LLM Remote Assistant
An ambitious full-stack project to control a PC remotely via natural language. The core vision involves an LLM that dynamically generates and executes secure Python scripts in a sandboxed environment, controlled by a cross-platform mobile client.
View on GitHub (In Progress)True Dynamic Execution
Instead of relying on a limited set of pre-defined commands, this assistant's core logic is truly dynamic. The LLM interprets the user's natural language request and generates executable Python scripts in real-time. This groundbreaking approach allows for an almost limitless combination of actions, turning complex, multi-step tasks into simple commands.
Security by Design: A Three-Layered Approach
Executing dynamically generated code presents significant security challenges. To address this, I designed a robust, three-layered security architecture. The first layer is a highly restrictive system prompt for the primary LLM. Before execution, the script is then vetted by a secondary AI safety model. Finally, the code is run within an isolated sandbox environment, preventing any access to critical system files or unauthorized network activity.
Seamless Cross-Platform Control
The system's architecture is designed for true accessibility, consisting of two core components: a cross-platform mobile client (built with React Native/Flutter) serving as the command center, and a lightweight Python agent running on the host PC. This decoupled design ensures a seamless and responsive experience, allowing you to securely control your machine from anywhere, on any modern smartphone.