voicebot_flutter 0.4.0

Drop-in voice and text store assistant for Flutter apps. The same VoiceBot assistant that runs as the web widget, native in your app over WebSocket + PCM16.

0.4.0

Optional, themeable UI kit — exported from a separate library so the headless core stays UI-free.

• New package:voicebot_flutter/ui.dart barrel (separate from voicebot_flutter.dart): import it only if you want the prebuilt widgets; headless apps never pull in Material UI from the core.
• VoicebotLauncher — turnkey all-in-one: floats a VoiceButton over every screen via a root OverlayEntry, manages the ChatPanel (open/minimize/close), and owns the client/session lifecycle. Drop it once near the app root. Wires minimize ≠ close (HANDOFF §4): minimize collapses to the launcher and keeps the session alive (no end); close (X) calls endSession() + tears down. Bot/server end auto-collapses and resets for the next open. Programmatic open() / minimizePanel() / close() / toggle() via a GlobalKey<VoicebotLauncherState>.
• VoiceButton — floating circular launcher (FAB) reflecting the live phase from the status + audioState streams: idle (mic) · connecting (spinner) · listening (animated pulse ring) · speaking (accent). Themeable size/position/colors/elevation/icon. Exported standalone for custom layouts.
• ChatPanel — the assistant window: header (title + minimize + close), scrollable transcript (user vs bot bubbles, live streaming partials with the partial-replace rule), footer (multiline text input + send + mic toggle/mute). Animated slide-up + fade; respects safe area + keyboard insets. Exported standalone.
• VoicebotTheme — sensible defaults that blend with the ambient Material ColorScheme (override any field), copyWith, and a VoicebotThemeProvider inherited widget. VoicebotLabels holds every user-facing string (localizable — nothing hardcoded in the widgets). The kit is fully replaceable: every widget binds only to the public VoicebotSession streams.
• Example rewritten: a mock storefront with the floating VoicebotLauncher as the primary demo (tool handlers + deep links + live context wired via onClientReady/onSessionStarted); headless usage still available via the core API.
• Pure Dart/Flutter (material/widgets only) — no new deps, no native code; the android/iOS compile gates are untouched. dart analyze --fatal-infos clean; widget tests added for VoiceButton / ChatPanel / VoicebotLauncher (minimize-keeps-alive vs close-sends-end).
• Cross-SDK parity: this UI model (VoicebotLauncher / VoiceButton / ChatPanel / VoicebotTheme) is the shared spec the RN SDK mirrors so docs describe one model.

0.3.0

Native audio device-readiness (VOIC-46) — the voice path is completed for real devices.

• iOS: setPreferredIOBufferDuration(0.02) to hold capture→WS-frame latency within the ≤40 ms AC; the interruption handler now honors .shouldResume and restarts the engine/tap after a call/Siri; route changes rebuild the AVAudioConverter against the new input format (Bluetooth/wired headset) so capture doesn't silently garble.
• Android: OnAudioFocusChangeListener pauses the mic uplink + playback on incoming call / focus loss and auto-resumes on gain — gated by a separate focusPaused flag so a user-initiated mute is preserved; AudioDeviceCallback re-asserts MODE_IN_COMMUNICATION on BT/headset connect/disconnect.
• Self-VAD / speakerphone: corrected the false "avoided by WS choice" note — the mitigation is hardware AEC (already enabled, .voiceChat / MODE_IN_COMMUNICATION + AcousticEchoCanceler); documented as a device-dependent known limitation.
• CI: added android-build (flutter build apk) and ios-build (flutter build ios --no-codesign) jobs — the native modules now compile on every push (the Android module had never been compiled before).
• Synced the iOS podspec version to the package version.
• Open: physical-device test matrix (iOS 16+ / Android 24+) is the only remaining VOIC-46 AC — the native code compiles in CI, but a hardware run is required to sign off latency / AEC / interruption behavior.

