Kesha Voice Kit

Offline microphone dictation powered by the Kesha Voice Kit CLI. Records locally, transcribes locally, and copies text to your clipboard.

Anton Yakutovich

14 Installs

Install Extension

Overview

Kesha Voice Kit - Raycast extension

Offline microphone dictation for Raycast, powered by the Kesha Voice Kit CLI. Recording and transcription stay local; no API keys, no cloud roundtrips.

Commands

Dictate to Clipboard

Starts recording from the default microphone and shows a live view while you speak:

Signal meter — input level plus the microphone's name and format, so you can tell at a glance whether the right device is live.
Auto-stop on silence — after ~30 s without speech the status flips to Idle, and ~15 s later recording stops on its own; speaking again resets the timer.
Stop and Transcribe — or let the configured time limit end the session.

Transcription runs locally (NVIDIA Parakeet TDT, multilingual) and the transcript is copied to the clipboard automatically. A running transcription can be cancelled from the action panel.

Prerequisites

Install the Kesha Voice Kit CLI, then fetch the local engine + ASR models:

brew install drakulavich/tap/kesha-voice-kit
kesha install

Prefer bun? bun add -g @drakulavich/kesha-voice-kit@latest installs the same CLI.

kesha install downloads ~2.5 GB of engine and speech-to-text models on first run, so give it a few minutes. Verify the result before using the extension:

kesha record --help

macOS may ask for microphone permission the first time the command records audio. Grant access to Raycast when prompted.

Preferences

Preference	Default	When to set
`kesha` binary path	empty (auto-detect)	If `kesha` is installed somewhere Raycast cannot find.
Max recording seconds	`300`	If you want shorter or longer dictation sessions.

Source and contributions

The extension source lives alongside the main CLI at https://github.com/drakulavich/kesha-voice-kit/tree/main/raycast. Issues and feature requests go in the main repo's issue tracker.

Versions4

Silence auto-stop now works - August 8, 2026

Stop a dictation session after the idle period as documented. The check for whether anyone is speaking used a level 800x below any real room's noise floor, so every microphone counted as speech and the session always ran to the full maximum duration unless stopped by hand.
Show a signal level that matches what you hear: a silent room read about 30% before.
Measure the level from the loudest input channel rather than the average of all of them, so a microphone plugged into one input of a multi-channel audio interface is no longer read as 30% quieter than it is and cut off mid-sentence.

Sturdier setup detection - August 3, 2026

Detect whether the engine is installed from the CLI's machine-readable status output instead of matching its wording, so a reworded CLI message can no longer be mistaken for a broken install. Older CLIs keep working through the previous check.
Refuse to start a session when the engine is installed but cannot run, instead of recording audio that could never be transcribed, and say how to repair it.
Tell a CLI/extension version mismatch apart from a broken engine, so the suggested fix matches the actual problem.

Setup guidance and faster failure feedback - July 27, 2026

Rewrite the "kesha CLI not found" message as numbered setup steps, leading with Homebrew so the guidance works without knowing about bun, and include the required kesha install step.
Add actions to the error view: copy the error text, open extension preferences, or open the setup guide.
Check the CLI and engine before recording starts, so an unfinished setup shows the exact remaining command instead of a recording that cannot produce a transcript.
Warn within ~8 s when the microphone delivers no signal, pointing at the macOS Microphone permission, instead of leaving the session silent until it times out.
Stop recording after the silence auto-stop even when the level meter never starts, so a session with a dead meter no longer runs to the full duration.

Initial Version - July 24, 2026

Add Dictate to Clipboard command: records from the default microphone, transcribes locally with Kesha Voice Kit, and copies the transcript to the clipboard.
Auto-detect the kesha binary across common global install locations so Raycast's GUI environment does not require manual PATH setup.
Add a max recording duration preference to prevent runaway microphone sessions.

Compatibility

macOS

Subscribe to our newsletter.

Get product updates and news in your inbox. No spam.

Kesha Voice Kit

Offline microphone dictation powered by the Kesha Voice Kit CLI. Records locally, transcribes locally, and copies text to your clipboard.

Anton Yakutovich

14 Installs

Install Extension

Overview

Kesha Voice Kit - Raycast extension

Offline microphone dictation for Raycast, powered by the Kesha Voice Kit CLI. Recording and transcription stay local; no API keys, no cloud roundtrips.

