# Terms of Service Effective 2026-09-20 · Operator: Zaitgeist Innovation Group GmbH · Contact: info@zaitgeist.com This is a free, public, experimental API. It serves the open-weights **Laya** decision models (Apache-2.0, by Convai Innovations) behind an interface compatible with Cloudflare Workers AI's `typesafe/jev` model. **By sending a request you accept these terms. If you do not accept them, do not use the service.** ## 1. The deal, in one paragraph You may use this API free of charge, without an account. In exchange, **everything you submit is stored and may be used to train and evaluate machine-learning models.** If that is not acceptable for your data, do not send it here - the models are open weights and you can host them yourself. ## 2. What is stored, and what is not **Stored** for every accepted request that is not an exact repeat of an earlier one: the `state` and `questions` you sent, the answers returned, the checkpoint that answered, token counts, timings, a timestamp, and a random request id (returned to you in the `X-Request-Id` response header). **Also stored:** requests that are rejected *only* because they exceed the model's context window (`max_tokens_exceeded`), if the state is at most 256k tokens. They are kept without answers, because inputs that are too long for today's model are what a longer-context successor must be trained on. The error message says so when it happens. Requests rejected for any other reason (invalid input, unknown model, rate limits, oversize bodies) are not stored. **Cached** for up to 7 days, so that identical requests need not be recomputed: a one-way hash (SHA-256) of the request together with the numeric probabilities that were returned. The cache contains none of your text - not the state, not the questions, not the option labels. The `X-Laya-Cache` response header tells you whether an answer came from it. **Not stored:** your IP address, the `Authorization` header or any other header, cookies, or any account identifier. The API needs no credentials; if you send a bearer token out of habit it is ignored and discarded. The one exception is a token the operator has issued to a gateway that resells this endpoint: it is matched in memory against the issued ones to pick that gateway's rate-limit budget, and is no more written down than yours. Your IP address is held only in memory, as a salted hash, for rate limiting, and is gone when the process ends. **Where it is processed:** requests are received (TLS is terminated) and computed in the European Economic Area. The infrastructure is operated by Modal Labs, Inc., a company based in the United States, which may keep its own short-lived access logs under its own policies and does not document the physical location of its storage volumes - which is one reason stored records are encrypted before they are written, as described next. **How it is protected:** records are encrypted with public-key cryptography (libsodium sealed boxes) before they are written to disk. The server holds only the encryption key. The decryption key is kept offline by the operator, so the stored data cannot be read from the serving infrastructure, even if it were compromised. ## 3. Licence you grant For everything you submit, you grant the operator a worldwide, non-exclusive, royalty-free, perpetual and irrevocable licence to store, reproduce, annotate and use it to train, evaluate and improve machine-learning models, and to publish and license models trained on it (including as open weights). Your raw submissions will not be sold, and will not be published in identifiable form. You confirm that you have the right to grant this licence for everything you submit. ## 4. What you must not submit Because submissions are retained and used for training, **do not submit**: - personal data about yourself or anyone else (names, e-mail addresses, phone numbers, account or customer identifiers, health, financial or other sensitive information) - use synthetic or properly anonymised data instead; - secrets: passwords, API keys, tokens, private keys; - confidential or proprietary information, or anything you are contractually or legally barred from sharing; - unlawful content. ## 5. Acceptable use No abuse. In particular you agree not to: - exceed or circumvent the rate limits (currently 60 requests per minute and 5000 per day per client), including by rotating addresses; - attack, probe, overload or disrupt the service, or attempt to access data that is not yours; - submit deliberately poisoned or adversarial data intended to degrade models trained on it; - use the service for unlawful purposes; - resell the service or present it as your own, or as TypeSafe's or Cloudflare's. The operator may block any client, at any time, without notice. ## 6. Do not rely on the answers The models are experimental. Answers are probabilistic, can be wrong, and **can be wrong with high reported confidence**. Inputs longer than the models were trained on are accepted (see the `X-Laya-Beyond-Trained-Context` response header) but are less reliable. Do not use the output as the sole basis for decisions with legal, financial, medical, safety or similarly significant effects on people. Keep a human in the loop. ## 7. Deletion E-mail the contact above with the `X-Request-Id` values of the requests concerned and the matching records will be deleted from the stored data. Records cannot be located without their request id, because nothing else that identifies you is kept. Deletion cannot undo the influence of a record on a model that has already been trained. ## 8. No warranty, no liability The service is provided "as is" and "as available", without warranties of any kind, and without any commitment to availability, latency, accuracy or continuity. It may change, be rate-limited, or be shut down at any time. To the fullest extent permitted by law, the operator is not liable for any damages arising from your use of, or inability to use, the service. Nothing in these terms excludes liability that cannot be excluded by law. ## 9. Independence and trademarks This service is not affiliated with, endorsed by, or connected to TypeSafe AI, Cloudflare, or Convai Innovations. "Jev", "TypeSafe" and "Cloudflare" are trademarks of their respective owners and are used only to describe API compatibility. Requests naming the model `typesafe/jev` are answered by Laya, never by Jev, and the `model` field of every response says so. ## 10. Changes These terms may change. The current version is always at `https://zaitlabs--laya-jev.eu-west.modal.run/terms`; continued use after a change means acceptance of the new version.