0.2.1

CI/CD + pub.dev OIDC automated publishing; pubspec metadata (homepage/repository) + api base -> api.monoverse.tech.

Store compliance (VOIC-52 / T9).

• Added docs/STORE_SUBMISSION.md — copy-paste App Store + Play Store checklist for merchants embedding the SDK (mic string, privacy manifest, Apple 5.1.2 third-party-AI disclosure, export/encryption, RECORD_AUDIO, Play Data safety, GenAI policy, foreground-only note).
• ios/Resources/PrivacyInfo.xcprivacy now declares NSPrivacyCollectedDataTypeAudioData (App Functionality, not linked, not tracking). Native code audited: no required-reason APIs used (flutter_secure_storage uses the Keychain, which is not a required-reason API and ships its own manifest), so NSPrivacyAccessedAPITypes stays empty. README + example mic string aligned to the third-party-AI disclosure.

0.2.0

Auth model corrected to web-parity + Mode B + context enrichment (VOIC-44 / VOIC-49).

• Auth (breaking): MobileIssueTokenProvider (Mode A) now signs with the EXACT PROTOCOL §1 canonical preimage HMAC_SHA256(secret, "{METHOD}\n{path}\n{ts}\n{nonce}\n{sha256hex(body)}").hex() and sends X-VoiceBot-{Tenant-Id,Timestamp,Nonce,Signature} + X-VoiceBot-App-Id (timestamp = epoch seconds). The default signer is now the real HMAC-SHA256 (crypto added as a direct dep); still injectable for hardware-key delegation. AppCredential reshaped to {tenantId, appId, sharedSecret} (mirror the web key). JsonPoster.body is now List<int> (raw signed bytes).
• Auth — Mode B: new PublishableKeyTokenProvider calling POST /api/v1/mobile/issue-token-public with {api_key:"vb_pk_…", app_id, attestation?, lang?}.
• Context enrichment (no backend change): session.updateContext(AppContext) → client_context frame and session.sendCartEvent(CartEvent) → cart_event frame (PROTOCOL §10). New zero-PII AppContext (screen/route/product/category/locale/custom) and CartEvent / CartLineItem types. Retained context auto-re-sends on resume.
• New exported HMAC helpers: buildSignaturePreimage, computeSignature, defaultHmacSigner, IssueTokenSigner.
• README rewritten into a full SDK landing page; tests expanded to 55.

0.1.0

First milestone (VOIC-43 / VOIC-44 / VOIC-46 / VOIC-48): headless core + text path

• native-audio scaffolding.

• Headless core: VoicebotClient.init, VoicebotSession sealed state machine (connecting → ready → streaming → reconnecting → ended/failed).
• WsTransport ported from the web widget ws-client.ts: exponential backoff, ~25s heartbeat ping, fatal-close-code set {4400, 4401, 4403, 4503, 1008, 1011}, reconnect capped at 3 attempts (VOIC-44 AC).
• Protocol codec for all docs/PROTOCOL.md envelopes (user_text, action_ack, binary uplink; ready / session_state / transcript / turn_complete / barge_in / tool_call / action / error inbound).
• ToolHandlerRegistry with registerToolHandler + registerDeepLinkMap; unregistered tool auto-replies {success: false, error: "no_handler"}.
• AudioSource engine-agnostic PCM16 in/out interface (VOIC-44 boundary) plus a native PlatformAudioSource backed by AVAudioEngine (iOS) / AudioRecord (Android).
• Lifecycle: minimize() (UI-only, keeps WS) vs endSession() (sends end, tears down).
• Stream-based events: transcripts, audio state, tool calls, actions, status, errors.
• Native scaffolding: iOS Swift (AVAudioEngine + AVAudioConverter + .voiceChat AEC, 24 kHz scheduled-buffer playback, interruption/route observers) and Android Kotlin (AudioRecord VOICE_COMMUNICATION + AcousticEchoCanceler, AudioTrack 24 kHz, audio-focus/device callbacks). Device-tuning TODOs marked inline.