For the complete documentation index, see llms.txt. This page is also available as Markdown.

Platform and security

Platforms

Every SDK feature, including the food scanner and voice capture, runs on iOS 15 and later (requirements). The SDK doesn't support macOS, Mac Catalyst, tvOS, watchOS, or visionOS.

Credentials

  • Keep your API key on your backend. The only exception is local development in a Debug build.

  • The SDK keeps client tokens in memory. Don't write them to disk or logs.

  • Configure your token endpoint's URL explicitly; don't ship a fallback URL.

  • Authenticate the token request with the app's existing session. The token endpoint rules cover the backend side.

  • Create a new client when a different account signs in (Client lifecycle).

User and health data

Meal photos, food logs, glucose profiles, CGM readings, and predictions can be sensitive. Collect and keep as little as you can, get the permissions your product needs, and keep these values out of analytics and crash reports. Use opaque end-user IDs, never email addresses or names.

Camera privacy

The food scanner needs NSCameraUsageDescription. Write a purpose string that matches the feature, and don't ask for camera access before the user starts scanning.

Microphone and speech privacy

Voice capture needs both NSMicrophoneUsageDescription and NSSpeechRecognitionUsageDescription. VoiceCaptureSession asks for access only when you call startRecording(). Recorded audio stays in a temporary local file only while Apple Speech transcribes it; then the SDK deletes the file. The SDK doesn't send audio or transcripts to January.

Last updated

Was this helpful?