Privacy Policy
Scope
This policy covers OneTriangle's website, routing services, and inference service,
including our OpenAI-compatible API, managed model infrastructure, and KV-cache-aware
inference features. An order form, data processing agreement, or enterprise agreement
may set additional terms for a customer deployment.
Information you provide
We collect contact, company, billing, and account information when you contact us,
book a demo, create an account, or purchase the service. We also process the model,
routing, and deployment settings you configure.
Inference data
Requests to the OneTriangle API may include prompts, messages, files, model settings,
and other inputs you submit. Because we provide inference, we must process those
inputs to generate and return model outputs. Together, these inputs and outputs are
“Customer Content.” You are responsible for deciding what Customer Content to send to
the service.
KV-cache-aware processing
For an eligible model pair, the service may prefill an input on a smaller source model,
create key-value cache tensors from that input, translate selected cache state into the
target model's attention space, and continue decoding on the target model. These cache
states are temporary technical representations derived from Customer Content and are
handled as Customer Content. If a transfer is unavailable or does not clear applicable
compatibility, quality, or latency gates, the service may use ordinary target-model
prefill instead.
Usage and service data
We collect operational metadata such as token counts, request timing, latency, model
and model-pair selection, cache-transfer and fallback status, errors, and estimated or
billed cost. We use this data to operate, secure, meter, troubleshoot, and improve the
service. We do not sell personal data, and we do not use Customer Content or derived
cache states to train AI models or fit cache-translation maps unless you expressly
agree in writing.
Infrastructure and sharing
Depending on your deployment, inference may run on OneTriangle-managed infrastructure,
dedicated infrastructure, or approved model, cloud, and GPU providers. We share data
with those providers and with vendors that support hosting, monitoring, communications,
and billing only as needed to provide the service. We may also disclose information
when required by law or to protect the service, our customers, or others.
Retention and security
We retain account, billing, and usage records, including the operational metadata
described above, for as long as needed to provide the service, meet legal obligations,
and resolve disputes. Prompt and response content and derived cache state are stored
for approximately 24 hours to complete, secure, and troubleshoot inference, then deleted,
unless a written agreement provides otherwise. Retention periods are not currently
user-configurable. We use technical and organizational safeguards, including
encryption in transit and at rest where applicable. No method of storage or
transmission is completely secure.
| Data | Retention |
|---|
| Prompts and model outputs | Approximately 24 hours, then deleted |
| Derived KV-cache state | Approximately 24 hours, then deleted |
| Operational metadata (token counts, timing, model selection, errors, billed cost) | Retained to operate, secure, meter, and troubleshoot the service |
| Account and billing records | As long as needed for the service and legal obligations |
Your choices and rights
You may request access to, correction of, or deletion of your personal data. Depending
on your location, you may have additional privacy rights. To make a request or ask
about how we handle inference data, email hannah@onetriangle.ai.
Changes
If we change this policy, we will post the updated version here and revise the date above.
Terms of Service
Agreement
By using OneTriangle's website, API, managed inference, routing, or related services,
you agree to these terms. If you use the service for an
organization, you represent that you have authority to bind it. A signed order form or
other written agreement controls if it conflicts with these online terms.
The service
OneTriangle provides model routing and OpenAI-compatible inference services for
managed and open-weight models. Where supported, the service may compute a
prompt on a source model, translate its KV cache, and decode on a compatible target
model. OneTriangle may select models and infrastructure according to your configuration and
may use ordinary prefill when cache transfer is unavailable or does not clear its gates.
Customer Content
You keep ownership of Customer Content. You grant OneTriangle the limited rights needed to
process Customer Content, create temporary cache state, route requests, return
outputs, and otherwise provide and secure the service. You
represent that you have the rights and permissions needed to submit Customer Content for
that processing.
Your responsibilities
You are responsible for account credentials, API keys, Customer Content, deployment
settings, and your use of model outputs. You
must not use the service to violate law or third-party rights, introduce malicious code,
evade usage limits, interfere with the service, probe another customer's environment,
or facilitate abuse. You remain responsible for reviewing outputs before relying on them
in sensitive or high-impact uses.
Models, cache transfer, and outputs
Model outputs are probabilistic and may be inaccurate, incomplete, or unsuitable for
your use. A quality or compatibility gate is an operational control, not a guarantee
that transferred-cache output will be identical to full target-model prefill or correct
for every request. Model availability, eligible source-target pairs, fallback behavior,
performance, and cost effects may vary by workload, configuration, and infrastructure.
Third-party services
The service may rely on third-party models, cloud infrastructure, GPU providers, and
software. Their availability and behavior may affect the service. Where your
configuration routes a request to a third-party provider, that provider may process the
request subject to its applicable terms and any data-processing commitments in your
written agreement with OneTriangle.
Preview and research features
Features identified as research, alpha, beta, preview, or not production-ready are
offered for testing and may change, fail, or be discontinued. Do not use those
features for production or high-impact decisions unless OneTriangle has approved that use
in writing.
Fees
Paid service is billed as stated in your order form or plan. Usage charges may depend on
input and output tokens, selected models, infrastructure, or other metered resources.
Model-provider and dedicated-infrastructure charges may be separate. Fees are
non-refundable except as required by law or stated in your written agreement.
Intellectual property
OneTriangle and its licensors own the service, cache-translation technology, software,
documentation, and related intellectual property. Except for the limited rights needed
to provide the service, neither party receives ownership of the other party's property.
Feedback may be used to improve the service without restriction or payment.
Confidentiality
Each party will protect the other's non-public information using reasonable care and
use it only to perform or receive the service. This obligation does not cover
information that is public through no breach, already known without restriction,
independently developed, or lawfully received from another source.
Disclaimers and liability
The service is provided “as is” and “as available.” To the maximum extent permitted by
law, OneTriangle disclaims implied warranties and is not liable for indirect, incidental,
special, consequential, or exemplary damages. OneTriangle's total liability arising from
the service is limited to the fees you paid OneTriangle for the service in the twelve months
before the event giving rise to the claim, unless a written agreement says otherwise.
Suspension and termination
You may stop using the service at any time. We may suspend or terminate access for a
material breach, security risk, unlawful use, non-payment, or conduct that could harm
the service or others. Upon termination, payment obligations and provisions that by
their nature should survive will remain in effect.
Changes and contact
We may update these terms by posting a revised version and date. Material changes apply
prospectively. Questions about these terms or deployment-specific terms may be sent to
hannah@onetriangle.ai.
