1. Service provider
The name, address, country, and direct support channels of the provider operating this NULIQ deployment are published in the business card below and on the contact page. Missing details are clearly marked; new payments are disabled until they are complete.
2. Information processed
- Account: name, email, securely hashed password, account creation date, and records of acceptance of the terms.
- Access: session and API-key hashes, key names, creation and last-use information. Technical request information is also processed for security and rate limiting.
- Usage: query time, input byte count, model version, 0/1 result, cost, balance, and transaction identifiers. Raw API and Playground query text is not saved in the application’s usage table.
- Training: uploaded labeled TXT content is processed as a temporary file during training. Raw training files are removed when the job ends. Model weights, training reports, and text fingerprints used to check training/test overlap are retained.
- Verified feedback: when you explicitly save a correct answer, the selected question text and its 0/1 label are retained privately for your account, up to 500 examples and the training size limit. They are used only when you request a new training version. They are not merged into the shared model. You can clear saved feedback in the Playground; this does not delete existing model weights or cancel already queued training jobs. A private keyed input fingerprint is stored with usage to verify that feedback belongs to your request.
- Quality testing: test content is processed for evaluation. Aggregate metrics, test-set fingerprints, and evaluation reports are stored.
- Payments: prepaid balance, amount, checkout and provider event identifiers, payment status, and transaction records. NULIQ does not store card numbers or security codes.
This describes the application’s data processing. Hosting and network providers may separately process connection and security logs. NULIQ does not guarantee anonymity for uploaded data or zero retention across every system.
3. Purpose and sharing
Data is processed to operate your account and API access, train and evaluate your models, calculate usage, resolve errors, and prevent abuse. This application does not use your account data for advertising targeting.
NULIQ Decision Core: queries are processed by NULIQ’s own model on our server. Query text is not sent to an external AI inference provider. Email and phone requests are processed by NULIQ’s trained neural contact classifiers. They estimate format plausibility and do not query mailboxes or phone networks.
Private models: TXT training, saved feedback and quality tests are processed on the NULIQ server. They create or evaluate separate private models. They do not automatically retrain the shared Decision Core. When you enable automatic learning, explicitly submitted verified feedback is used to train and test a private adapter. We retain the request text, label and evaluation split for this feature, up to 500 examples and 256 KiB per account. An adapter is activated only after passing its evaluation gate. Disabling the feature returns inference to the shared Core; resetting it deletes its saved examples and adapter. Separately saved TXT-training feedback and TXT models are managed separately.
The hosting provider processes data to run the service. When payments are enabled, Stripe processes amounts, account/transaction identifiers, and information you provide during payment. Stripe’s own privacy policy also applies. Information may need to be shared with authorities in response to mandatory legal requests.
Shared NULIQ chat contributions (optional)
Ordinary chat requests are not saved for shared training. If you rate a chat answer and separately check the shared-training consent box, we save the conversation context sent with that answer, the answer, your rating, any correction, model version, consent version and account association. These contributions may be reviewed and used to improve a shared model serving all users. The raw contributions are not publicly published. Do not contribute personal, confidential or third-party private information.
Consented contributions pass automatic checks for duplicates, conflicting labels and suspicious or sensitive text. Pending examples need matching contributions from at least three accounts aged three days, or operator approval. With enough eligible data, a CPU training job builds a candidate. Only candidates passing separate accuracy, regression and speed tests are automatically activated. Ratings do not immediately change model weights; contributions may be excluded. Temporary training exports are deleted after each run. Limits are 500 contributions per account and 100,000 overall. Use Withdraw contribution beneath a submitted answer to remove its source record while it remains on screen, or contact support later. Withdrawal does not undo completed training or automatically erase previously created training exports or backups. Private decision feedback and TXT training remain separate.
4. Retention and deletion
Account records, model files, and related reports are retained to provide the service. Deleting an inactive custom model from the console removes its weights, training fingerprints, and related evaluation reports. Usage and payment records are not automatically deleted with the model.
Send account and personal data requests through the support channels. Account deletion is currently handled through support. Requests are assessed after identity verification and review of applicable retention obligations. Accounting, security, dispute, and backup records are not guaranteed to be deleted at the same time. Ask the service provider for its specific retention periods and hosting region.
5. Cookies and security
An essential session cookie is used for login. This version does not install advertising or third-party analytics cookies within the application. Stripe’s checkout page may use the provider’s own cookies and security checks.
Passwords and API keys are stored as hashes in the database. Internet access must use HTTPS. No system can guarantee absolute security; do not upload unnecessary personal or sensitive data.
6. Your requests
Use the contact channels to request information, correction, deletion, or other actions relating to data processing. Applicable rights and obligations depend on the laws governing the provider and user. This notice makes no claim of certification or official compliance approval.
Open console