Acceptable use policy
Umbra sells API access to language models, for fiction, roleplay and creative writing. This policy covers the narrow set of uses that will get an account closed.
We do not read your prompts. Enforcement is driven by reports, by upstream providers telling us about a problem, and by usage patterns — never by us reading the content of your requests. See the privacy policy.
Never, under any circumstances
These end an account immediately and without warning.
- Sexual content involving minors. Any depiction, in any framing, fictional or otherwise, including content describing characters as minors or as childlike in a sexual context. There is no roleplay exception, no "aged-up" exception, and no ambiguity here.
- Content sexualising real people without consent, and any sexual content depicting a real identifiable person.
- Material that facilitates serious real-world harm — instructions for weapons capable of mass casualties, or for synthesising controlled substances or explosives.
- Content produced to harass, threaten, stalk or defame a real person, including generating messages to send to someone who has asked you to stop.
- Child sexual abuse material of any kind, which we report to the appropriate authorities.
Not allowed
- Reselling. Your key is for your own use. Do not resell access, share a key across a group, or operate a public bot or proxy on it. If you want to build something for other people, ask us.
- Evading limits. Multiple accounts to multiply trial credit or rate limits, or to get around a suspension.
- Deliberate abuse of the service. Automated retry loops you have not bounded, load testing without asking, or anything designed to exhaust shared capacity.
- Spam, phishing, malware, or generating content to deceive people about who they are talking to for commercial or political gain.
- Scraping model outputs to train a competing model, where the upstream provider's terms prohibit it.
- Presenting model output as human where doing so is illegal in your jurisdiction, or as professional medical, legal or financial advice.
Explicitly fine
Stated because the list above is long and the reason people are here can get lost in it:
- Adult fiction and erotic roleplay between adult characters, including dark themes, violence, and morally objectionable characters.
- Fiction involving real historical figures, in a non-sexual context.
- Long-running character roleplay, character cards, lorebooks, group chats — the whole SillyTavern workflow. That is the product.
- Personal automation, scripting, and building things on the API for yourself.
Individual model terms still apply
Each model in the catalogue is operated by someone else, under their own usage policy. A request can be refused upstream even when this policy permits it, and when that happens you are not charged for it beyond the tokens actually used.
How we enforce this
For most breaches we will contact you first and give you a chance to stop. For anything in the first list, or for repeated evasion, we close the account immediately. Where an account is closed for a breach, unused paid time is not refunded — see the refund policy.
We will comply with lawful requests from authorities. What we can hand over is limited to what we hold, which is described in the privacy policy and does not include the content of your requests, because we never store it.
Reporting
If you believe someone is using Umbra for anything in the first list, tell us at umbradevelopers1@gmail.com. We act on those first.