Security

SDKs

Krisp does not access, collect, or store any audio data. The Krisp SDK is expressly designed to operate solely on the SDK customer’s premises, with all processing—including any optional generation of metadata such as voice and noise durations, noise levels, and voice energy metrics—performed exclusively on the customer-side. The SDK accesses the network only for license verification needs. Metadata generation is only activated upon the customer's explicit instruction, and any metadata produced is not stored or transmitted. Consequently, neither audio data nor any derived metadata is made visible to or stored by Krisp, ensuring that the content of customer and end-user communications remains entirely private.

API

The Krisp API works differently from the SDKs: audio (Speech-to-Text, Voice Translation, Voice Isolation) or text (Text-to-Speech) is processed in Krisp's cloud. Your application sends it to wss://streaming.krisp.ai over a TLS-encrypted WebSocket connection, authenticated with a short-lived session token (see Authentication).

To compare the two models, see SDK or API: choosing an integration.