FAQ

Does backend config require app restart?

Not always. SDK periodically refreshes remote config via GET /v1/sdk/config.

Default refresh interval: 60 seconds (production), 15 seconds (debug)
AI.getConfig() always returns latest merged snapshot
App restart only needed if app is idle and no refresh loop running

To force immediate refresh, restart the app.

Where should non-secret config live?

Use backend YAML (backend/config/kycis.yaml) for product/runtime settings.

Reserve environment variables and secret stores for:

API keys
LiveKit credentials
Database passwords
Signing secrets

How do I verify what the agent currently knows?

Inspect backend context:

GET /v1/assistant/context/{session_id}

Response includes:

Current screen
latest_event payload
last_validation_failure
Component hints from component_input
KYC knowledge base rules

Also check /api/activity for recent events.

Can I disable voice and use text-only?

Yes. Voice is optional:

// Don't set voice listener
// Don't call startAssistant()

// SDK still provides:
// - Event tracking
// - Trigger evaluation
// - Validation telemetry

Backend handles voice dispatch via LiveKit. Skip if not needed.

What's the difference between `error_reported` and `validation_failed`?

Event	SDK Method	Sets `last_validation_failure`	Use For
`error_reported`	`AI.trackError()`	No	Network, UX, generic errors
`validation_failed`	`AI.trackValidationFailure()`	Yes	Field-level validation failures

Validation failures populate session state for trigger and context.

Why is `session_id` sometimes "unknown"?

In current Android SDK implementation, AI.setUser() may send session_id: "unknown" before a real session exists.

This event may not associate with the real SDK session in backend store until session is established.

How do I mask/redact PII in hints?

Always set masked = true and redact values:

AI.reportComponentInput(
  componentId = "phone_field",
  hint = "98****7654",  // Redacted
  masked = true
)

Backend feature integration.assistant_echo_component_values controls whether agent can echo hints.

Can I have multiple assistant prompts per session?

Yes, with feature flag:

if (AI.getConfig().isFeatureEnabled("allow_multiple_assistant_prompt")) {
  // Multiple prompts allowed with cooldown
}

Default: one prompt per app session (with cooldown).

What's the difference between `screen` and `screen_id`?

Both identify current screen:

screen — semantic name (e.g., "pan_entry")
screen_id — technical identifier (e.g., "com.app.PanActivity")

Use consistently. Backend uses either for context building.

How do I test trigger behavior?

Enable backend flags: trigger.v2_scoring, trigger.policy_engine
Set passiveEvalEnabled = true with short interval
Create conditions:
- Stay on screen 20+ seconds → high_time_spent
- Cause 2+ validation errors → repeated_errors
Check /api/activity for kind: "trigger"

Why does backend restart clear all data?

Session store is in-memory only by default:

# backend/app/services/store.py
class InMemorySessionStore:
  # Data lost on restart

For persistence, enable SQLite:

SESSION_STORE_BACKEND=sqlite
SESSION_STORE_SQLITE_PATH=.kycis-session-store.sqlite3

Can I use my own UI for confirmation dialog?

Yes, control trigger behavior:

RuntimePolicy(
  triggerStartMode = TriggerStartMode.COOLDOWN,
  // Or IMMEDIATE for direct start
)

// Then implement your own UI
fun onTriggerDetected() {
  showCustomDialog()
}

Skip CONFIRM_UI mode for custom UI.

FAQ

On this page