Kalam v2.6.0 — Hindi + English, reliability overhaul

Highlights

Hindi + English, properly. New Sarvam AI engine (Saaras v3) for the best Hindi/Hinglish accuracy — code-mix mode keeps Hinglish natural. Free API key from dashboard.sarvam.ai; automatic fallback to local Whisper. New local Large Turbo model (openai whisper-large-v3-turbo) for offline Hindi.

It just works now. Fixed the two big reliability bugs: audio was resampled from a hardcoded 48kHz (garbage output with AirPods/USB mics) and the tail of every recording could be lost to an async race. Permissions now survive app updates thanks to stable code signing. Every failure produces a clear notification instead of silence.

New:

• Hold ⌘⇧Space to talk, release to transcribe (tap-to-toggle still works)
• Personal dictionary — fix names/acronyms once, applies to every transcript
• Auto-stop after 2 min of silence or 10 min total
• Launch at login
• Live menu bar state (red = recording, purple = transcribing) and a cursor-following indicator with elapsed time
• Sarvam API key stored in the macOS Keychain
• Refreshed UI with harmonized Hindi+English typography

Install

curl -sL https://raw.githubusercontent.com/harshvardhaniimi/kalam/main/install.sh | bash

Or download Kalam.zip below, move to Applications, run xattr -cr /Applications/Kalam.app, then right-click → Open.

Requires macOS 14+ on Apple Silicon.

v2.1.1: Fix model initialization

Bug Fix

• Fix model initialization failure: WhisperKit stores downloaded models in a HuggingFace hub cache structure with nested directories. The app was looking in a flat path that didn't exist. Now correctly stores and resolves the actual download paths returned by WhisperKit.
• Fix hotkey toggle bug: The global hotkey toggle in Settings was double-flipping due to both the Toggle binding and toggleGlobalHotkey() each toggling the value.

What Changed

• ModelManager now persists actual model paths via UserDefaults
• Added fallback scanning of HuggingFace cache to find previously downloaded models
• Removed redundant .toggle() call in toggleGlobalHotkey()

Installation

curl -sL https://raw.githubusercontent.com/harshvardhaniimi/whisper-mac/main/install.sh | bash

Or download Kalam.zip below, unzip, and run:

xattr -cr ~/Downloads/Kalam.app
open ~/Downloads/Kalam.app

Kalam v2.1.0 — Real Whisper Transcription

What's New

Kalam now uses real OpenAI Whisper models for transcription, replacing the previous Apple Speech Recognition backend. This is a major upgrade to transcription quality and language support.

Changes

• Real Whisper models — Powered by WhisperKit running CoreML-optimized Whisper models on Apple Neural Engine
• Working model picker — Download and switch between Tiny, Base, Small, Medium, and Large-v3 models
• 99 language support — Full Whisper language coverage (30+ in the UI picker)
• Auto-download — Base model (~140 MB) downloads automatically on first launch
• Requires macOS 14.0+ (Sonoma) — needed for WhisperKit's CoreML optimizations

Install

curl -sL https://raw.githubusercontent.com/harshvardhaniimi/whisper-mac/main/install.sh | bash

Or download Kalam.zip, unzip, move to /Applications, then:

xattr -cr /Applications/Kalam.app

Right-click → Open (first time only).

Note

First launch will download the base Whisper model (~140 MB). This requires an internet connection. After that, everything runs locally.

Kalam v2.0.0 — Rebranded from WhisperMac

What's New

Kalam (कलम, "pen" in Hindi) — the app formerly known as WhisperMac has been rebranded.

Changes

• Renamed from WhisperMac to Kalam throughout
• New app icon with कलम Devanagari accent
• Dual-build system for future App Store support
• Centralized branding via release.config.sh
• Cleaned up repo — legacy files moved to archive/

Install

curl -sL https://raw.githubusercontent.com/harshvardhaniimi/whisper-mac/main/install.sh | bash

Or download Kalam.zip, unzip, move to /Applications, then:

xattr -cr /Applications/Kalam.app

Right-click → Open (first time only).

Existing Users

If you had WhisperMac installed, run the uninstall script first:

curl -sL https://raw.githubusercontent.com/harshvardhaniimi/whisper-mac/main/uninstall.sh | bash

Then install Kalam fresh.

v1.0.2 - New Hotkey: Cmd+Shift+Space

What's New

New Global Hotkey: Cmd+Shift+Space

• No Accessibility permission required for the hotkey itself!
• More reliable than the previous Ctrl+Ctrl approach
• Uses the standard macOS Carbon API for global hotkeys

Bug Fixes

• Fixed history view layout issue in popover
• Removed duplicate AppState instance

New Features

• Uninstall script - Completely remove WhisperMac and all data
• Improved Settings view with version and author info

Documentation

• Updated all docs to reflect new hotkey
• Added feedback section and contribution guidelines
• Acknowledged "vibe coded" development with Claude Code

Installation

Quick Install (Recommended)

curl -sL https://raw.githubusercontent.com/harshvardhaniimi/whisper-mac/main/install.sh | bash

Manual Install

1. Download WhisperMac.zip below
2. Extract and move to Applications
3. Run: xattr -cr /Applications/WhisperMac.app
4. Right-click → Open

Uninstall

curl -sL https://raw.githubusercontent.com/harshvardhaniimi/whisper-mac/main/uninstall.sh | bash

Feedback Welcome!

This app was entirely vibe coded with Claude Code. We welcome bug reports, feature requests, and pull requests!

---

Created with Claude Code by Dr. Harshvardhan

WhisperMac v1.0.1

What's New

• Beautiful new app icon with waveform design
• Purple gradient matching the app's accent color

Full Changelog

v1.0.0...v1.0.1

WhisperMac v1.0.0

Initial release of WhisperMac - local speech-to-text for macOS.

Features:

• Press Ctrl+Ctrl anywhere to record and transcribe
• Visual recording indicator follows your cursor
• Text automatically inserted at cursor position
• 100% local - no cloud, no data leaves your Mac

Installation:

1. Download and extract the zip
2. Right-click WhisperMac.app → Open (first time only)
3. Grant Accessibility permissions in System Settings