AURA: Offline Autonomous Voice Agent

Most voice assistants force a compromise: either give up your private voice data and sensitive files to cloud servers, or settle for rigid, basic voice commands. AURA eliminates this trade-off by bringing autonomous, multi-step voice intelligence directly onto local hardware. Built to run locally with optional AMD ROCm hardware acceleration, AURA listens to spoken instructions, reasons through complex multi-step goals, executes local tools, and speaks back in real time. Whether you need to search private project files, draft summaries, calculate figures, or check calendar schedules, AURA plans and carries out the work while keeping your audio and documents entirely on your device. Key highlights of AURA include: • Autonomous Agent Engine: AURA breaks down broad voice requests into clear tool actions—such as searching local documents, querying file contents, and generating drafts—with explicit permission controls before executing modifications. • Hardware-Accelerated Local RAG: Indexes local PDF, Markdown, and text files using vector embeddings to answer questions with verifiable source citations, powered locally on AMD GPUs. • Real-Time Event Architecture: Uses FastAPI and WebSockets for low-latency voice streaming, instant status timelines, and dynamic feedback. • Privacy-First Design: Full offline capability with zero reliance on cloud services or external databases, persisting data strictly in local storage. AURA demonstrates how powerful, practical, and private personal AI assistance can be when built directly for local desktop hardware.