Terms

The service

Sigilline scores one message against rules you supply. One call returns a verdict: allow, flag, or block. It also returns the winning rule, a confidence score, and a score for each rule.

The service uses a purpose-built decision model, not a general chat model. It does not write a reply.

You choose the rules. Sigilline does not ship a fixed list of what to block.

The agreement

These terms and the Acceptable Use Policy are the agreement. The policy controls what you may send. These terms control the rest.

You accept them when you create a workspace. We record the version you accepted. If you use the API, you accept them for that use.

Who we are

Sigilline is the name of the service. "We" means the operator of the service. The legal name and mailing address will be added before this version leaves draft.

Accounts

Sign-in is an email code. The code confirms the address. It is not a password.

A new account creates one organization and one project after you name the workspace and accept these terms. A returning account goes to the dashboard.

You are responsible for the people who can sign in to the workspace.

Keys

A key is a secret your server sends with each call. We show a new secret once. Store it then. We keep a hash and a prefix, not the secret.

Test keys run the same model and do not count toward your allowance. Live keys count. Call the API from your server, not from a browser.

You can revoke a key. Calls that use it start failing right away. You are responsible for a key that leaks.

Your content

You keep the rights in the messages, rules, and examples you send. You grant us a license to process them only to run the service: to score the call, to return the verdict, and, if you later turn history on, to store what those settings say.

You confirm you have the right to send that text, including text from the people on your platform.

What a verdict means

A verdict applies your rules. Among rules at or above their threshold, the strictest action wins: allow, then flag, then block. If none qualify, the verdict is allow.

An allow verdict means no rule you sent reached its threshold. It does not mean the message is safe, lawful, or fit to publish.

Confidence is the winning rule's score, or a score derived when nothing matched. It is not a probability and it is not legal advice. English is the supported language. A message in another language still receives a verdict, and confidence on that response is capped.

You decide what your product does with the verdict. We do not ban, mute, or message your users.

What we keep

By default we do not store the message, earlier turns, the rule text sent on that call, the examples sent on that call, the sender role, or the channel.

For every call we do keep the verdict, the winning rule key, the language code (the one you send, or "en" when you send none), latency, the rule count, the model version, the key prefix, and the request id. That record has no message text. Usage and billing are calculated from it.

When you send an Idempotency-Key, the response can be replayed for 24 hours. The stored response does not include the message.

Sign-in email contains only the code. Logs hold a status, a request id, and latency. They do not hold the request body.

One workspace cannot read another's rows.

History

Message history is off. A paid plan may include the option. It cannot be turned on until you choose how long messages are kept and the plan allows it. Nothing in these terms turns it on.

If history is turned on later, the stored text is encrypted before it is written, kept only for the period you chose, and deletable by you. The number of days is not set in this draft.

Deleting the account removes stored message text, keys, and memberships. Usage counts may remain as billing records, with no message text. A contact request may remain so we can reply.

Billing

The free plan includes 1,000 decisions a month. Paid plans are 5,000, 10,000, and 100,000 decisions a month. The price is the price shown at checkout and on the pricing page. Volume above that is arranged through the contact form.

A decision is one live call that returns a verdict. Test keys, a replay of a stored idempotent response, and calls we refuse or cannot score do not count. Unused decisions expire at the end of the period. They do not roll forward.

Past the allowance, the next live call is refused and is not billed. You manage a paid plan in the billing portal. Cancelling stops the next renewal. The current period runs to its end unless we suspend the workspace.

Availability

The service is built to return a decision in under a second, round trip. That describes the product. It is not a promise of uptime, and it does not create a service credit.

Suspension

You will follow the Acceptable Use Policy. We may refuse a call, revoke a key, or close a workspace that breaks these terms or that policy, if a bill fails, or if the law requires it. We will tell you unless the law or the risk says not to.

Warranties

The service is provided as is. We do not warrant that a verdict is accurate, complete, or fit for a particular use. We do not warrant that the service will catch every message you hoped to catch.

Liability

We are not liable for indirect or consequential loss, including lost profits, lost users, or a verdict you acted on.

Our total liability arising out of the service in any 12-month period is limited to the fees you paid us for the service in that period. If you paid nothing, that limit is zero until a floor is set. This draft sets no dollar floor.

These limits do not apply where the law forbids them, including liability for fraud.

You will cover claims a third party brings against us from the text you send, the rules you set, or the way you enforce a verdict, except to the extent the claim is caused by our fraud.

Law and changes

You may not assign the workspace without our consent, except to a successor of the whole business. We may assign these terms with the service.

The governing law and the courts for a dispute are not chosen in this draft. They will be named before this version leaves draft.

We change these terms by posting a new version on this page and changing the version string. The version you accepted stays recorded on the workspace. A material change for a paid plan is posted before it applies. The version string stays draft until the open items at the top of this page are filled in.