Local-First & Offline-Capable AI Web Apps: Service Workers, IndexedDB, SQLite WASM & ONNX

Why should a note-taking app, semantic search engine, or field-service PWA stop working the moment a user enters an airplane or subway tunnel? By combining Service Workers (Serwist), IndexedDB / OPFS, SQLite compiled to WebAssembly, and ONNX Runtime Web inside dedicated Web Workers, a Senior Freelance Full Stack & AI Programmer can build web applications that run 100% offline with zero server latency.

1. Architecture of a Local-First AI PWA (Journal: Private Markdown Notes)

1. Client-Side AES-GCM Encryption: User notes and local API keys are encrypted in the browser using Web Crypto API (PBKDF2 key derivation + AES-GCM-256).
2. In-Browser Neural Embeddings: Quantized Transformer models run inside a background Web Worker via ONNX Runtime Web (`wasm` / `webgpu`), turning Markdown paragraphs into 384-dimensional float vectors without sending private text to any third-party cloud.
3. SQLite WASM Vector Index: Vectors are persisted locally and queried in under 10 milliseconds for offline Retrieval-Augmented Generation (RAG).

2. Deep-Dive Local-First Engineering Tutorials

• Offline RAG: Building a High-Performance SQLite Vector Index Inside a Web Worker
• Client-Side ML: High-Performance Embeddings with ONNX Runtime Web & Transformers.js
• Client-Side Cryptography: Securing Local API Keys with PBKDF2 & AES-GCM