blurt

    Git Repo
    AssemblyAI

    Native macOS dictation application that converts speech to text using the AssemblyAI API and pastes it into active applications.

    About blurt

    Blurt is an open-source macOS dictation tool that converts speech directly into polished, formatted text inside whatever application currently has focus.

    1. For the Non-Technical Reader

    Imagine having a professional editor sitting next to you who listens as you speak, strips out your "ums" and "ahs," fixes your punctuation, and immediately types the cleaned-up sentence right where your cursor is positioned. Instead of typing out long emails, notes, or messages, you simply tap a single key, speak naturally, and see refined text appear in real time across any Mac application without needing a monthly subscription.

    2. For the Technical Reader

    Blurt is built natively in Swift 6 using AppKit and SwiftUI, completely bypassing Electron or web view overhead.

    • Architecture: Core logic lives in BlurtEngine, a dependency-free Swift package managing mic capture, a single synchronous HTTPS call to AssemblyAI's dictation API, and clipboard insertion via a synthesized ⌘V keypress with full prior state restoration.

    • Latency & Pre-warming: Delivers an end-to-end round trip of ~1 second by pre-warming HTTPS connections during audio capture and transitioning states instantly at key-up.

    • Processing Pipeline: Single-step audio processing leverages AssemblyAI's server-side LLM cleanup for disfluency removal and punctuation in one pass, eliminating the need for local model downloads or background daemons.

    • Accuracy & Multilingual: Outperforms Whisper on AssemblyAI benchmarks with 30% fewer hallucinations, supporting automatic language detection and code-switching across 18 languages.

    • Licensing & Requirements: Licensed under MIT; operates on a bring-your-own (BYO) AssemblyAI API key model.

    3. Why It Matters

    Blurt provides a blueprint for lightweight, privacy-conscious desktop AI utility design. By keeping the client 100% native and open-source while offloading heavy ML inference and LLM text polishing to a specialized cloud API, it eliminates heavy local compute demands and subscription fatigue. Only minimal immediate text context is shared with the API, giving users control over their data footprint.

    Explore the project on GitHub or visit AssemblyAI Blurt.