Nabha AI Features Addendum
This Addendum forms part of the Terms of Service and applies whenever your organisation uses an AI-assisted feature of Nabha. It is also the disclosure of AI sub-processing referred to in the Data Processing Addendum and the Privacy Policy. Capitalised terms have the meaning given in the Terms.
1. What the AI features do
1.1 AI onboarding. When you connect a data source, Nabha reads sample payloads that a user in your organisation submits and suggests how to interpret them: field meaning, units, scale and offset, data types, timestamp format and a parsing recipe. Suggestions are applied to your workspace only after a user in your organisation reviews and confirms them.
1.2 AI assistant. A chat assistant that answers questions about your telemetry and configuration, reading the tags and time ranges the question concerns.
1.3 ML jobs and diagnostics. Anomaly detection and predictive models run inside Nabha against your time-series store; the ML job itself sends nothing to an AI provider. When you ask for a diagnosis, the AI assistant reads the job's output alongside the raw telemetry, using the model you have picked.
2. Your duty to verify
2.1 AI-generated output can be wrong, including in ways that matter: a wrong unit, a ×10 scale, a flipped sign, a mis-identified tag, or an anomaly that is not one. A competent person in your organisation must review every suggestion, diagnosis or prediction before relying on it, and must re-check when devices, firmware or payload formats change.
2.2 Nabha is a monitoring and analytics tool, not a safety system (Terms, Section 2.2). AI output must not be the sole basis for any safety-related, process-control or financial decision.
2.3 We give no warranty as to the accuracy, completeness or fitness of AI-generated output, and Section 10 of the Terms (Liability) applies to it.
3. What is sent, and to whom
3.1 When an AI-assisted feature runs, we send to the provider your organisation has selected only what that feature needs: the sample payloads, telemetry excerpts, ML job outputs or questions involved, together with schema-level context (tag names, data types, units, value ranges and any description you have written). Account data, credentials, billing data and other organisations' data are never sent.
3.2 The providers you can choose from, with their legal names and locations, are listed on the sub-processors page: Mistral AI (France), DeepSeek (China), Anthropic (United States) and OpenAI (United States). Each provider handles a request under its own API terms, which govern how long it keeps the request and whether it may use the content; we rely on those terms and have no separate agreement with any provider. If you have a data-residency or confidentiality constraint, choose a provider accordingly, or do not use the AI features.
3.3 Your controls. Your organisation chooses the provider in the app and can change it at any time; only the selected provider receives data. AI-assisted features run only when a user in your organisation uses them. If nobody in your organisation uses them, nothing is sent to any provider. Manual configuration is always available.
3.4 We may add, replace or remove a provider or model on 30 days' notice on the sub-processors page. If you object on reasonable data-protection grounds, you may stop using the AI features or terminate under Section 4.2 of the Data Processing Addendum.
4. How we handle it
We use the data sent to an AI-assisted feature only to produce your result. We do not use your data to train models, ours or anyone else's. We keep a record of each request (time, feature, model and metered usage) for billing, security and audit, and may keep request content for a limited period to cache repeated requests and to troubleshoot, after which it is deleted under the retention schedule in the Privacy Policy.
5. Usage, availability and labelling
AI usage is metered and included in your plan's allowance as shown on the pricing page; extra usage can be purchased. We may throttle or pause AI-assisted features during abuse or a provider outage; manual configuration remains available. AI-assisted features are not covered by the Service Level & Support Policy. AI-generated content is labelled as such in the product.
6. Ownership
As between you and us, you own your inputs, your outputs and the confirmed configuration produced from suggestions. We own the platform, its prompts, models and improvements (Terms, Section 8). Aggregated, de-identified statistics about AI feature quality may be used to improve the Service (Terms, Section 6.3).