Terms of service

These terms govern your use of the Dydema EEG API, this developer site and its dashboard (together, the service). By creating an API key or using the service you accept them. The dashboard asks you to confirm that you accept them before you can create a key.

You must be 13 or older to create an account.

You may send data only from people who are 18 or older. You must not submit EEG or any other data from anyone under 18.

By sending data you confirm that you have the rights, consents and approvals needed to send it to us and to have it processed as these terms describe.

You are responsible for having the consent and ethics approval your study requires before sending recordings. Remove names, dates of birth and other direct identifiers, including from session ids. Embeddings can identify individuals: in our tests people were re-identified across sessions far above chance. Treat them as personal data.

You grant Dydema a licence to process the data you send in order to provide the service to you.

Dydema may also use submitted data and data derived from it, such as embeddings, fitted classifier heads and usage records, to operate, secure, evaluate and improve the service and its models.

Labs that cannot upload data can license the model weights to run on their own premises: contact info@dydema.com.

Requests are processed in the United States, and usage records and stored answers are kept in the United States.

• The recording in a request is used to answer that request and is not written to storage.
• Each request leaves one log row (time, status, model, windows, cost, latency and the key used, never any signal), shown on your dashboard’s Logs page and deleted after 30 days.
• Usage and billing records hold counts and amounts, never signal, and are kept as the record your bill is built from.
• When you send an Idempotency-Key, the successful answer is stored under your account so a retry can be replayed: the key is honored for 24 hours, and the stored answer is deleted after 30 days.
• A deleted stored object keeps a backup version for 90 days after that.

Task models (hosted classifiers)

For a fitted task model we store the fitted classifier (its coefficients or classical pipeline), the class names, the training report, the record of how its model was chosen, and the channel names, sampling rate and units it was fitted on. We do not store your EEG signal. While a model created with defer_fit is collecting trials, we keep the features computed from them (never the signal) until it is fitted, and delete them then, or after 7 days if it is never fitted.

A task model is kept until its expiry date or until you delete it. If your account has insufficient credit when a month’s hosting falls due, that month is not charged and the model keeps working; if the account still has insufficient credit 30 days later, we delete the overdue models, after an email warning 7 days before. A task model created with a test key is deleted automatically after 30 days.

Deleted task models are removed from service immediately; backup copies are purged within 90 days.

Nothing served here is a medical device, and nothing here is validated for diagnosis or clinical decisions. A task model returns labels for the task you describe, and the models return features, never diagnoses: any classifier built on them, including a task model, is the builder’s, and so is its validation. Results with missing or substituted channels are not comparable.

Do not use the service to:

• break the law, or process data you have no right to process;
• make clinical or diagnostic decisions about a person;
• get around rate limits, spending limits, metering or billing, or reach another account’s data;
• probe, overload or disrupt the service, or test its security without our written permission;
• extract the models’ weights, or reconstruct the models from their outputs.

The restrictions below apply to every size, every task model and every output of the service. They also implement the prohibited-use restrictions of the third-party licence acknowledged below. You must not use the service, its models, task models or outputs:

• to intentionally extract, reconstruct, recover or infer subject-level EEG recordings, personal data or other information relating to an identifiable individual from a model, its parameters, representations or outputs; to perform model-inversion, training-data extraction, membership-inference or substantially similar attacks for the purpose of learning information about identifiable individuals or determining whether an identifiable individual or record was included in a model’s training data; to use embeddings, representations, outputs or parameters to re-identify an individual from data that has been anonymised, pseudonymised, de-identified or otherwise separated from their identity; or to assist anyone in doing so (aggregate scientific analysis that is not intended, and is not reasonably likely, to identify or recover information concerning particular individuals is not prohibited);
• to identify, authenticate or track a natural person on the basis of EEG, neural activity or representations derived from such data; to conduct covert or non-consensual surveillance of natural persons; to infer or profile a person’s cognitive, emotional, psychological, neurological, behavioral or other sensitive state without a valid research, medical, legal or other appropriate basis and, where applicable, the person’s informed consent; or to create or materially contribute to systems intended for population-scale surveillance or behavioral profiling of natural persons;
• to unlawfully discriminate against, or facilitate unfair or harmful differential treatment of, individuals or groups based on protected characteristics, health status, disability or sensitive characteristics inferred from neural data; to perform social scoring or comparable evaluation of persons that results in unjustified or disproportionate adverse treatment; to make or materially support decisions concerning employment, education, insurance, credit, housing, law enforcement or access to essential services where the use of neural information would be unlawful, discriminatory or inconsistent with applicable safeguards; or to exploit vulnerabilities associated with age, disability, health, economic circumstances or other conditions in a manner intended or reasonably likely to cause significant harm;
• to intentionally manipulate or coerce individuals through deceptive, subliminal or exploitative techniques in a manner that materially impairs their ability to make informed decisions and causes or is reasonably likely to cause significant harm; to intentionally facilitate physical or psychological harm to individuals; or to intentionally develop or deploy systems whose principal purpose is otherwise incompatible with the safety, dignity, autonomy or fundamental rights of natural persons;
• in violation of applicable law or regulation, including requirements concerning privacy and personal-data protection, human-subject research, medical or clinical use, discrimination, cybersecurity or artificial-intelligence systems.

