Acceptable use policy

The Saanora Platform gives you API access to Saanora models, starting with Clara. You can use them to build almost anything: routing and triage, agent tool choice, moderation queues, review workflows, internal tools and products of your own.

This policy sets out the few things you can't do, and the extra care we ask for when Clara's answers affect people. It applies alongside the platform terms and covers every request made with your organization's API keys, including requests your own users make through your product.

The person who signs up is responsible for their organization. If you invite members, make sure they know these rules too.

Don't use the platform or its models for any of the following:

• Anything illegal, or helping someone else break the law.
• Fraud, scams or deception, such as phishing, impersonating someone, or misleading people about who they are dealing with.
• Harassing, threatening, stalking or intimidating anyone, or discriminating against people where the law forbids it.
• Violating other people's privacy: sending records about people you have no right to process, building secret profiles of individuals, or monitoring people without a lawful reason.
• Creating, spreading or targeting malware, spam, or attacks on other people's systems.
• Abusing the service itself. The rules for that are in Protecting the platform below.

If the records you send contain personal data, you are responsible for having the right to use it that way and for telling people how you use it. We process records in memory to compute answers and don't store them. Data handling and the privacy policy explain what we keep.

Clara returns probabilities over options you define. It is good at this, and it can still be wrong. Some decisions matter too much to hand to a model alone.

Clara must not be the sole basis for a decision that has legal or similarly significant effects on a person unless a human meaningfully reviews it first. That includes decisions about:

• Credit, lending and other financial products.
• Employment: hiring, promotion, pay, discipline or dismissal.
• Housing.
• Insurance.
• Healthcare.
• Education, including admissions and grading.
• Law enforcement.
• Access to essential services, such as utilities, benefits or public services.

Meaningful review means a person who has the authority and the information to change the outcome actually looks at the case. Rubber-stamping the model's answer is not review.

Before Clara acts on anything automatically, validate it on your own labeled data: run it on a few hundred of your real records, compare its answers with the right ones, and pick a threshold for each question. Read the known limitations and how to use the probabilities. Where answers must stay stable, for example when they feed thresholds or audits, we recommend consistent mode. It reads the record once for each question, so an answer does not change with the other questions in the request. We don't keep your records or answers, so keep your own records of any decision you may need to audit.

You are responsible for following the laws on automated decisions that apply where you and your users are, including any duty to tell people a model was involved or to let them ask for a human.

To keep the service fast, fair and safe for every customer:

• Keep your API keys secret. Don't publish them, commit them to public repositories, or ship them in browser or mobile code. Anyone with a key can spend your credit. If a key leaks, revoke it in the console and create a new one. Authentication has the details.
• Don't circumvent rate limits or spend limits, for example by spreading traffic across extra organizations or accounts to get around your tier. If you need more capacity, the rate limits page explains how tiers rise.
• Don't run load tests, penetration tests, vulnerability scans or other security testing against the platform without our written permission. If you think you have found a security issue, tell us at support@saanora.com.
• Don't try to extract, distill or reverse engineer the model. That includes using its outputs to train or tune another model, systematically probing it to reconstruct its behavior, and trying to recover its weights, architecture or training data. The model is closed and its weights are never released. Testing Clara on your own data to measure how well it works for you is fine and encouraged.
• Don't interfere with or disrupt the service, other customers' requests, or the systems that run it, and don't try to access data that belongs to another organization.

Building products on the API is what it is for. You can put Clara behind a feature in your app, a workflow you sell, or a service you run for clients, and charge for what you build.

Reselling raw API access is not allowed. That means you can't sell, rent or give your keys to anyone outside your organization, or offer a pass-through service whose main purpose is to let others call Saanora models with their own records and questions.

When your users reach Clara through your product, you are responsible for their use of it under this policy.

If you see the platform or its models being misused, write to support@saanora.com. Tell us what you saw, when it happened, and any request id, organization or product involved. Include the request id (the x-request-id header on API responses) if you have one; it helps us find the request.

Use the same address to report a security issue or a leaked key that isn't yours.

If we find a use that breaks this policy, we will warn you first where it is practical and lawful, and where we can, give you a chance to fix it.

We may suspend or end access for a key, a project, an organization or an account that breaks this policy. Where a use is serious, illegal, or puts other people or the platform at risk, we may act first and contact you afterwards where we can.

If you think we got it wrong, write to support@saanora.com and we will look at it again.