Shre Keyboard

AI-powered keyboard for iOS and Android with voice-to-text and full Shre AI platform integration.

Features

• Voice-to-Text: On-device STT (Apple Speech / Google SpeechRecognizer) with cloud Whisper fallback via shre-router
• Shre AI Agent Access: Talk to any Shre agent from any app — the keyboard routes intent through shre-router
• Smart Compose: Context-aware text generation, replies, translations
• Workspace-Aware: Knows your stores, agents, and data through shre-context
• Conversation Memory: Persistent sessions via Working Memory (Tier 2)
• Cost Tracking: Per-request billing through shre-meter

Architecture

shre-keyboard (device)
├── Voice capture (on-device mic)
├── Local STT (Apple Speech / Google SpeechRecognizer)
├── Shre Client SDK (shared Kotlin Multiplatform)
│   ├── Auth (shre-auth, API key / device token)
│   ├── Chat (shre-router /v1/chat — agent routing)
│   ├── Audio (shre-router /v1/audio/* — Whisper STT)
│   └── Context (shre-context — workspace injection)
└── Platform keyboard extension (IME / UIInputViewController)

Project Structure

shre-keyboard/
├── android/          # Android IME (Input Method Service)
│   └── app/
├── shared/           # Kotlin Multiplatform — Shre client SDK
│   └── src/
│       ├── commonMain/   # Shared API client, models, auth
│       ├── androidMain/  # Android-specific (OkHttp, speech)
│       └── iosMain/      # iOS-specific (URLSession, Speech)
├── ios/              # iOS Keyboard Extension (Swift)
│   └── ShreKeyboard/
└── docs/             # Architecture docs

Getting Started

Android

cd android
./gradlew installDebug

Then enable "Shre Keyboard" in Settings > System > Languages & Input > On-screen keyboard.

iOS

# Install xcodegen (one-time)
brew install xcodegen

# Generate Xcode project
cd ios
xcodegen generate

# Open in Xcode
open ShreKeyboard.xcodeproj

Build to a real device (keyboard extensions don't work in Simulator), then:

1. Settings > General > Keyboard > Keyboards > Add New Keyboard > Shre Keyboard
2. Tap Shre Keyboard > Allow Full Access (required for mic + network)

Voice Engine (3-Tier STT)

Tier 1: On-device (Apple Speech / Google)  — 0ms, free, instant partials
Tier 2: NVIDIA Riva (Shadow PC, GPU)       — ~50ms via Tailscale
Tier 3: Whisper via shre-router (cloud)    — ~300ms fallback

Tier	Provider	Latency	Languages	Cost
1	Apple Speech / Google STT	0ms	50+	Free
2	NVIDIA Riva (RTX 4070 Ti)	~50ms	11 native	Free (self-hosted)
3	faster-whisper / OpenAI	~300ms	99+	Metered

TTS cascade: Riva TTS > PersonaPlex > ElevenLabs > OpenAI TTS

Shre Platform Integration

Feature	Endpoint	Service
Voice transcription	POST /v1/audio/transcriptions	shre-router (cascade)
Voice session (WS)	POST /v1/voice/session	shre-router (Riva/PersonaPlex/OpenAI)
TTS synthesis	POST /v1/audio/speech	shre-router (cascade)
AI chat / agents	POST /v1/chat	shre-router
Voice engine status	GET /v1/voice/status	shre-router
Authentication	POST /v1/auth/device	shre-auth
User context	GET /v1/context	shre-context
Usage metering	automatic	shre-meter

Multilingual Support (50+ languages)

Speak in any supported language. Three output modes:

• = Same: Insert in the language you spoke
• -> EN: Auto-translate to English before inserting
• -> Target: Translate to any chosen language

Regions: English (3), Americas/Europe (20), South Asia (12), East/SE Asia (9), Middle East/Africa (8)

Configuration

Copy .env.example to .env and set:

• SHRE_API_URL: shre-router URL (default https://127.0.0.1:5497)
• SHRE_API_KEY: Generated from MIB007 dashboard

Or configure in the app's settings screen (shared via App Group on iOS, SharedPreferences on Android).

License

Proprietary — Nirlab Inc.