Voice AI / Space
    Submit
    Back

    openwhispr

    View on GitHub
    Git Repo
    OpenWhispr

    Open-source cross-platform desktop application providing voice-to-text dictation, meeting transcription, and AI notes using local or cloud models.

    About openwhispr

    OpenWhispr is an open-source, cross-platform voice-to-text dictation and meeting intelligence desktop application designed as a privacy-first alternative to proprietary tools like WisprFlow and Granola.

    For the Non-Technical Reader

    Imagine having a personal assistant built directly into your keyboard shortcuts. Instead of typing out emails, meeting notes, or prompts, you hold down a hotkey, speak naturally, and clean, formatted text instantly appears at your cursor across any desktop application. It acts as both a seamless voice dictation tool and a meeting recorder that automatically recognizes who is speaking, organizes key ideas into notes, and lets you query your desktop AI assistant using real-time voice commands.

    For the Technical Reader

    OpenWhispr is designed around a local-first, privacy-focused architecture with cloud fallback capabilities (Bring Your Own Key / BYOK):

    • STT Engine Flexibility: Offers fully offline transcription leveraging local speech-to-text models like NVIDIA Parakeet, Whisper, Cohere Transcribe, and Orukeet, keeping audio data strictly on-device.

    • Speaker Diarization & Fingerprinting: Features local on-device speaker labeling and voice fingerprinting optimized via ONNX Runtime (with semantic search and live speaker ID available natively on modern architectures).

    • Context-Aware Voice Commands: Includes a dedicated Voice Assistant hotkey that can optionally append active screen captures as multimodal visual context directly to model calls (supporting models like GPT-5, Claude, Gemini, or local LLMs).

    • Cross-Platform Support: Binaries available for macOS (Apple Silicon & Intel), Windows, and Linux, with zero telemetry or data collection.

    Why It Matters

    As voice dictation and meeting summarization become central productivity workflows, concerns over continuous audio harvesting and enterprise data privacy have escalated. OpenWhispr provides an open-source framework that decouples speech recognition from invasive SaaS telemetry. By supporting both local hardware acceleration and cloud LLMs via BYOK, it lowers operating costs for teams while guaranteeing full control over sensitive voice data.

    Check out the full codebase on the OpenWhispr GitHub Repository.