Outputs may be used, shared or distributed only where doing so does not disclose personal data obtained through conduct prohibited above, and does not otherwise violate these terms or applicable law. Wizard also refuses, at the dry run and at creation, tasks that recognise emotions or affect and clinical or diagnostic tasks, whatever their basis.

Research and product use is otherwise allowed under these terms.

Dydema may use third-party models and software in operating this service, including REVE, under the REVE Responsible Use License v1.0. This is a service-wide notice and does not identify the technology used for any particular feature, model size, task model or request.

REVE was developed by the BRAIN (BRoader Artificial Intelligence) research team, Lab-STICC (CNRS UMR 6285), at IMT Atlantique. Copyright © 2026 Institut Mines-Télécom Atlantique Bretagne-Pays de la Loire (IMT Atlantique). IMT Atlantique, BRAIN, Lab-STICC and CNRS do not endorse, certify or sponsor Dydema or Wizard.

A key is shown once, at creation; we keep only a one-way hash of it, and it cannot be recovered. Keep your keys secret: requests made with your key are billed to your account. You may revoke any key at any time, and revocation takes effect on the next request. Tell us at security@dydema.com if you think your account or a key has been misused.

The service is prepaid. You buy credit on the dashboard’s Billing page, from $5 per purchase, and each request’s cost is taken off your balance at the prices on the pricing page. Requests may pause when you have insufficient credit. Auto-recharge is optional: when you turn it on, your saved card is charged the amount you chose whenever your balance falls below the level you set. Every response is metered, and those metering records are what your bill is assembled from. Payments are processed by Stripe.

Task models (hosted classifiers): creating one is charged once, at the create fee on the pricing page, which covers fitting it; dry runs and failed fits are not charged. Hosting is charged from your prepaid credit in whole 30-day months up front, at the monthly price on the pricing page, starting when the model is fitted and ending at its expiry or deletion; deleting a model stops the months after it. Predictions are billed per window at the price on the pricing page, whichever model or method it runs. Task models created with test keys are not charged.

The task models, predictions and other results the service returns are yours to use, subject to these terms. They are statistical estimates computed from the signal you sent. You are responsible for how you use them and for any conclusion or classifier you build on them.

Serving is decided by us, not by the payment processor. A failed payment suspends your account from the next request; your keys are kept and resume when the payment clears. We may also suspend or close an account that breaks these terms or puts the service or other people at risk. You may stop using the service at any time: revoke your keys, and email support@dydema.com to close your account.

The service is provided as is and as available, without warranties of any kind, express or implied, including fitness for a particular purpose, accuracy and uninterrupted availability, to the extent the law allows.

To the extent the law allows, Dydema is not liable for indirect, incidental, special or consequential damages, or for lost profits, data or research results, arising from your use of the service. Our total liability for any claim is limited to the amount you paid us for the service in the twelve months before the claim.

When these terms change we publish a new version here with a new version number and effective date, and the dashboard asks you to accept it before you can create keys or use your account there. Keys you already hold keep working in the meantime; continuing to use the service after a change means you accept the new version.

Questions about these terms: legal@dydema.com.