A powerful, privacy-first AI study companion optimized for Arm-based devices. Study faster, think deeper, learn privately.
Pinguin is an offline-first AI study companion built specifically for university students who value privacy and performance. Running entirely on-device with Arm-optimized AI models, Pinguin transforms your study materials into an intelligent knowledge base you can query naturally—no internet required, no data leaving your device.
Why Pinguin for Arm?
- Native Arm64 Performance: Built and optimized specifically for Arm architecture, leveraging efficient instruction sets for faster inference
- On-Device AI: All processing happens locally using Ollama's Arm-native builds—your data never leaves your device
- Energy Efficient: Arm's power efficiency means longer battery life during study sessions
- Windows on Arm: Optimized for Windows 11 on Arm devices (Snapdragon X Elite, Surface Pro X, etc.)
- Multi-Format Support: PDF, DOCX, EPUB, TXT, and more
- OCR for Scanned Documents: Extract text from images and scanned PDFs using Tesseract
- Smart Chunking: Advanced text segmentation preserves context and meaning
- Metadata Extraction: Automatic extraction of document structure and metadata
- Semantic Search: Vector-based retrieval finds relevant information across all your documents
- Context-Aware Responses: LLM generates answers grounded in your study materials
- Source Attribution: Every answer includes references to source documents
- Multiple Query Modes: Optimize for precision, recall, or balanced retrieval
- 100% Offline: No internet connection required after initial setup
- Local AI Models: Ollama runs LLMs and embeddings entirely on your device
- Fast Inference: Arm-optimized models deliver quick responses
- Secure Storage: ChromaDB vector store keeps your data local and encrypted
- Course Organization: Group documents by courses and subjects
- Chat History: Review past conversations and insights
- Batch Processing: Upload multiple documents at once
- Clean Interface: Distraction-free UI built with Material-UI
Pinguin leverages a modern, efficient tech stack optimized for Arm devices:
Frontend
- Electron (Arm64 native builds)
- React 18 with TypeScript
- Material-UI for responsive design
Backend
- Python FastAPI server
- LangChain for RAG orchestration
- ChromaDB vector database
- Ollama for local LLM inference
AI Models
- Embedding models: nomic-embed-text, mxbai-embed-large
- LLMs: llama3.2, qwen2.5, phi3, and more
- All models run via Ollama's Arm-native builds
Document Processing
- Tesseract OCR (Arm64 builds)
- Poppler PDF utilities
- Custom extractors for various formats
- Windows 11 on Arm: Snapdragon X Elite, Surface Pro X, or other Arm64 Windows devices
- Ollama: Download from ollama.com (Windows Arm64 build)
- 4GB+ RAM: Recommended for optimal performance
- 5GB+ Storage: For models and document storage
-
Download Pinguin
- Get the latest Arm64 installer from GitHub Releases
- Download:
Pinguin-Setup-1.0.0-arm64.exe
-
Install Ollama
- Visit ollama.com
- Download Ollama for Windows (ARM64)
- Run the installer
-
Run Pinguin Installer
- Double-click the installer and follow the prompts
- Pinguin will automatically detect Ollama on first launch
-
First-Run Setup
- Select your preferred LLM (e.g., llama3.2:3b for speed, llama3.2:7b for quality)
- Choose an embedding model (nomic-embed-text recommended)
- Models will download automatically via Ollama
-
Start Learning
- Upload your study materials (PDFs, documents, notes)
- Ask questions and get AI-powered answers from your content
Detailed build instructions are available in docs/ARM_BUILD_GUIDE.md.
# Clone the repository
git clone https://github.com/Kehn-Marv/Pinguin.git
cd Pinguin
# Install dependencies
npm install
cd backend
pip install -r requirements.txt
cd ..
# Build for Windows on Arm
npm run makePinguin is actively being improved. Current known issues:
- First Query Latency: Initial queries may take 1-2 minutes as models load into memory. Subsequent queries are faster (30-50 seconds depending on complexity).
- UI State Sync: Occasional UI glitches with message display. Workaround: Navigate between chats to refresh state. Fix planned for v1.1.
- Scanned Document Processing: OCR processing can take 20-30 minutes for large scanned PDFs. For best experience, use text-based PDFs when possible.
- File Format Support: Currently supports PDF, DOCX, EPUB, and TXT. Additional formats coming in future releases.
See KNOWN_ISSUES.md for details and workarounds.
- Setup Instructions - Step-by-step installation guide
- Project Overview - Complete project summary and architecture
- Technical Documentation - Detailed technical implementation
- Arm Build Guide - Building from source on Windows on Arm
- API Reference - Backend API documentation
- Known Issues - Current limitations and workarounds
- Contributing Guide - How to contribute to Pinguin
Pinguin is optimized for Arm architecture and delivers excellent performance:
- Fast Startup: < 5 seconds on Snapdragon X Elite and similar devices
- Quick Inference: 25-40 tokens/second with 3B models on Arm CPUs
- Efficient Memory: Runs comfortably in 4GB RAM
- Low Power: Extended battery life thanks to Arm efficiency
- Native Builds: All components compiled for Arm64
- Exam Preparation: Query your lecture notes and textbooks instantly
- Research: Quickly find relevant information across multiple papers
- Note Organization: Transform scattered notes into a searchable knowledge base
- Language Learning: Practice with AI using your own learning materials
- Professional Development: Build a personal knowledge base from courses and books
We welcome contributions! Pinguin is open source and built for the community.
See CONTRIBUTING.md for guidelines on:
- Reporting bugs
- Suggesting features
- Submitting pull requests
- Code style and standards
Pinguin is licensed under the MIT License. See LICENSE for details.
Built with love for the Arm AI Developer Challenge 2025.
Special thanks to:
- Ollama for making local AI accessible
- LangChain for RAG infrastructure
- ChromaDB for vector storage
- The Arm developer community
Developer: Kehn Marv
Email: kehnmarv30@gmail.com
Repository: github.com/Kehn-Marv/Pinguin
Made with passion for students who value privacy and performance.