JAMES is a fully local, browser-native AI assistant designed with privacy as the foundational principle. By leveraging WebAssembly and WebGPU, JAMES runs entirely client-side, ensuring your data never leaves your device. No cloud, no API calls, no accounts—just your machine.
- 100% Private & Local AI: All processing happens directly within your browser. There is no server communication for model inference.
- Frictionless Access: Start chatting instantly. No sign-ups, logins, or accounts are required.
- Browser-Powered Performance: Utilizes WASM and WebGPU for fast, hardware-accelerated client-side model execution.
- Integrated Python Runtime: Powered by Pyodide, allowing JAMES to execute Python code securely within the browser environment.
- Offline-Capable: Once the model is cached locally, no active internet connection is needed to chat.
The most significant feature of JAMES is the Persistent Personal Memory system. While traditional local LLMs wipe their context the moment you close the tab, JAMES is designed to remember you across sessions without compromising your privacy.
- Client-Side Encryption: All memories, notes, and session histories are encrypted and stored exclusively on your device using IndexedDB.
- Contextual Continuity: JAMES seamlessly retrieves relevant past interactions to maintain a continuous, living relationship, making conversations increasingly tailored over time.
- Zero Server Footprint: Because the memory storage is strictly local, you get all the benefits of a personalized, long-term AI thought partner without ever creating a profile on a corporate server.
- Frontend: Vanilla JavaScript, HTML5, CSS3 (No heavy frameworks)
- Compute: WebGPU for accelerated local model execution
- Runtime Environments: ONNX Runtime Web & Pyodide (WASM)
- Storage: IndexedDB for encrypted, persistent chat history and memory
Try it live at: chatbotjames.onrender.com
(Note: Because the model runs entirely locally, initial load times may vary based on your hardware and network speed as the browser caches the required WebAssembly files and model weights).
Author: Andrey Lopukhov