Thanks for your feedback.

Some key definitions in this article include:

• Data Processing: The act of performing computational operations (such as transcription, inference, or synthesis) on customer data. No customer data is stored during processing; data exists only temporarily in volatile memory (RAM) and is purged immediately after processing is completed.
• Data Residency and Storage: Refers to where customer data is persistently stored (“at rest”). This includes configuration, tenant-specific settings, and stored logs.
• Media Residency: Media (such as audio streams) remain in their originating location for as long as possible and are not persistently moved or stored outside the region, even if processing occurs in another geography.
• Sub-processor: A third-party service provider engaged to perform specific functions (e.g., LLM inference, speech recognition) on customer data, under strict contractual and technical controls.

For customers operating in regulated regions such as Singapore, we distinguish between where data is stored (Residency) and where data is computed (Processing).

A fundamental aspect of our AI architecture is the commitment to ephemeral processing and zero data retention. Our approach prioritizes data privacy and security by ensuring that customer information is never stored or retained during AI operations.

• ASR & TTS Visibility: During real-time transcription (ASR) or speech synthesis (TTS), data exists only in the volatile memory (RAM) of the processing engine and is purged immediately after use. No data is stored or retained after processing.
• Access Control: No human employees (internal or sub-processor) have access to raw audio or text streams during the inference phase.
• No Secondary Usage: We maintain strict contractual and technical barriers to ensure customer data—including prompts and audio—is never used to train, retrain, or improve foundation models owned by sub-processors.

We conduct rigorous third-party risk assessments on all sub-processors.

• Third-party Risk Assessments: All sub-processors undergo rigorous assessment.
• Encryption: Data sent to sub-processors is encrypted in transit, processed only in RAM, and never stored.
• Auditability: All API calls to sub-processors are logged for audit purposes (metadata only; sensitive payloads are excluded from logs).
• Incident Accountability: In the event of a data incident at a sub-processor, [Your Company] maintains primary accountability and handles all customer notifications and remediation per our standard Data Processing Addendum (DPA).

To ensure transparency, the following table outlines exactly how long data is held within our ecosystem:

Our AI Assistant uses a Retrieval-Augmented Generation (RAG) architecture. This ensures the AI provides responses based on your specific "Ground Truth" documents rather than internal model training.

• URL/Document Updates: When a knowledge source is updated, the system re-indexes the content and replaces previous versions.
• Latency: Updated content typically reflects in the AI’s responses within [X] minutes.
• Cache Handling: Active sessions use the context available at the start of the session; however, all subsequent sessions are forced to query the most recent index, preventing the reuse of stale or "hallucinated" data.