Published Paper Details:

In the domain of personal computing and remote system administration, users often rely on fragmented GUI applications or complex Command Line Interfaces (CLI) to execute routine tasks, monitor system health, and manage files. Furthermore, relying on cloud-based AI assistants for local machine control introduces significant data privacy and security vulnerabilities. This project, titled V.A.L.K.Y.R.I.E. (Voice-Actuated Logic Kernel for Yielding Remote Interface Execution), addresses these challenges by developing a secure, local-first AI agent capable of executing comprehensive system operations through a conversational Telegram interface. The system combines natural language processing, a ReAct (Reasoning and Acting) agent architecture, and real-time telemetry to deliver an intuitive, hands-free control environment. The primary objective of the project is to build an end-to-end orchestration platform that translates text or voice commands into secure system actions. Utilizing LangGraph and LiteLLM, the core agent dynamically parses user intent and routes tasks to a suite of over 15 specialized local tools. These capabilities range from system health monitoring and media capture to file management, process termination, and sandboxed Python/shell execution. By processing instructions locally and supporting models via Ollama, the system ensures zero data leakage while maintaining high-fidelity command execution. The project explores a dual-layered architectural methodology: a robust Python backend for logic execution and a responsive frontend for system observability. The backend integrates aiogram for secure Telegram communication, faster-whisper for Speech-to-Text transcription, and FastAPI to serve REST endpoints and WebSockets. To provide absolute transparency into the agent's operations, a Next.js 16 dashboard was developed. This interactive web interface provides live system vitals, an event feed, and deep reasoning traces that visualize the LLM's decision-making process step-by-step. Throughout development, critical security and synchronization challenges were addressed. Strict user ID whitelisting, directory sandboxing, protected process constraints, and execution timeouts were implemented to safeguard the host machine against unauthorized access and prompt injection attacks. Modular code design ensures clear abstraction between the Telegram middleware, the LangGraph orchestrator, and the FastAPI service. In conclusion, this project demonstrates how combining local Large Language Models, robust API design, and asynchronous task execution can revolutionize remote system management. The resulting application not only exemplifies the power of AI-driven automation and secure UX design but also establishes a scalable framework for future extensions, such as dynamic Model Context Protocol (MCP) integrations and multi-device management.