Back to Parayu Blog
Product Updates

Previewing Parayu Local Engine: Privacy-First Offline Voice Dictation for macOS & Windows

A preview of the next generation of our offline voice engine running whisper.cpp locally for fast, secure, zero-internet AI dictation.

June 30, 20265 min readSebin Sabu, Founder, Parayu

We built Parayu because we believe your voice should belong to you. In a world where cloud-based AI tools record, stream, and process your personal dictation on remote servers, we wanted to build a complete offline, zero-compromise alternative.

We are preparing the Parayu Local Engine for public release. Powered by an optimized compilation of whisper.cpp and on-device CoreML/DirectML execution, Parayu lets you type with your voice anywhere on your system without an internet connection.

Why Offline Voice Recognition Matters

Most modern voice dictation apps require an active internet connection to stream your audio to cloud API endpoints. This introduces three critical problems: network latency, internet dependency, and massive privacy risks. If you are drafting a private email, a proprietary code block, or client notes, sending that audio to a third party is a security vulnerability.

Parayu solves this by running state-of-the-art quantized Whisper models directly on your CPU and GPU. Your audio is transcribed locally in memory, pasted into your active application, and immediately destroyed. Zero data is cached, zero data is uploaded, and it works perfectly on a flight or in areas with poor cellular reception.

Dual Model Architecture

The new engine supports two primary configurations optimized for different hardware profiles:

  • MAX Model (4-bit Quantized): Offers maximum accuracy with extremely low memory requirements. It uses 4-bit integer quantization to reduce the model footprint while retaining 99% of the transcription quality, running flawlessly on standard consumer laptops.
  • PRO Model (16-bit Float): Designed for high-end Apple Silicon and NVIDIA rigs. It processes unquantized weights to achieve state-of-the-art accuracy, adapting easily to thick accents and noisy background environments.

Malayalam Native Support

As part of this rollout, we are promoting native Malayalam dictation support. Parayu can listen to Malayalam speech and output native script directly, completely offline. We have optimized column sorting and custom dictionary mappings to ensure accuracy is higher than standard translation engines.

The Parayu Local Engine will be available as a free download for macOS and Windows once the release window is finalized. Join us in making dictation fast, secure, and private.

Try private, offline voice dictation

Parayu runs speech recognition locally on macOS and Windows, with Malayalam and multilingual workflows built in.