Sahai is a sophisticated, production-grade mobile application that redefines the concept of an AI assistant. Built on a foundation of React Native and Expo, Sahai integrates state-of-the-art Large Language Models (LLMs) with advanced local processing to provide a private, fast, and multimodal experience.
Whether you're in a high-stakes meeting, analyzing complex documents, or needing instant visual context from your camera, Sahai acts as your invisible co-pilot—always listening, always ready.
Transform how you participate in meetings. Sahai doesn't just record; it understands.
- Real-time Question Detection: Uses advanced NLP to detect questions directed at you or the group.
- Autonomous Answering: Generates context-aware answers to detected questions in the background.
- Live Diarization: (Planned) Persistent voice enrollment for personalized speaker filtering.
Sahai combines the power of cloud AI with the privacy of local models. Cloud models are not eliminated; instead, local models are included alongside cloud models to provide secure, on-device processing where needed.
- Cloud Models: Leverages Groq and Gemini for high-performance reasoning and processing.
- Local Reasoning: Powered by Qwen 2.5 for fast, secure, and private text generation.
- Local Vision: Uses Phi (3.8B) for private, on-device visual analysis.
- Local Transcription: Employs OpenAI Whisper for accurate, offline audio transcription.
Your documents, now searchable and conversational.
- Massive Context Support: Upload and index up to 50 documents (PDFs, TXT, Code files).
- On-Device Vector-ish Engine: Uses a custom BM25 scoring algorithm for lightning-fast, offline-first context retrieval.
- Privacy First: Your data is indexed and queried locally before being sent as relevant context to the AI.
See the world through the eyes of AI.
- AI Camera: Real-time analysis of visual data using Llama Vision APIs.
- Video Intelligence: Extracts key frames for temporal analysis of events or demonstrations.
- All-in-One Chat: A unified interface for Text, Audio, Video, and Image inputs.
- Smart Caching: Sophisticated memory management for document chunks and image assets.
- Offline-First Resilience: Core search and indexing functions work without an internet connection.
- Haptic UI: Deeply integrated haptic feedback for a premium native feel.
| Layer | Technologies |
|---|---|
| Frontend | React Native, Expo Router, Reanimated, Expo Blur, Expo Haptics |
| AI Models | Groq, Gemini, Qwen 2.5 (Local Reasoning), Phi 3.8B (Local Vision) |
| Services | OpenAI Whisper (Local Speech-to-Text), Custom Question Detection |
| Data Engine | Local File System (Expo FS), BM25 Retrieval, Persistent JSON Storage |
| Language | TypeScript (Strict Mode) |
Sahai is designed with a service-oriented architecture to ensure modularity and speed:
- RagContextStore: Handles the chunking, indexing, and persistent storage of user documents.
- QuestionDetector: A dedicated service that monitors transcription streams for specific intent.
- ChatSessionStore: Manages complex multi-modal conversation histories and context injection.
- Vision Engine: Optimized frame-extraction pipeline for video-to-AI analysis.
- Node.js (v18+)
- npm or bun
- Expo Go app on your mobile device or an Emulator
-
Clone the repository:
git clone https://github.com/laxmanfhaneendra/Sahai.git cd sahai -
Install dependencies:
npm install
-
Configure Environment Variables: Create a
.envfile in the root directory and add your API keys:EXPO_PUBLIC_GROQ_API_KEY=your_groq_key_here EXPO_PUBLIC_GEMINI_API_KEY=your_gemini_key_here
-
Start the development server:
npx expo start
Document manager |
Live meet mode |
Chat interface |
Additional Feature 1 |
Additional Feature 2 |
We welcome contributions! Please follow these steps:
- Fork the Project
- Create your Feature Branch (
git checkout -b feature/AmazingFeature) - Commit your Changes (
git commit -m 'Add some AmazingFeature') - Push to the Branch (
git push origin feature/AmazingFeature) - Open a Pull Request