Initialization

Initialization has two independent steps, deliberately separated:

1. open() — wires platform key storage, biometrics, and attestation into the Rust core and returns your SenseCryptAuthenticator handle. Cheap.
2. initializeInferenceEngine() — parses the license and loads the face detection, matching, and liveness models. Costs 200–500 ms of model-load wall time.

They are separate because the authentication flows that don't touch a camera — key rotation, revocation, listing keys — need only step 1. Run step 2 during your splash screen so the cost is paid before the user taps anything.

Opening the SDK

import SenseCryptAuthenticatorSDK

let authn = try SenseCryptAuthn.open(
    config: AuthenticatorConfig(
        portalUrl: "https://your-tenant.sensecrypt.com",
        appLabel: "Acme",
        tenantId: nil,
        enableLogging: false
    )
)

import ai.seventhsense.sensecrypt.authenticator.SenseCryptAuthenticatorSDK
import ai.seventhsense.sensecrypt.authenticator_ffi.AuthenticatorConfig

val authn = SenseCryptAuthenticatorSDK.open(
    context = context.applicationContext,
    config = AuthenticatorConfig(
        portalUrl = "https://your-tenant.sensecrypt.com",
        appLabel = "Acme",
        tenantId = null,
        enableLogging = BuildConfig.DEBUG,
    ),
    playIntegrityCloudProjectNumber = BuildConfig.PLAY_INTEGRITY_CLOUD_PROJECT_NUMBER,
)

AuthenticatorConfig

enableLogging defaults to off so a host app that never sets it — which is every release build — emits nothing to logcat or the Xcode console. Pass your debug flag (BuildConfig.DEBUG, #if DEBUG). Panic reporting stays on regardless of this setting.

Failures from open()

open() throws AuthnInitError:

Loading the inference engine

do {
    try await authn.initializeInferenceEngine()
} catch let error as SenseCryptInitError {
    // See the table below.
}

try {
    authn.initializeInferenceEngine(context)
} catch (e: SenseCryptInitException) {
    // See the table below.
}

This call stages the SDK's bundled model assets and your mobile.lic into app-private storage, then constructs the capture engine — parsing the license and loading the detection, recognition, and liveness pipelines.

It is idempotent: a second call after success is a no-op. Construction runs off the main thread, so it will not block your UI dispatcher during the 200–500 ms model load. On a first run the asset staging step adds noticeably more — expect several seconds on a slow device or an emulator, since the models are copied out of the package before anything loads.

Failures from initializeInferenceEngine()

Throws SenseCryptInitError (Swift) / SenseCryptInitException (Kotlin):

Reading device state

Once open, the handle exposes what this device already holds — useful for deciding whether to show enrollment or sign-in:

let keys: [Key] = authn.keys
let current: Key? = authn.currentKey
let branding: Branding? = authn.brandingFor(keyId: someKeyId)
let entries: [AuditEntry] = authn.auditLog(limit: 50)

val keys: List<Key> = authn.keys
val current: Key? = authn.currentKey
val branding: Branding? = authn.brandingFor(keyId)
val entries: List<AuditEntry> = authn.auditLog(limit = 50)

keys is filtered to keys that are actually usable: records whose local hardware key was wiped are dropped, and keys surviving a reinstall are cleared by open() as a clean-slate measure. An empty list means this device has no enrollment and you should start a registration flow.

See Reference for the fields on Key, Branding, and AuditEntry.

Related

• Install — getting the artifacts and license in place.
• Login — the first flow most apps implement.
• Errors — the complete error surface.