| Field | Policy | Detail |
|---|---|---|
| Prompt retention | Not retained by default Prompts (text inputs) are processed transiently in memory to serve a request and are not written to durable storage. During active debugging or incident investigation, limited content may be captured temporarily under access controls and deleted promptly afterwards (within 7 days). | Prompts (text inputs) are processed transiently in memory to serve a request and are not written to durable storage. During active debugging or incident investigation, limited content may be captured temporarily under access controls and deleted promptly afterwards (within 7 days). |
| Completion retention | Not retained by default Generated completions are returned to the caller and are not persisted for analytics, evaluation, or training, except for limited, temporary capture during active debugging (deleted within 7 days). | Generated completions are returned to the caller and are not persisted for analytics, evaluation, or training, except for limited, temporary capture during active debugging (deleted within 7 days). |
| Training usage | Never Customer prompts and completions are never used to train, fine-tune, re-train, or evaluate models, and are not added to any training corpus, benchmark, or dataset. | Customer prompts and completions are never used to train, fine-tune, re-train, or evaluate models, and are not added to any training corpus, benchmark, or dataset. |
| Primary inference region | India Inference is served from GPU-backed infrastructure located in India. Certain supporting services (such as DNS, CDN, TLS termination, and monitoring) may process limited request metadata in other regions. | Inference is served from GPU-backed infrastructure located in India. Certain supporting services (such as DNS, CDN, TLS termination, and monitoring) may process limited request metadata in other regions. |
| Operational logs | Up to 30 days Minimised request metadata (timestamps, model id, token counts, status codes, latency) is retained for up to 30 days for reliability and abuse prevention (unless a longer period is required by law), then deleted or aggregated into non-identifying statistics. | Minimised request metadata (timestamps, model id, token counts, status codes, latency) is retained for up to 30 days for reliability and abuse prevention (unless a longer period is required by law), then deleted or aggregated into non-identifying statistics. |
| Data deletion | On request You can request deletion of account-associated data at any time, subject to legal retention obligations. | You can request deletion of account-associated data at any time, subject to legal retention obligations. |
| Sale of data | Never We do not sell, rent, or trade personal or customer data to any third party. | We do not sell, rent, or trade personal or customer data to any third party. |
Scope
This Data Policy gives an explicit, field-by-field account of how AndiIndia handles data processed through the inference API. It is intended to give partners, marketplaces, and customers verifiable answers, and it complements our Privacy Policy and Terms of Service. Where there is a conflict on a data-handling specific, this page governs.
Definitions
- Prompt / input: the text content and parameters you send to a model.
- Completion / output: the content a model generates in response to a prompt.
- Operational metadata: non-content information about a request, such as timestamps, model id, token counts, status codes, and latency.
Training and model improvement
We do not use customer prompts or completions to train, fine-tune, re-train, or evaluate any model, and we do not share such content with third parties for those purposes. Model improvements come from publicly available or properly licensed datasets, never from customer traffic.
Retention
By default, prompt and completion content is not persisted after a response is returned, and we do not maintain content logs of inference traffic. The only exception is limited, temporary capture during active debugging or incident investigation, which is access-controlled and deleted promptly afterwards (within 7 days).
Operational metadata (described above) is retained for up to 30 days for reliability, billing reconciliation, and abuse prevention, unless a longer period is required by law, after which it is deleted or aggregated into non-identifying statistics.
Regional processing and residency
Inference is served from infrastructure located in India, which supports data-residency needs for customers who require in-region model processing. Certain supporting services that are necessary to operate and protect the platform - such as DNS, content delivery (CDN), TLS termination, and monitoring - may process limited request metadata outside India. Prompt and completion content is processed by the inference infrastructure in India.
Subprocessors
We use a minimal set of infrastructure subprocessors (such as cloud GPU and hosting providers) strictly necessary to operate the Services. They are bound by confidentiality and security obligations and may not use customer data for their own purposes.
Security
We apply encryption in transit, access controls, the principle of least privilege, and monitoring appropriate to the sensitivity of the data. Access to systems is restricted to authorised personnel on a need-to-know basis.
Abuse prevention
To keep the platform safe and reliable, we may process operational metadata and apply automated safeguards to detect and prevent abuse, fraud, and security threats. These safeguards operate on metadata and signals rather than on retained content.
Requests and contact
For data deletion, residency, or processing questions, contact support@andiindia.in. We respond to verified requests within a reasonable period. See also our Privacy Policy and Terms of Service.