Real-time fact-checking app that listens to conversations, transcribes speech, and instantly verifies factual claims with AI-powered web search.

• Live Transcription - Deepgram Flux streaming with end-of-turn detection
• AI Fact-Checking - Perplexity, Groq, or Gemini (configurable)
• iOS Live Activity - Dynamic Island + Lock Screen updates
• Deep Dive - Tap any claim for detailed research with sources
• Dark/Light Theme - System-aware theming

Tap mic → Audio streams to Deepgram → Transcription on pause →
LLM extracts claims → Fact-checks with web search → Live Activity updates

app/                    # Expo Router pages
components/             # UI components (MicButton, ClaimCard, etc)
services/               # API integrations
  ├── deepgramFlux.ts   # Real-time transcription
  ├── perplexity.ts     # Fact-checking (default)
  ├── groq.ts           # Fast fact-checking (no search)
  ├── gemini.ts         # Fact-checking + deep dive
  └── liveActivity.ts   # iOS native bridge
hooks/                  # React hooks
  ├── useFactChecker.ts # Main orchestration
  └── useFluxTranscription.ts
native-src/             # Swift code for Live Activities
config.ts               # API keys & provider selection

1. Install dependencies

   npm install

2. Configure API keys in config.ts:

   export const DEEPGRAM_API_KEY = 'your-key';
   export const PERPLEXITY_API_KEY = 'your-key';
   export const GROQ_API_KEY = 'your-key';      // optional
   export const GEMINI_API_KEY = 'your-key';    // optional

3. Run development build

   npx expo prebuild
   npx expo run:ios
   # or
   npx expo run:android

Switch LLM providers in config.ts:

export const LLM_PROVIDER: 'perplexity' | 'groq' | 'gemini' = 'perplexity';

Requires iOS 16.2+. Shows:

• Dynamic Island - Mic icon + claim count (compact), full claim (expanded)
• Lock Screen - Current claim with timer

The app auto-stops when Live Activity is dismissed.

holdthatthought://start-listening
holdthatthought://stop-listening
holdthatthought://toggle-listening

All LLM services return:

interface FactCheck {
  id: string;
  claim: string;           // Short topic label
  verdict: Verdict;        // 'true' | 'false' | 'partial' | 'unverified'
  confidence: number;
  explanation: string;     // Brief correction or answer
  sources: Source[];
  timestamp: number;
}

Typical latency (end-to-end from speech to result):

• Perplexity: 2.2-3s (with web search)
• Groq: 200-500ms (no web search)
• Gemini: 2-6s (with web search)

MIT