Kalamv1.0.0 / MAR 2026

Punjabi [ ਕਲਮ ] · /kə.ləm/ · A Traditional Writing Instrument

Speak messy.Type clean (privately).

A dictation app that knows ‘um’ isn’t a word and ‘scratch that’ is an instruction.

Download for MacOSAvailable for MacOSInstallation instructions
View on GitHub
Vector Quill Pen

Installation Guide

01

Drag and Drop

Open the downloaded DMG file and drag the Kalam app icon into your Applications folder.

02

First Launch (Gatekeeper)

Kalam is an independent, non-commercial release. For the first launch, Right-click (or Control-click) the app and select Open. This one-time step allows macOS to verify the software.

03

Onboarding

Follow the 4-step setup guide to configure your microphone, accessibility settings, and download your local transcription models.

The Dictation Engine.

Audio Stream

Live

Raw Text Output

The Cleanup Crew.

Messy Input

um I think we should ship on friday you know

Clean Output

I think we should ship on Friday

Works Everywhere.

Speak naturally and watch your words appear perfectly in any text field — emails, documents, code, chat.

Slack
Notion
Xcode
Terminal
Mail
VS Code
Safari

Writing at the speed of sound.

Watch your thoughts hit the page as fast as you can speak them. Speaking is ~4x faster than typing, keeping you in flow while your hands do something else.

Typing40 wpm
Speaking160 wpm

avg. words per minute

01 / Global Shortcuts

Always ready, at your cursor.

Configure a global hotkey to start dictating instantly, no matter what app you're in. Optimized for speed, designed for workflows that never stop.

⌥
Space
Start Recording

02 / Custom Dictionary

Personal Dictionary.

User-definable word replacements and AI post-processing formatting. Handles industry-specific terms, brands, and smart text expansion automatically.

SpokenOutput
"swift ui"SwiftUI
"gpt four oh"GPT-4o
"l g t m"Looks good to me!

03 / On-Device AI

On-Device AI.

Powered by latest Automatic Speech Recognition (ASR) models running on the Apple Neural Engine. Zero latency, zero cloud dependency, absolute privacy.

Apple Silicon

CPU
GPU
NEURAL ENGINE ✦

04 / Multilingual

25+ languages.

Speak in your native language or switch between them naturally. All processing happens locally with no data ever leaving your machine.

English
Español
中文
日本語
Français
Deutsch
Português
한국어
العربية
+16 more

Frequently asked.

Never. Kalam is compiled without any network entitlements. Your audio is processed entirely on-device using Apple's Neural Engine and is wiped from memory the instant transcription is complete. No data is ever transmitted, stored, or accessible to us or any third party.

Yes — completely. Kalam is free and open-source under the MIT license. There are no subscriptions, no usage limits, no premium tiers. Download it, inspect the source, and use it however you like.

If you can type into it, Kalam can fill it. It works at the OS level, injecting clean text directly into any focused text field — Slack, Notion, Xcode, Terminal, email clients, browsers, code editors, and everything in between.

Kalam currently supports 25+ languages on-device, powered by Apple's multilingual speech recognition stack. This includes English, Spanish, French, German, Japanese, Mandarin, Hindi, Arabic, Portuguese, and many more — all without a network connection.

What happens on your Mac, stays on your Mac.

Completely Isolated

Kalam runs in a strictly contained environment. It operates entirely on-device, only accessing the specific hardware resources it absolutely needs to function.

Verifiably Offline

We chose absolute privacy over auto-download convenience. Kalam is compiled entirely without network access capabilities, meaning it has zero ability to connect to the internet or phone home.

Leaves No Trace

Your voice data exists only for the millisecond it takes to transcribe. Once converted to text, the audio is instantly wiped from memory. Nothing is ever saved to disk.

Vector Quill Pen

Your pen keeps your words between you and the page.

So does Kalam.

Free, open-source, and available now.

Download for MacOSAvailable for MacOSView Source