
Are reasoning tokens billed?
Yes, at the output rate. They are reported in usage.completion_tokens_details.reasoning_tokens.
Am I charged if I cancel a streaming request?
Only for tokens generated before the connection closed.
Do you offer invoicing?
Yes, for enterprise customers on monthly terms. Talk to sales.
Support
How do I contact support?
Email [email protected] and include the request id from the response headers. We reply Monday–Saturday, 09:00–18:00 Singapore time (UTC+8).
How do I report a security issue?
Email [email protected] with the subject line “Security report”. See Security.