1. Who we are, and what this covers
tokn.supply ("tokn", "we") is an AI inference service operated as a sole proprietorship
based in New York, United States. These terms govern your use of the tokn website, API,
and account pages (the "Service"). By creating an account or using an API key, you agree
to them. If you are using the Service for an organization, you represent that you can
bind that organization.
2. The Service, plainly
tokn provides access to open-weight AI models over an OpenAI-compatible API, metered per
token and paid from a prepaid credit balance. The Service is currently a private beta:
access is by invitation, the catalogue is small, and payments are processed by Stripe and
charge the card you provide, including automatic top-ups you switch on. We may change
the models offered, their prices, and the features of the Service; price changes never
apply retroactively to tokens already served.
3. Your content, and what we may do with it
"Input" is what you send to a model; "Output" is what the model returns. You own your
Input and, as between you and us, you own the Output. You grant us only the license
required to provide the Service: to transmit, process, and meter your Input and Output
for the purpose of generating and delivering your Output to you, and for nothing else.
Zero retention of content. We do not store, log, review, sell, or
disclose the content of your Input or Output, and we do not use it — ever — to train,
fine-tune, or improve any model or service. Content exists in memory on the machine
serving your request for the duration of that request only. This is not only a promise:
the Service is built so that the components between you and the serving machine cannot
read content at all, and we check before every release that no content reaches any log
or database. The technology page describes the mechanism and
its limits precisely, including the one place content necessarily exists in memory — the
machine serving your request.
What we do record is operational metadata: request identifiers,
model name, token counts, timings, request state, error codes, and cryptographic hashes
used to verify billing. Metadata never contains content.
4. Accounts and API keys
Your API key is your credential. Keep it secret; anything done with it is attributed to
your account, and usage it incurs is yours. We store only a keyed hash of the key and
cannot recover it — a lost key is replaced, not retrieved. One account per person or
organization. We may revoke keys used in breach of these terms.
5. Acceptable use
Do not use the Service to break the law, to infringe others' rights, to generate child
sexual abuse material, to conduct or facilitate attacks on other systems, or to probe,
overload, or circumvent the Service's own limits and security. We cannot read your
content, so enforcement acts on what we can see: traffic patterns, metadata, and
reports. We may suspend or terminate accounts for violations.
Models from other publishers. The models we serve are published by
others under their own licences, and your use of a model through the Service is subject
to its licence as well as to these terms. Llama 3.2 is licensed under the
Llama 3.2 Community License, Copyright © Meta Platforms, Inc.
All Rights Reserved; your use of Llama models must also comply with Meta's
Acceptable Use Policy. BAAI bge-m3 is published under the MIT
License.
6. Credits and payment
The Service is prepaid. Credits are purchased through Stripe's hosted checkout — your
card details never reach us — and are drawn down per token at the prices published at
the time of each request. A request refused for capacity or failed before
producing output is not charged. Credits are not transferable and bear no interest.
On account termination, unused purchased credits are refundable on request at their
purchase price; promotional or starter credits are not.
7. Output, disclaimed
Models generate text statistically. Output may be wrong, incomplete, offensive, or
similar to output generated for others, and is not professional advice of any kind.
You are responsible for evaluating Output before relying on it or publishing it.
8. No warranty, no SLA
The Service is provided "as is" and "as available", without warranties of any kind,
express or implied. This is a private beta: capacity is deliberately small, and a full
fleet refuses new work rather than queuing it. We publish no service-level agreement
and offer no uptime commitment.
9. Liability
To the maximum extent permitted by law, our total liability arising out of or relating
to the Service is limited to the greater of one hundred US dollars ($100) or the
amounts you paid us in the six (6) months before the event giving rise to the claim,
and we are not liable for indirect, incidental, special, consequential, or exemplary
damages, or for lost profits, data, or goodwill.
10. Indemnification
You will defend and hold us harmless from claims arising out of your Input, your use of
Output, or your breach of these terms.
11. Termination
You may stop using the Service at any time and may request account deletion at
hello@tokn.supply. We may suspend or terminate
the Service or your access to it, with notice where practicable. Sections 3, 7, 9, 10,
and 12 survive termination.
12. Governing law; changes; contact
These terms are governed by the laws of the State of New York, USA, excluding its
conflict-of-law rules, and disputes belong to the state or federal courts located in
New York. We may update these terms; material changes will be dated at the top of this
page, and continued use after a change is acceptance. Questions:
hello@tokn.supply.