SYS://PRIVACY68.AI
[ 04 // PROJECT MISSION & CREATOR ]

About PRIVACY68 Voice Assistant

Building an uncompromised, hyper-fast, private desktop intelligence engine.

// THE PRIVACY-FIRST PHILOSOPHY

PRIVACY68 includes local speech-processing components, but the current Agent Brain uses the model identifier ollama/gemma4:31b-cloud. Model requests may be processed by a remote provider; do not assume prompts or command text remain on-device.

The command executor handles a small number of direct cases, such as greetings and pending security confirmations. Most other requests go to the Agent Brain, which asks LiteLLM for Python code and executes it in a subprocess with the application's user permissions.

P

Designed & Developed by Prinson

AI Architect & Full-Stack Systems Engineer focusing on local AI inference, high-performance desktop automation, and low-latency speech pipelines.

[ NEURAL ENGINE CORE ]

Technology Stack & Libraries

SPEECH-TO-TEXT

faster-whisper (CTranslate2)

CTranslate2 PyTorch execution engine running Whisper Medium.en on CUDA float16 GPU memory.

VOICE ACTIVITY

Silero VAD (ONNX Runtime)

Ultra-lightweight 32ms frame chunk VAD filter that rejects ambient noise and keyboard clicks.

VOICE SECURITY

CAM++ Speaker Embedder

512-dimensional neural speaker embedding model verifying voice lock cosine similarity (0.50 threshold).

AGENT BRAIN

LiteLLM · gemma4:31b-cloud

Requests generated Python through LiteLLM, then runs returned scripts locally in a subprocess. The configured model identifier is cloud-tagged.

DASHBOARD UI

WebView2 + Tailwind CSS

Modern glassmorphic Tailwind Control Center Dashboard running natively inside pywebview.

Version Release History

  • v1.4.3 · CURRENT RELEASE

    Reliable Windows Commands & Packaging

    Fixes a missing tokenizer plugin in the frozen build, routes Chrome and WhatsApp launch commands through their built-in plugins, and improves command error reporting.

  • v1.4.1 · PREVIOUS RELEASE

    Agentic Brain & Professional CLI Installation

    Uses an Agent Brain with LiteLLM model requests and generated-code execution, alongside the CLI installer.

  • v1.1.0 · LEGACY RELEASE

    Mobile Responsiveness & Neural Engine Enhancements

    Full responsive mobile navigation drawer, fluid typography, Silero VAD, and CUDA acceleration.

  • v1.0.0 · INITIAL LAUNCH

    Core Voice Assistant Engine

    Initial release with faster-whisper and CAM++ voice authentication.