Commands

Dictate to Clipboard

Starts recording from the default microphone and shows a live view while you speak:

Signal meter — input level plus the microphone's name and format, so you can tell at a glance whether the right device is live.
Auto-stop on silence — after ~30 s without speech the status flips to Idle, and ~15 s later recording stops on its own; speaking again resets the timer.
Stop and Transcribe — or let the configured time limit end the session.

Transcription runs locally (NVIDIA Parakeet TDT, multilingual) and the transcript is copied to the clipboard automatically. A running transcription can be cancelled from the action panel.

Prerequisites

Install the Kesha Voice Kit CLI, then fetch the local engine + ASR models:

brew install drakulavich/tap/kesha-voice-kit
kesha install

Prefer bun? bun add -g @drakulavich/kesha-voice-kit@latest installs the same CLI.

kesha install downloads ~2.5 GB of engine and speech-to-text models on first run, so give it a few minutes. Verify the result before using the extension:

kesha record --help

macOS may ask for microphone permission the first time the command records audio. Grant access to Raycast when prompted.

Preferences

Preference	Default	When to set
`kesha` binary path	empty (auto-detect)	If `kesha` is installed somewhere Raycast cannot find.
Max recording seconds	`300`	If you want shorter or longer dictation sessions.

Source and contributions

The extension source lives alongside the main CLI at https://github.com/drakulavich/kesha-voice-kit/tree/main/raycast. Issues and feature requests go in the main repo's issue tracker.

Versions4

Silence auto-stop now works - August 8, 2026

Stop a dictation session after the idle period as documented. The check for whether anyone is speaking used a level 800x below any real room's noise floor, so every microphone counted as speech and the session always ran to the full maximum duration unless stopped by hand.
Show a signal level that matches what you hear: a silent room read about 30% before.
Measure the level from the loudest input channel rather than the average of all of them, so a microphone plugged into one input of a multi-channel audio interface is no longer read as 30% quieter than it is and cut off mid-sentence.

Sturdier setup detection - August 3, 2026

Detect whether the engine is installed from the CLI's machine-readable status output instead of matching its wording, so a reworded CLI message can no longer be mistaken for a broken install. Older CLIs keep working through the previous check.
Refuse to start a session when the engine is installed but cannot run, instead of recording audio that could never be transcribed, and say how to repair it.
Tell a CLI/extension version mismatch apart from a broken engine, so the suggested fix matches the actual problem.

Setup guidance and faster failure feedback - July 27, 2026

Rewrite the "kesha CLI not found" message as numbered setup steps, leading with Homebrew so the guidance works without knowing about bun, and include the required kesha install step.
Add actions to the error view: copy the error text, open extension preferences, or open the setup guide.
Check the CLI and engine before recording starts, so an unfinished setup shows the exact remaining command instead of a recording that cannot produce a transcript.
Warn within ~8 s when the microphone delivers no signal, pointing at the macOS Microphone permission, instead of leaving the session silent until it times out.
Stop recording after the silence auto-stop even when the level meter never starts, so a session with a dead meter no longer runs to the full duration.

Initial Version - July 24, 2026

Add Dictate to Clipboard command: records from the default microphone, transcribes locally with Kesha Voice Kit, and copies the transcript to the clipboard.
Auto-detect the kesha binary across common global install locations so Raycast's GUI environment does not require manual PATH setup.
Add a max recording duration preference to prevent runaway microphone sessions.

Compatibility

macOS

Kesha Voice Kit

Kesha Voice Kit - Raycast extension

Commands

Dictate to Clipboard

Prerequisites

Preferences

Source and contributions

Commands1

Versions4

Silence auto-stop now works - August 8, 2026

Sturdier setup detection - August 3, 2026

Setup guidance and faster failure feedback - July 27, 2026

Initial Version - July 24, 2026

Subscribe to our newsletter.

Kesha Voice Kit

Kesha Voice Kit - Raycast extension

Commands

Dictate to Clipboard

Prerequisites

Preferences

Source and contributions

Commands1

Versions4

Silence auto-stop now works - August 8, 2026

Sturdier setup detection - August 3, 2026

Setup guidance and faster failure feedback - July 27, 2026

Initial Version - July 24, 2026